附章:Raw各指标说明(下) ⭐教你读懂 Nsight Compute 报告 系列合集⭐
附章:Raw各指标说明(下)
本附章由AI(GPT 6 Astra ultra)主写,整理reduce_ill报告的Raw指标,按硬件与用途分组供查找。本报告由ncu 2025.1.1采集;表格依据报告中的指标定义、对应版本的官方说明和section计算公式整理,先列出截图中的19项身份与规则字段,再列出原报告中可读取的1,624项指标及辅助记录。时间统一换算为微秒,字节数按byte列出,小数保留6位有效数字;具体显示精度可能与界面缩写不同。
值后面的{N}表示该项还保存了N个实例,例如不同指令位置、不同参数取值或不同采样时刻的数据。普通硬件计数按名称中的avg、max、min、sum分别取同类单元的平均、最大、最小或总和;每周期速率、峰值百分比则在此基础上使用名称指定的周期分母。其中,指标名中的active使用该硬件单元的活跃周期,elapsed使用kernel执行期间该单元流逝的全部周期;例如SM的活跃周期要求至少有一个warp驻留。throughput这类复合指标会先计算各组成计数器的峰值利用率,再取其中最大的百分比,表示最繁忙的组成部分。时间线、理论曲线和逐指令平均值则需要按实例读取,各点聚合后的默认值仅作为工具保存的汇总记录。
附表也保留工具内部用于分组、作图和地址关联的辅助字段,并在说明中注明用途。下列单位按统计含义展开:sector是32字节访问块,wavefront是数据管线一次处理的工作包,inst表示指令次数。报告中少数字段继承了不适用的单位标签,本表根据其明确公式与含义注明sector、wavefront、线程数或每周期运算量。NUMA与名称列表采用实例中的实际内容;N/A表示没有可用的汇总值,若同时列出{N},仍可读取各实例的数据。缺少公开定义的临时字段只说明可确认的用途,个人身份与关联地址已隐去。
SM执行资源
Tensor路径指标中的.peak_sustained给出每周期持续峰值,.peak_sustained_elapsed.per_second把该峰值按流逝周期频率换算成每秒上限。这里的次数指数学运算次数,单位分别写成次/cycle和次/s。例如INT1路径的整卡峰值为1,343,488次/cycle,乘本报告SM流逝周期频率约1.39019 × 109 cycle/s,得到约1.86770 × 1015次/s。该上限与本次实际完成多少Tensor运算分开读取,本kernel的实测Tensor运算数为0。
| 项 | 值 | 说明 |
|---|---|---|
sm__cycles_active.avg |
437993 cycle | SM至少有一个warp驻留的周期数;在同类硬件单元间取平均 |
sm__cycles_active.max |
441,248 cycle | SM至少有一个warp驻留的周期数;在同类硬件单元间取最大值 |
sm__cycles_active.min |
435,302 cycle | SM至少有一个warp驻留的周期数;在同类硬件单元间取最小值 |
sm__cycles_active.sum |
35,915,388 cycle | SM至少有一个warp驻留的周期数;在同类硬件单元间求和 |
sm__cycles_elapsed.avg |
442680 cycle | SM自身时钟的流逝周期数;在同类硬件单元间取平均 |
sm__cycles_elapsed.avg.per_second |
1.39019e+09 Hz | SM单元的流逝周期数按同类单元取平均值,再除以kernel执行时长,得到对应的时钟频率 |
sm__cycles_elapsed.max |
442,842 cycle | SM自身时钟的流逝周期数;在同类硬件单元间取最大值 |
sm__cycles_elapsed.max.per_second |
1.3907e+09 Hz | SM单元的流逝周期数按同类单元取最大值,再除以kernel执行时长,得到对应的时钟频率 |
sm__cycles_elapsed.min |
442,486 cycle | SM自身时钟的流逝周期数;在同类硬件单元间取最小值 |
sm__cycles_elapsed.min.per_second |
1.38958e+09 Hz | SM单元的流逝周期数按同类单元取最小值,再除以kernel执行时长,得到对应的时钟频率 |
sm__cycles_elapsed.sum |
36,299,778 cycle | SM自身时钟的流逝周期数;在同类硬件单元间求和 |
sm__cycles_elapsed.sum.per_second |
1.13995e+11 Hz | 全部SM单元的流逝周期数相加,再除以kernel执行时长,得到各单元时钟频率的合计值 |
sm__inst_executed.avg.pct_of_peak_sustained_elapsed |
14.1104 % | warp级执行指令数;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed.avg.per_cycle_active |
0.570457 inst/cycle | warp级执行指令数;在同类硬件单元间取平均并除以对应单元活跃周期数,得到每活跃周期的处理量 |
sm__inst_executed.avg.per_cycle_elapsed |
0.564416 inst/cycle | warp级执行指令数;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__inst_executed.max.pct_of_peak_sustained_elapsed |
14.6892 % | warp级执行指令数;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed.max.per_cycle_active |
0.593855 inst/cycle | warp级执行指令数;在同类硬件单元间取最大值并除以对应单元活跃周期数,得到每活跃周期的处理量 |
sm__inst_executed.max.per_cycle_elapsed |
0.587566 inst/cycle | warp级执行指令数;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__inst_executed.min.pct_of_peak_sustained_elapsed |
13.8417 % | warp级执行指令数;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed.min.per_cycle_active |
0.559594 inst/cycle | warp级执行指令数;在同类硬件单元间取最小值并除以对应单元活跃周期数,得到每活跃周期的处理量 |
sm__inst_executed.min.per_cycle_elapsed |
0.553668 inst/cycle | warp级执行指令数;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__inst_executed.sum.pct_of_peak_sustained_elapsed |
14.1104 % | warp级执行指令数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed.sum.per_cycle_active |
46.7775 inst/cycle | warp级执行指令数;在同类硬件单元间求和并除以对应单元活跃周期数,得到每活跃周期的处理量 |
sm__inst_executed.sum.per_cycle_elapsed |
46.2821 inst/cycle | warp级执行指令数;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__inst_executed_pipe_adu.avg.pct_of_peak_sustained_active |
4.37936 % | ADU管线处理的warp级执行指令数;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_adu.avg.pct_of_peak_sustained_elapsed |
4.33299 % | ADU管线处理的warp级执行指令数;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_adu.max.pct_of_peak_sustained_active |
4.55898 % | ADU管线处理的warp级执行指令数;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_adu.max.pct_of_peak_sustained_elapsed |
4.51071 % | ADU管线处理的warp级执行指令数;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_adu.min.pct_of_peak_sustained_active |
4.29596 % | ADU管线处理的warp级执行指令数;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_adu.min.pct_of_peak_sustained_elapsed |
4.25047 % | ADU管线处理的warp级执行指令数;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_adu.sum.pct_of_peak_sustained_active |
4.37936 % | ADU管线处理的warp级执行指令数;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_adu.sum.pct_of_peak_sustained_elapsed |
4.33299 % | ADU管线处理的warp级执行指令数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_alu.avg.pct_of_peak_sustained_active |
11.9634 % | ALU管线处理的warp级执行指令数;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_alu.max.pct_of_peak_sustained_active |
12.4541 % | ALU管线处理的warp级执行指令数;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_alu.min.pct_of_peak_sustained_active |
11.6158 % | ALU管线处理的warp级执行指令数;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_alu.sum.pct_of_peak_sustained_active |
11.9634 % | ALU管线处理的warp级执行指令数;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_cbu.avg.pct_of_peak_sustained_active |
0.0513206 % | CBU管线处理的warp级执行指令数;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_cbu.max.pct_of_peak_sustained_active |
0.0534256 % | CBU管线处理的warp级执行指令数;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_cbu.min.pct_of_peak_sustained_active |
0.0498296 % | CBU管线处理的warp级执行指令数;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_cbu.sum.pct_of_peak_sustained_active |
0.0513206 % | CBU管线处理的warp级执行指令数;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_cbu_pred_on_any.avg.pct_of_peak_sustained_elapsed |
0.203109 % | CBU管线处理的warp级执行指令数,至少有一个活跃线程满足执行条件;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_cbu_pred_on_any.max.pct_of_peak_sustained_elapsed |
0.211439 % | CBU管线处理的warp级执行指令数,至少有一个活跃线程满足执行条件;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_cbu_pred_on_any.min.pct_of_peak_sustained_elapsed |
0.197208 % | CBU管线处理的warp级执行指令数,至少有一个活跃线程满足执行条件;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_cbu_pred_on_any.sum.pct_of_peak_sustained_elapsed |
0.203109 % | CBU管线处理的warp级执行指令数,至少有一个活跃线程满足执行条件;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_fma.avg.pct_of_peak_sustained_active |
4.06573 % | FMA管线处理的warp级执行指令数;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_fma.max.pct_of_peak_sustained_active |
4.23249 % | FMA管线处理的warp级执行指令数;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_fma.min.pct_of_peak_sustained_active |
3.94761 % | FMA管线处理的warp级执行指令数;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_fma.sum.pct_of_peak_sustained_active |
4.06573 % | FMA管线处理的warp级执行指令数;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_fma_type_fp16.avg.pct_of_peak_sustained_active |
0 % | FMA管线处理的warp级执行指令数,只计FP16操作;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_fp64.max.pct_of_peak_sustained_active |
0 % | FP64管线处理的warp级执行指令数;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_fp64.min.pct_of_peak_sustained_active |
0 % | FP64管线处理的warp级执行指令数;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_fp64.sum.pct_of_peak_sustained_active |
0 % | FP64管线处理的warp级执行指令数;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_fp64_op_dmma.avg.pct_of_peak_sustained_active |
0 % | FP64管线处理的warp级执行指令数,只计双精度矩阵乘加;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_fp64_op_dmma.max.pct_of_peak_sustained_active |
0 % | FP64管线处理的warp级执行指令数,只计双精度矩阵乘加;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_fp64_op_dmma.min.pct_of_peak_sustained_active |
0 % | FP64管线处理的warp级执行指令数,只计双精度矩阵乘加;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_fp64_op_dmma.sum.pct_of_peak_sustained_active |
0 % | FP64管线处理的warp级执行指令数,只计双精度矩阵乘加;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_fp64_op_fp64.avg.pct_of_peak_sustained_active |
0 % | FP64管线处理的warp级执行指令数,只计普通FP64操作;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_fp64_op_fp64.max.pct_of_peak_sustained_active |
0 % | FP64管线处理的warp级执行指令数,只计普通FP64操作;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_fp64_op_fp64.min.pct_of_peak_sustained_active |
0 % | FP64管线处理的warp级执行指令数,只计普通FP64操作;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_fp64_op_fp64.sum.pct_of_peak_sustained_active |
0 % | FP64管线处理的warp级执行指令数,只计普通FP64操作;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_ipa.avg.pct_of_peak_sustained_elapsed |
0 % | IPA管线处理的warp级执行指令数;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_ipa.max.pct_of_peak_sustained_elapsed |
0 % | IPA管线处理的warp级执行指令数;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_ipa.min.pct_of_peak_sustained_elapsed |
0 % | IPA管线处理的warp级执行指令数;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_ipa.sum.pct_of_peak_sustained_elapsed |
0 % | IPA管线处理的warp级执行指令数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_lsu.avg.pct_of_peak_sustained_active |
16.1489 % | LSU管线处理的warp级执行指令数;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_lsu.avg.pct_of_peak_sustained_elapsed |
15.9779 % | LSU管线处理的warp级执行指令数;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_lsu.max.pct_of_peak_sustained_active |
16.8112 % | LSU管线处理的warp级执行指令数;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_lsu.max.pct_of_peak_sustained_elapsed |
16.6332 % | LSU管线处理的warp级执行指令数;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_lsu.min.pct_of_peak_sustained_active |
15.8414 % | LSU管线处理的warp级执行指令数;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_lsu.min.pct_of_peak_sustained_elapsed |
15.6736 % | LSU管线处理的warp级执行指令数;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_lsu.sum.pct_of_peak_sustained_active |
16.1489 % | LSU管线处理的warp级执行指令数;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_lsu.sum.pct_of_peak_sustained_elapsed |
15.9779 % | LSU管线处理的warp级执行指令数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_tensor_op_hmma.avg.pct_of_peak_sustained_active |
0 % | Tensor管线处理的warp级执行指令数,只计HMMA矩阵操作;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_tensor_op_hmma.max.pct_of_peak_sustained_active |
0 % | Tensor管线处理的warp级执行指令数,只计HMMA矩阵操作;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_tensor_op_hmma.min.pct_of_peak_sustained_active |
0 % | Tensor管线处理的warp级执行指令数,只计HMMA矩阵操作;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_tensor_op_hmma.sum.pct_of_peak_sustained_active |
0 % | Tensor管线处理的warp级执行指令数,只计HMMA矩阵操作;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_tensor_op_imma.avg.pct_of_peak_sustained_active |
0 % | Tensor管线处理的warp级执行指令数,只计IMMA或BMMA矩阵操作;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_tensor_op_imma.max.pct_of_peak_sustained_active |
0 % | Tensor管线处理的warp级执行指令数,只计IMMA或BMMA矩阵操作;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_tensor_op_imma.min.pct_of_peak_sustained_active |
0 % | Tensor管线处理的warp级执行指令数,只计IMMA或BMMA矩阵操作;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_tensor_op_imma.sum.pct_of_peak_sustained_active |
0 % | Tensor管线处理的warp级执行指令数,只计IMMA或BMMA矩阵操作;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_tex.avg.pct_of_peak_sustained_active |
0 % | TEX管线处理的warp级执行指令数;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_tex.avg.pct_of_peak_sustained_elapsed |
0 % | TEX管线处理的warp级执行指令数;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_tex.max.pct_of_peak_sustained_active |
0 % | TEX管线处理的warp级执行指令数;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_tex.max.pct_of_peak_sustained_elapsed |
0 % | TEX管线处理的warp级执行指令数;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_tex.min.pct_of_peak_sustained_active |
0 % | TEX管线处理的warp级执行指令数;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_tex.min.pct_of_peak_sustained_elapsed |
0 % | TEX管线处理的warp级执行指令数;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_tex.sum.pct_of_peak_sustained_active |
0 % | TEX管线处理的warp级执行指令数;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_tex.sum.pct_of_peak_sustained_elapsed |
0 % | TEX管线处理的warp级执行指令数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_uniform.avg.pct_of_peak_sustained_active |
0.0912367 % | Uniform管线处理的warp级执行指令数;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_uniform.avg.pct_of_peak_sustained_elapsed |
0.0902705 % | Uniform管线处理的warp级执行指令数;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_uniform.max.pct_of_peak_sustained_active |
0.0949788 % | Uniform管线处理的warp级执行指令数;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_uniform.max.pct_of_peak_sustained_elapsed |
0.093973 % | Uniform管线处理的warp级执行指令数;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_uniform.min.pct_of_peak_sustained_active |
0.088586 % | Uniform管线处理的warp级执行指令数;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_uniform.min.pct_of_peak_sustained_elapsed |
0.0876479 % | Uniform管线处理的warp级执行指令数;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_uniform.sum.pct_of_peak_sustained_active |
0.0912367 % | Uniform管线处理的warp级执行指令数;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_uniform.sum.pct_of_peak_sustained_elapsed |
0.0902705 % | Uniform管线处理的warp级执行指令数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_xu.avg.pct_of_peak_sustained_active |
8.75872 % | XU管线处理的warp级执行指令数;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_xu.avg.pct_of_peak_sustained_elapsed |
8.66597 % | XU管线处理的warp级执行指令数;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_xu.max.pct_of_peak_sustained_active |
9.11796 % | XU管线处理的warp级执行指令数;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_xu.max.pct_of_peak_sustained_elapsed |
9.02141 % | XU管线处理的warp级执行指令数;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_xu.min.pct_of_peak_sustained_active |
8.50425 % | XU管线处理的warp级执行指令数;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_xu.min.pct_of_peak_sustained_elapsed |
8.4142 % | XU管线处理的warp级执行指令数;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_xu.sum.pct_of_peak_sustained_active |
8.75872 % | XU管线处理的warp级执行指令数;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_executed_pipe_xu.sum.pct_of_peak_sustained_elapsed |
8.66597 % | XU管线处理的warp级执行指令数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__inst_issued.avg.pct_of_peak_sustained_active |
14.2785 % | warp级发射指令数;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_issued.avg.per_cycle_active |
0.571141 inst/cycle | warp级发射指令数;在同类硬件单元间取平均并除以对应单元活跃周期数,得到每活跃周期的处理量 |
sm__inst_issued.max.pct_of_peak_sustained_active |
14.8633 % | warp级发射指令数;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_issued.max.per_cycle_active |
0.594531 inst/cycle | warp级发射指令数;在同类硬件单元间取最大值并除以对应单元活跃周期数,得到每活跃周期的处理量 |
sm__inst_issued.min.pct_of_peak_sustained_active |
13.8642 % | warp级发射指令数;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_issued.min.per_cycle_active |
0.554567 inst/cycle | warp级发射指令数;在同类硬件单元间取最小值并除以对应单元活跃周期数,得到每活跃周期的处理量 |
sm__inst_issued.sum.pct_of_peak_sustained_active |
14.2785 % | warp级发射指令数;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__inst_issued.sum.per_cycle_active |
46.8335 inst/cycle | warp级发射指令数;在同类硬件单元间求和并除以对应单元活跃周期数,得到每活跃周期的处理量 |
sm__instruction_throughput.avg.pct_of_peak_sustained_active |
14.2785 % | 假定SMSP间负载均衡时的SM核心指令吞吐汇总;各组成计数器先按avg取得同类单元的平均计数,再除以相同统计范围内活跃周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__instruction_throughput.max.pct_of_peak_sustained_active |
14.8633 % | 假定SMSP间负载均衡时的SM核心指令吞吐汇总;各组成计数器先按max取得同类单元的最大计数,再除以相同统计范围内活跃周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__instruction_throughput.min.pct_of_peak_sustained_active |
13.9899 % | 假定SMSP间负载均衡时的SM核心指令吞吐汇总;各组成计数器先按min取得同类单元的最小计数,再除以相同统计范围内活跃周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__instruction_throughput.sum.pct_of_peak_sustained_active |
14.2785 % | 假定SMSP间负载均衡时的SM核心指令吞吐汇总;各组成计数器先按sum取得同类单元的合计计数,再除以相同统计范围内活跃周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__instruction_throughput_internal_activity.avg.pct_of_peak_sustained_elapsed |
0 % | 假定SMSP间负载均衡时的SM核心指令内部活动吞吐汇总;各组成计数器先按avg取得同类单元的平均计数,再除以相同统计范围内全部流逝周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__instruction_throughput_internal_activity.max.pct_of_peak_sustained_elapsed |
0 % | 假定SMSP间负载均衡时的SM核心指令内部活动吞吐汇总;各组成计数器先按max取得同类单元的最大计数,再除以相同统计范围内全部流逝周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__instruction_throughput_internal_activity.min.pct_of_peak_sustained_elapsed |
0 % | 假定SMSP间负载均衡时的SM核心指令内部活动吞吐汇总;各组成计数器先按min取得同类单元的最小计数,再除以相同统计范围内全部流逝周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__instruction_throughput_internal_activity.sum.pct_of_peak_sustained_elapsed |
0 % | 假定SMSP间负载均衡时的SM核心指令内部活动吞吐汇总;各组成计数器先按sum取得同类单元的合计计数,再除以相同统计范围内全部流逝周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__issue_active.avg.pct_of_peak_sustained_elapsed |
14.1273 % | SMSP实际发射指令的周期数;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__issue_active.max.pct_of_peak_sustained_elapsed |
14.7059 % | SMSP实际发射指令的周期数;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__issue_active.min.pct_of_peak_sustained_elapsed |
13.7174 % | SMSP实际发射指令的周期数;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__issue_active.sum.pct_of_peak_sustained_elapsed |
14.1273 % | SMSP实际发射指令的周期数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__maximum_warps_avg_per_active_cycle |
48 warp | 按本次启动资源限制计算的每SM理论最大驻留warp数 |
sm__maximum_warps_per_active_cycle_pct |
100 % | 本次配置的理论最大驻留warp数除以硬件warp上限,再乘100,得到理论occupancy |
sm__memory_throughput.avg.pct_of_peak_sustained_elapsed |
15.9779 % | 假定SMSP间负载均衡时的SM访存指令吞吐汇总;各组成计数器先按avg取得同类单元的平均计数,再除以相同统计范围内全部流逝周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__memory_throughput.max.pct_of_peak_sustained_elapsed |
16.6332 % | 假定SMSP间负载均衡时的SM访存指令吞吐汇总;各组成计数器先按max取得同类单元的最大计数,再除以相同统计范围内全部流逝周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__memory_throughput.min.pct_of_peak_sustained_elapsed |
15.6736 % | 假定SMSP间负载均衡时的SM访存指令吞吐汇总;各组成计数器先按min取得同类单元的最小计数,再除以相同统计范围内全部流逝周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__memory_throughput.sum.pct_of_peak_sustained_elapsed |
15.9779 % | 假定SMSP间负载均衡时的SM访存指令吞吐汇总;各组成计数器先按sum取得同类单元的合计计数,再除以相同统计范围内全部流逝周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__memory_throughput_internal_activity.avg.pct_of_peak_sustained_elapsed |
0 % | 假定SMSP间负载均衡时的SM访存指令内部活动吞吐汇总;各组成计数器先按avg取得同类单元的平均计数,再除以相同统计范围内全部流逝周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__memory_throughput_internal_activity.max.pct_of_peak_sustained_elapsed |
0 % | 假定SMSP间负载均衡时的SM访存指令内部活动吞吐汇总;各组成计数器先按max取得同类单元的最大计数,再除以相同统计范围内全部流逝周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__memory_throughput_internal_activity.min.pct_of_peak_sustained_elapsed |
0 % | 假定SMSP间负载均衡时的SM访存指令内部活动吞吐汇总;各组成计数器先按min取得同类单元的最小计数,再除以相同统计范围内全部流逝周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__memory_throughput_internal_activity.sum.pct_of_peak_sustained_elapsed |
0 % | 假定SMSP间负载均衡时的SM访存指令内部活动吞吐汇总;各组成计数器先按sum取得同类单元的合计计数,再除以相同统计范围内全部流逝周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__mio2rf_writeback_active.avg.pct_of_peak_sustained_elapsed |
2.69712 % | MIO结果写回寄存器接口的活跃周期数;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__mio2rf_writeback_active.max.pct_of_peak_sustained_elapsed |
2.80846 % | MIO结果写回寄存器接口的活跃周期数;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__mio2rf_writeback_active.min.pct_of_peak_sustained_elapsed |
2.61848 % | MIO结果写回寄存器接口的活跃周期数;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__mio2rf_writeback_active.sum.pct_of_peak_sustained_elapsed |
2.69712 % | MIO结果写回寄存器接口的活跃周期数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__mio_inst_issued.avg.pct_of_peak_sustained_elapsed |
6.77029 % | 从MIO控制部分发往MIO的指令数;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__mio_inst_issued.max.pct_of_peak_sustained_elapsed |
7.04798 % | 从MIO控制部分发往MIO的指令数;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__mio_inst_issued.min.pct_of_peak_sustained_elapsed |
6.64136 % | 从MIO控制部分发往MIO的指令数;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__mio_inst_issued.sum.pct_of_peak_sustained_elapsed |
6.77029 % | 从MIO控制部分发往MIO的指令数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__mio_pq_read_cycles_active.avg.pct_of_peak_sustained_elapsed |
2.70812 % | MIO操作数队列向执行管线发送寄存器操作数的周期数;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__mio_pq_read_cycles_active.max.pct_of_peak_sustained_elapsed |
2.81919 % | MIO操作数队列向执行管线发送寄存器操作数的周期数;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__mio_pq_read_cycles_active.min.pct_of_peak_sustained_elapsed |
2.62944 % | MIO操作数队列向执行管线发送寄存器操作数的周期数;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__mio_pq_read_cycles_active.sum.pct_of_peak_sustained_elapsed |
2.70812 % | MIO操作数队列向执行管线发送寄存器操作数的周期数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__mio_pq_write_cycles_active.avg.pct_of_peak_sustained_elapsed |
2.70812 % | 寄存器操作数写入MIO操作数队列的周期数;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__mio_pq_write_cycles_active.max.pct_of_peak_sustained_elapsed |
2.81919 % | 寄存器操作数写入MIO操作数队列的周期数;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__mio_pq_write_cycles_active.min.pct_of_peak_sustained_elapsed |
2.62944 % | 寄存器操作数写入MIO操作数队列的周期数;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__mio_pq_write_cycles_active.sum.pct_of_peak_sustained_elapsed |
2.70812 % | 寄存器操作数写入MIO操作数队列的周期数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.avg |
0 | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.avg.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.avg.peak_sustained |
512 次/cycle | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式的每周期持续峰值;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.avg.peak_sustained_elapsed.per_second |
7.11776e+11 次/s | Tensor路径的每秒持续峰值运算速率,输入BF16、输出FP32、关闭稀疏模式;在同类SM间取平均,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.avg.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.avg.per_second |
0 次/s | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式;在同类硬件单元间取平均并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.max |
0 | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.max.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.max.peak_sustained |
512 次/cycle | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式的每周期持续峰值;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.max.peak_sustained_elapsed.per_second |
7.11776e+11 次/s | Tensor路径的每秒持续峰值运算速率,输入BF16、输出FP32、关闭稀疏模式;在同类SM间取最大值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.max.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.max.per_second |
0 次/s | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式;在同类硬件单元间取最大值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.min |
0 | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.min.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.min.peak_sustained |
512 次/cycle | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式的每周期持续峰值;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.min.peak_sustained_elapsed.per_second |
7.11776e+11 次/s | Tensor路径的每秒持续峰值运算速率,输入BF16、输出FP32、关闭稀疏模式;在同类SM间取最小值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.min.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.min.per_second |
0 次/s | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式;在同类硬件单元间取最小值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.sum |
0 | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式;在同类硬件单元间求和 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.sum.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.sum.peak_sustained |
41,984 次/cycle | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式的每周期持续峰值;在同类硬件单元间求和 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.sum.peak_sustained_elapsed.per_second |
5.83656e+13 次/s | Tensor路径的每秒持续峰值运算速率,输入BF16、输出FP32、关闭稀疏模式;在同类SM间求和,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.sum.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_off.sum.per_second |
0 次/s | Tensor路径的数学运算数,输入BF16、输出FP32、关闭稀疏模式;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.avg |
0 | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.avg.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.avg.peak_sustained |
1,024 次/cycle | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式的每周期持续峰值;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.avg.peak_sustained_elapsed.per_second |
1.42355e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入BF16、输出FP32、启用稀疏模式;在同类SM间取平均,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.avg.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.avg.per_second |
0 次/s | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式;在同类硬件单元间取平均并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.max |
0 | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.max.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.max.peak_sustained |
1,024 次/cycle | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式的每周期持续峰值;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.max.peak_sustained_elapsed.per_second |
1.42355e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入BF16、输出FP32、启用稀疏模式;在同类SM间取最大值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.max.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.max.per_second |
0 次/s | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式;在同类硬件单元间取最大值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.min |
0 | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.min.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.min.peak_sustained |
1,024 次/cycle | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式的每周期持续峰值;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.min.peak_sustained_elapsed.per_second |
1.42355e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入BF16、输出FP32、启用稀疏模式;在同类SM间取最小值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.min.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.min.per_second |
0 次/s | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式;在同类硬件单元间取最小值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.sum |
0 | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式;在同类硬件单元间求和 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.sum.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.sum.peak_sustained |
83,968 次/cycle | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式的每周期持续峰值;在同类硬件单元间求和 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.sum.peak_sustained_elapsed.per_second |
1.16731e+14 次/s | Tensor路径的每秒持续峰值运算速率,输入BF16、输出FP32、启用稀疏模式;在同类SM间求和,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.sum.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_bf16_dst_fp32_sparsity_on.sum.per_second |
0 次/s | Tensor路径的数学运算数,输入BF16、输出FP32、启用稀疏模式;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.avg |
0 | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.avg.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.avg.peak_sustained |
1,024 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式的每周期持续峰值;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.avg.peak_sustained_elapsed.per_second |
1.42355e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入FP16、输出FP16、关闭稀疏模式;在同类SM间取平均,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.avg.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.avg.per_second |
0 次/s | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式;在同类硬件单元间取平均并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.max |
0 | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.max.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.max.peak_sustained |
1,024 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式的每周期持续峰值;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.max.peak_sustained_elapsed.per_second |
1.42355e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入FP16、输出FP16、关闭稀疏模式;在同类SM间取最大值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.max.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.max.per_second |
0 次/s | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式;在同类硬件单元间取最大值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.min |
0 | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.min.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.min.peak_sustained |
1,024 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式的每周期持续峰值;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.min.peak_sustained_elapsed.per_second |
1.42355e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入FP16、输出FP16、关闭稀疏模式;在同类SM间取最小值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.min.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.min.per_second |
0 次/s | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式;在同类硬件单元间取最小值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.sum |
0 | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式;在同类硬件单元间求和 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.sum.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.sum.peak_sustained |
83,968 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式的每周期持续峰值;在同类硬件单元间求和 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.sum.peak_sustained_elapsed.per_second |
1.16731e+14 次/s | Tensor路径的每秒持续峰值运算速率,输入FP16、输出FP16、关闭稀疏模式;在同类SM间求和,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.sum.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_off.sum.per_second |
0 次/s | Tensor路径的数学运算数,输入FP16、输出FP16、关闭稀疏模式;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.avg |
0 | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.avg.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.avg.peak_sustained |
2,048 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式的每周期持续峰值;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.avg.peak_sustained_elapsed.per_second |
2.8471e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入FP16、输出FP16、启用稀疏模式;在同类SM间取平均,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.avg.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.avg.per_second |
0 次/s | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式;在同类硬件单元间取平均并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.max |
0 | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.max.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.max.peak_sustained |
2,048 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式的每周期持续峰值;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.max.peak_sustained_elapsed.per_second |
2.8471e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入FP16、输出FP16、启用稀疏模式;在同类SM间取最大值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.max.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.max.per_second |
0 次/s | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式;在同类硬件单元间取最大值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.min |
0 | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.min.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.min.peak_sustained |
2,048 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式的每周期持续峰值;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.min.peak_sustained_elapsed.per_second |
2.8471e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入FP16、输出FP16、启用稀疏模式;在同类SM间取最小值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.min.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.min.per_second |
0 次/s | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式;在同类硬件单元间取最小值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.sum |
0 | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式;在同类硬件单元间求和 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.sum.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.sum.peak_sustained |
167,936 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式的每周期持续峰值;在同类硬件单元间求和 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.sum.peak_sustained_elapsed.per_second |
2.33463e+14 次/s | Tensor路径的每秒持续峰值运算速率,输入FP16、输出FP16、启用稀疏模式;在同类SM间求和,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.sum.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp16_dst_fp16_sparsity_on.sum.per_second |
0 次/s | Tensor路径的数学运算数,输入FP16、输出FP16、启用稀疏模式;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.avg |
0 | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.avg.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.avg.peak_sustained |
512 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式的每周期持续峰值;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.avg.peak_sustained_elapsed.per_second |
7.11776e+11 次/s | Tensor路径的每秒持续峰值运算速率,输入FP16、输出FP32、关闭稀疏模式;在同类SM间取平均,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.avg.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.avg.per_second |
0 次/s | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式;在同类硬件单元间取平均并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.max |
0 | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.max.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.max.peak_sustained |
512 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式的每周期持续峰值;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.max.peak_sustained_elapsed.per_second |
7.11776e+11 次/s | Tensor路径的每秒持续峰值运算速率,输入FP16、输出FP32、关闭稀疏模式;在同类SM间取最大值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.max.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.max.per_second |
0 次/s | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式;在同类硬件单元间取最大值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.min |
0 | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.min.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.min.peak_sustained |
512 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式的每周期持续峰值;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.min.peak_sustained_elapsed.per_second |
7.11776e+11 次/s | Tensor路径的每秒持续峰值运算速率,输入FP16、输出FP32、关闭稀疏模式;在同类SM间取最小值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.min.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.min.per_second |
0 次/s | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式;在同类硬件单元间取最小值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.sum |
0 | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式;在同类硬件单元间求和 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.sum.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.sum.peak_sustained |
41,984 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式的每周期持续峰值;在同类硬件单元间求和 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.sum.peak_sustained_elapsed.per_second |
5.83656e+13 次/s | Tensor路径的每秒持续峰值运算速率,输入FP16、输出FP32、关闭稀疏模式;在同类SM间求和,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.sum.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_off.sum.per_second |
0 次/s | Tensor路径的数学运算数,输入FP16、输出FP32、关闭稀疏模式;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.avg |
0 | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.avg.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.avg.peak_sustained |
1,024 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式的每周期持续峰值;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.avg.peak_sustained_elapsed.per_second |
1.42355e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入FP16、输出FP32、启用稀疏模式;在同类SM间取平均,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.avg.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.avg.per_second |
0 次/s | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式;在同类硬件单元间取平均并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.max |
0 | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.max.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.max.peak_sustained |
1,024 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式的每周期持续峰值;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.max.peak_sustained_elapsed.per_second |
1.42355e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入FP16、输出FP32、启用稀疏模式;在同类SM间取最大值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.max.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.max.per_second |
0 次/s | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式;在同类硬件单元间取最大值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.min |
0 | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.min.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.min.peak_sustained |
1,024 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式的每周期持续峰值;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.min.peak_sustained_elapsed.per_second |
1.42355e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入FP16、输出FP32、启用稀疏模式;在同类SM间取最小值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.min.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.min.per_second |
0 次/s | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式;在同类硬件单元间取最小值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.sum |
0 | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式;在同类硬件单元间求和 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.sum.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.sum.peak_sustained |
83,968 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式的每周期持续峰值;在同类硬件单元间求和 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.sum.peak_sustained_elapsed.per_second |
1.16731e+14 次/s | Tensor路径的每秒持续峰值运算速率,输入FP16、输出FP32、启用稀疏模式;在同类SM间求和,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.sum.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp16_dst_fp32_sparsity_on.sum.per_second |
0 次/s | Tensor路径的数学运算数,输入FP16、输出FP32、启用稀疏模式;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp64.avg |
0 | Tensor路径的数学运算数,输入FP64;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_fp64.avg.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP64;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp64.avg.peak_sustained |
3.53103 次/cycle | Tensor路径的数学运算数,输入FP64的每周期持续峰值;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_fp64.avg.peak_sustained_elapsed.per_second |
4.9088e+09 次/s | Tensor路径的每秒持续峰值运算速率,输入FP64;在同类SM间取平均,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp64.avg.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP64;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp64.avg.per_second |
0 次/s | Tensor路径的数学运算数,输入FP64;在同类硬件单元间取平均并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp64.max |
0 | Tensor路径的数学运算数,输入FP64;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_fp64.max.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP64;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp64.max.peak_sustained |
3.53103 次/cycle | Tensor路径的数学运算数,输入FP64的每周期持续峰值;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_fp64.max.peak_sustained_elapsed.per_second |
4.9088e+09 次/s | Tensor路径的每秒持续峰值运算速率,输入FP64;在同类SM间取最大值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp64.max.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP64;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp64.max.per_second |
0 次/s | Tensor路径的数学运算数,输入FP64;在同类硬件单元间取最大值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp64.min |
0 | Tensor路径的数学运算数,输入FP64;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_fp64.min.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP64;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp64.min.peak_sustained |
3.53103 次/cycle | Tensor路径的数学运算数,输入FP64的每周期持续峰值;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_fp64.min.peak_sustained_elapsed.per_second |
4.9088e+09 次/s | Tensor路径的每秒持续峰值运算速率,输入FP64;在同类SM间取最小值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp64.min.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP64;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp64.min.per_second |
0 次/s | Tensor路径的数学运算数,输入FP64;在同类硬件单元间取最小值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_fp64.sum |
0 | Tensor路径的数学运算数,输入FP64;在同类硬件单元间求和 |
sm__ops_path_tensor_src_fp64.sum.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入FP64;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_fp64.sum.peak_sustained |
289.545 次/cycle | Tensor路径的数学运算数,输入FP64的每周期持续峰值;在同类硬件单元间求和 |
sm__ops_path_tensor_src_fp64.sum.peak_sustained_elapsed.per_second |
4.02522e+11 次/s | Tensor路径的每秒持续峰值运算速率,输入FP64;在同类SM间求和,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_fp64.sum.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入FP64;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_fp64.sum.per_second |
0 次/s | Tensor路径的数学运算数,输入FP64;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int1.avg |
0 | Tensor路径的数学运算数,输入INT1;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_int1.avg.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT1;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int1.avg.peak_sustained |
16,384 次/cycle | Tensor路径的数学运算数,输入INT1的每周期持续峰值;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_int1.avg.peak_sustained_elapsed.per_second |
2.27768e+13 次/s | Tensor路径的每秒持续峰值运算速率,输入INT1;在同类SM间取平均,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int1.avg.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT1;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int1.avg.per_second |
0 次/s | Tensor路径的数学运算数,输入INT1;在同类硬件单元间取平均并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int1.max |
0 | Tensor路径的数学运算数,输入INT1;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_int1.max.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT1;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int1.max.peak_sustained |
16,384 次/cycle | Tensor路径的数学运算数,输入INT1的每周期持续峰值;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_int1.max.peak_sustained_elapsed.per_second |
2.27768e+13 次/s | Tensor路径的每秒持续峰值运算速率,输入INT1;在同类SM间取最大值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int1.max.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT1;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int1.max.per_second |
0 次/s | Tensor路径的数学运算数,输入INT1;在同类硬件单元间取最大值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int1.min |
0 | Tensor路径的数学运算数,输入INT1;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_int1.min.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT1;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int1.min.peak_sustained |
16,384 次/cycle | Tensor路径的数学运算数,输入INT1的每周期持续峰值;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_int1.min.peak_sustained_elapsed.per_second |
2.27768e+13 次/s | Tensor路径的每秒持续峰值运算速率,输入INT1;在同类SM间取最小值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int1.min.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT1;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int1.min.per_second |
0 次/s | Tensor路径的数学运算数,输入INT1;在同类硬件单元间取最小值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int1.sum |
0 | Tensor路径的数学运算数,输入INT1;在同类硬件单元间求和 |
sm__ops_path_tensor_src_int1.sum.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT1;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int1.sum.peak_sustained |
1,343,488 次/cycle | Tensor路径的数学运算数,输入INT1的每周期持续峰值;在同类硬件单元间求和 |
sm__ops_path_tensor_src_int1.sum.peak_sustained_elapsed.per_second |
1,867,700,365,390,413 次/s | Tensor路径的每秒持续峰值运算速率,输入INT1;在同类SM间求和,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int1.sum.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT1;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int1.sum.per_second |
0 次/s | Tensor路径的数学运算数,输入INT1;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int4_sparsity_off.avg |
0 | Tensor路径的数学运算数,输入INT4、关闭稀疏模式;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_int4_sparsity_off.avg.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT4、关闭稀疏模式;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int4_sparsity_off.avg.peak_sustained |
4,096 次/cycle | Tensor路径的数学运算数,输入INT4、关闭稀疏模式的每周期持续峰值;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_int4_sparsity_off.avg.peak_sustained_elapsed.per_second |
5.69421e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入INT4、关闭稀疏模式;在同类SM间取平均,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int4_sparsity_off.avg.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT4、关闭稀疏模式;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int4_sparsity_off.avg.per_second |
0 次/s | Tensor路径的数学运算数,输入INT4、关闭稀疏模式;在同类硬件单元间取平均并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int4_sparsity_off.max |
0 | Tensor路径的数学运算数,输入INT4、关闭稀疏模式;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_int4_sparsity_off.max.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT4、关闭稀疏模式;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int4_sparsity_off.max.peak_sustained |
4,096 次/cycle | Tensor路径的数学运算数,输入INT4、关闭稀疏模式的每周期持续峰值;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_int4_sparsity_off.max.peak_sustained_elapsed.per_second |
5.69421e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入INT4、关闭稀疏模式;在同类SM间取最大值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int4_sparsity_off.max.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT4、关闭稀疏模式;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int4_sparsity_off.max.per_second |
0 次/s | Tensor路径的数学运算数,输入INT4、关闭稀疏模式;在同类硬件单元间取最大值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int4_sparsity_off.min |
0 | Tensor路径的数学运算数,输入INT4、关闭稀疏模式;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_int4_sparsity_off.min.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT4、关闭稀疏模式;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int4_sparsity_off.min.peak_sustained |
4,096 次/cycle | Tensor路径的数学运算数,输入INT4、关闭稀疏模式的每周期持续峰值;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_int4_sparsity_off.min.peak_sustained_elapsed.per_second |
5.69421e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入INT4、关闭稀疏模式;在同类SM间取最小值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int4_sparsity_off.min.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT4、关闭稀疏模式;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int4_sparsity_off.min.per_second |
0 次/s | Tensor路径的数学运算数,输入INT4、关闭稀疏模式;在同类硬件单元间取最小值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int4_sparsity_off.sum |
0 | Tensor路径的数学运算数,输入INT4、关闭稀疏模式;在同类硬件单元间求和 |
sm__ops_path_tensor_src_int4_sparsity_off.sum.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT4、关闭稀疏模式;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int4_sparsity_off.sum.peak_sustained |
335,872 次/cycle | Tensor路径的数学运算数,输入INT4、关闭稀疏模式的每周期持续峰值;在同类硬件单元间求和 |
sm__ops_path_tensor_src_int4_sparsity_off.sum.peak_sustained_elapsed.per_second |
4.66925e+14 次/s | Tensor路径的每秒持续峰值运算速率,输入INT4、关闭稀疏模式;在同类SM间求和,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int4_sparsity_off.sum.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT4、关闭稀疏模式;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int4_sparsity_off.sum.per_second |
0 次/s | Tensor路径的数学运算数,输入INT4、关闭稀疏模式;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int4_sparsity_on.avg |
0 | Tensor路径的数学运算数,输入INT4、启用稀疏模式;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_int4_sparsity_on.avg.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT4、启用稀疏模式;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int4_sparsity_on.avg.peak_sustained |
8,192 次/cycle | Tensor路径的数学运算数,输入INT4、启用稀疏模式的每周期持续峰值;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_int4_sparsity_on.avg.peak_sustained_elapsed.per_second |
1.13884e+13 次/s | Tensor路径的每秒持续峰值运算速率,输入INT4、启用稀疏模式;在同类SM间取平均,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int4_sparsity_on.avg.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT4、启用稀疏模式;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int4_sparsity_on.avg.per_second |
0 次/s | Tensor路径的数学运算数,输入INT4、启用稀疏模式;在同类硬件单元间取平均并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int4_sparsity_on.max |
0 | Tensor路径的数学运算数,输入INT4、启用稀疏模式;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_int4_sparsity_on.max.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT4、启用稀疏模式;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int4_sparsity_on.max.peak_sustained |
8,192 次/cycle | Tensor路径的数学运算数,输入INT4、启用稀疏模式的每周期持续峰值;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_int4_sparsity_on.max.peak_sustained_elapsed.per_second |
1.13884e+13 次/s | Tensor路径的每秒持续峰值运算速率,输入INT4、启用稀疏模式;在同类SM间取最大值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int4_sparsity_on.max.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT4、启用稀疏模式;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int4_sparsity_on.max.per_second |
0 次/s | Tensor路径的数学运算数,输入INT4、启用稀疏模式;在同类硬件单元间取最大值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int4_sparsity_on.min |
0 | Tensor路径的数学运算数,输入INT4、启用稀疏模式;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_int4_sparsity_on.min.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT4、启用稀疏模式;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int4_sparsity_on.min.peak_sustained |
8,192 次/cycle | Tensor路径的数学运算数,输入INT4、启用稀疏模式的每周期持续峰值;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_int4_sparsity_on.min.peak_sustained_elapsed.per_second |
1.13884e+13 次/s | Tensor路径的每秒持续峰值运算速率,输入INT4、启用稀疏模式;在同类SM间取最小值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int4_sparsity_on.min.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT4、启用稀疏模式;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int4_sparsity_on.min.per_second |
0 次/s | Tensor路径的数学运算数,输入INT4、启用稀疏模式;在同类硬件单元间取最小值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int4_sparsity_on.sum |
0 | Tensor路径的数学运算数,输入INT4、启用稀疏模式;在同类硬件单元间求和 |
sm__ops_path_tensor_src_int4_sparsity_on.sum.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT4、启用稀疏模式;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int4_sparsity_on.sum.peak_sustained |
671,744 次/cycle | Tensor路径的数学运算数,输入INT4、启用稀疏模式的每周期持续峰值;在同类硬件单元间求和 |
sm__ops_path_tensor_src_int4_sparsity_on.sum.peak_sustained_elapsed.per_second |
9.3385e+14 次/s | Tensor路径的每秒持续峰值运算速率,输入INT4、启用稀疏模式;在同类SM间求和,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int4_sparsity_on.sum.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT4、启用稀疏模式;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int4_sparsity_on.sum.per_second |
0 次/s | Tensor路径的数学运算数,输入INT4、启用稀疏模式;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int8_sparsity_off.avg |
0 | Tensor路径的数学运算数,输入INT8、关闭稀疏模式;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_int8_sparsity_off.avg.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT8、关闭稀疏模式;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int8_sparsity_off.avg.peak_sustained |
2,048 次/cycle | Tensor路径的数学运算数,输入INT8、关闭稀疏模式的每周期持续峰值;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_int8_sparsity_off.avg.peak_sustained_elapsed.per_second |
2.8471e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入INT8、关闭稀疏模式;在同类SM间取平均,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int8_sparsity_off.avg.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT8、关闭稀疏模式;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int8_sparsity_off.avg.per_second |
0 次/s | Tensor路径的数学运算数,输入INT8、关闭稀疏模式;在同类硬件单元间取平均并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int8_sparsity_off.max |
0 | Tensor路径的数学运算数,输入INT8、关闭稀疏模式;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_int8_sparsity_off.max.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT8、关闭稀疏模式;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int8_sparsity_off.max.peak_sustained |
2,048 次/cycle | Tensor路径的数学运算数,输入INT8、关闭稀疏模式的每周期持续峰值;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_int8_sparsity_off.max.peak_sustained_elapsed.per_second |
2.8471e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入INT8、关闭稀疏模式;在同类SM间取最大值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int8_sparsity_off.max.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT8、关闭稀疏模式;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int8_sparsity_off.max.per_second |
0 次/s | Tensor路径的数学运算数,输入INT8、关闭稀疏模式;在同类硬件单元间取最大值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int8_sparsity_off.min |
0 | Tensor路径的数学运算数,输入INT8、关闭稀疏模式;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_int8_sparsity_off.min.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT8、关闭稀疏模式;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int8_sparsity_off.min.peak_sustained |
2,048 次/cycle | Tensor路径的数学运算数,输入INT8、关闭稀疏模式的每周期持续峰值;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_int8_sparsity_off.min.peak_sustained_elapsed.per_second |
2.8471e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入INT8、关闭稀疏模式;在同类SM间取最小值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int8_sparsity_off.min.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT8、关闭稀疏模式;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int8_sparsity_off.min.per_second |
0 次/s | Tensor路径的数学运算数,输入INT8、关闭稀疏模式;在同类硬件单元间取最小值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int8_sparsity_off.sum |
0 | Tensor路径的数学运算数,输入INT8、关闭稀疏模式;在同类硬件单元间求和 |
sm__ops_path_tensor_src_int8_sparsity_off.sum.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT8、关闭稀疏模式;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int8_sparsity_off.sum.peak_sustained |
167,936 次/cycle | Tensor路径的数学运算数,输入INT8、关闭稀疏模式的每周期持续峰值;在同类硬件单元间求和 |
sm__ops_path_tensor_src_int8_sparsity_off.sum.peak_sustained_elapsed.per_second |
2.33463e+14 次/s | Tensor路径的每秒持续峰值运算速率,输入INT8、关闭稀疏模式;在同类SM间求和,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int8_sparsity_off.sum.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT8、关闭稀疏模式;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int8_sparsity_off.sum.per_second |
0 次/s | Tensor路径的数学运算数,输入INT8、关闭稀疏模式;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int8_sparsity_on.avg |
0 | Tensor路径的数学运算数,输入INT8、启用稀疏模式;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_int8_sparsity_on.avg.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT8、启用稀疏模式;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int8_sparsity_on.avg.peak_sustained |
4,096 次/cycle | Tensor路径的数学运算数,输入INT8、启用稀疏模式的每周期持续峰值;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_int8_sparsity_on.avg.peak_sustained_elapsed.per_second |
5.69421e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入INT8、启用稀疏模式;在同类SM间取平均,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int8_sparsity_on.avg.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT8、启用稀疏模式;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int8_sparsity_on.avg.per_second |
0 次/s | Tensor路径的数学运算数,输入INT8、启用稀疏模式;在同类硬件单元间取平均并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int8_sparsity_on.max |
0 | Tensor路径的数学运算数,输入INT8、启用稀疏模式;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_int8_sparsity_on.max.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT8、启用稀疏模式;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int8_sparsity_on.max.peak_sustained |
4,096 次/cycle | Tensor路径的数学运算数,输入INT8、启用稀疏模式的每周期持续峰值;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_int8_sparsity_on.max.peak_sustained_elapsed.per_second |
5.69421e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入INT8、启用稀疏模式;在同类SM间取最大值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int8_sparsity_on.max.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT8、启用稀疏模式;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int8_sparsity_on.max.per_second |
0 次/s | Tensor路径的数学运算数,输入INT8、启用稀疏模式;在同类硬件单元间取最大值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int8_sparsity_on.min |
0 | Tensor路径的数学运算数,输入INT8、启用稀疏模式;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_int8_sparsity_on.min.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT8、启用稀疏模式;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int8_sparsity_on.min.peak_sustained |
4,096 次/cycle | Tensor路径的数学运算数,输入INT8、启用稀疏模式的每周期持续峰值;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_int8_sparsity_on.min.peak_sustained_elapsed.per_second |
5.69421e+12 次/s | Tensor路径的每秒持续峰值运算速率,输入INT8、启用稀疏模式;在同类SM间取最小值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int8_sparsity_on.min.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT8、启用稀疏模式;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int8_sparsity_on.min.per_second |
0 次/s | Tensor路径的数学运算数,输入INT8、启用稀疏模式;在同类硬件单元间取最小值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_int8_sparsity_on.sum |
0 | Tensor路径的数学运算数,输入INT8、启用稀疏模式;在同类硬件单元间求和 |
sm__ops_path_tensor_src_int8_sparsity_on.sum.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入INT8、启用稀疏模式;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_int8_sparsity_on.sum.peak_sustained |
335,872 次/cycle | Tensor路径的数学运算数,输入INT8、启用稀疏模式的每周期持续峰值;在同类硬件单元间求和 |
sm__ops_path_tensor_src_int8_sparsity_on.sum.peak_sustained_elapsed.per_second |
4.66925e+14 次/s | Tensor路径的每秒持续峰值运算速率,输入INT8、启用稀疏模式;在同类SM间求和,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_int8_sparsity_on.sum.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入INT8、启用稀疏模式;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_int8_sparsity_on.sum.per_second |
0 次/s | Tensor路径的数学运算数,输入INT8、启用稀疏模式;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.avg |
0 | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.avg.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.avg.peak_sustained |
256 次/cycle | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式的每周期持续峰值;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.avg.peak_sustained_elapsed.per_second |
3.55888e+11 次/s | Tensor路径的每秒持续峰值运算速率,输入TF32、输出FP32、关闭稀疏模式;在同类SM间取平均,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.avg.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.avg.per_second |
0 次/s | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式;在同类硬件单元间取平均并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.max |
0 | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.max.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.max.peak_sustained |
256 次/cycle | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式的每周期持续峰值;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.max.peak_sustained_elapsed.per_second |
3.55888e+11 次/s | Tensor路径的每秒持续峰值运算速率,输入TF32、输出FP32、关闭稀疏模式;在同类SM间取最大值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.max.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.max.per_second |
0 次/s | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式;在同类硬件单元间取最大值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.min |
0 | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.min.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.min.peak_sustained |
256 次/cycle | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式的每周期持续峰值;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.min.peak_sustained_elapsed.per_second |
3.55888e+11 次/s | Tensor路径的每秒持续峰值运算速率,输入TF32、输出FP32、关闭稀疏模式;在同类SM间取最小值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.min.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.min.per_second |
0 次/s | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式;在同类硬件单元间取最小值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.sum |
0 | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式;在同类硬件单元间求和 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.sum.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.sum.peak_sustained |
20,992 次/cycle | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式的每周期持续峰值;在同类硬件单元间求和 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.sum.peak_sustained_elapsed.per_second |
2.91828e+13 次/s | Tensor路径的每秒持续峰值运算速率,输入TF32、输出FP32、关闭稀疏模式;在同类SM间求和,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.sum.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_off.sum.per_second |
0 次/s | Tensor路径的数学运算数,输入TF32、输出FP32、关闭稀疏模式;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.avg |
0 | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.avg.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.avg.peak_sustained |
512 次/cycle | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式的每周期持续峰值;在同类硬件单元间取平均 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.avg.peak_sustained_elapsed.per_second |
7.11776e+11 次/s | Tensor路径的每秒持续峰值运算速率,输入TF32、输出FP32、启用稀疏模式;在同类SM间取平均,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.avg.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.avg.per_second |
0 次/s | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式;在同类硬件单元间取平均并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.max |
0 | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.max.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.max.peak_sustained |
512 次/cycle | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式的每周期持续峰值;在同类硬件单元间取最大值 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.max.peak_sustained_elapsed.per_second |
7.11776e+11 次/s | Tensor路径的每秒持续峰值运算速率,输入TF32、输出FP32、启用稀疏模式;在同类SM间取最大值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.max.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.max.per_second |
0 次/s | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式;在同类硬件单元间取最大值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.min |
0 | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.min.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.min.peak_sustained |
512 次/cycle | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式的每周期持续峰值;在同类硬件单元间取最小值 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.min.peak_sustained_elapsed.per_second |
7.11776e+11 次/s | Tensor路径的每秒持续峰值运算速率,输入TF32、输出FP32、启用稀疏模式;在同类SM间取最小值,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.min.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.min.per_second |
0 次/s | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式;在同类硬件单元间取最小值并除以运行时长,得到每秒处理量 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.sum |
0 | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式;在同类硬件单元间求和 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.sum.pct_of_peak_sustained_elapsed |
0 % | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.sum.peak_sustained |
41,984 次/cycle | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式的每周期持续峰值;在同类硬件单元间求和 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.sum.peak_sustained_elapsed.per_second |
5.83656e+13 次/s | Tensor路径的每秒持续峰值运算速率,输入TF32、输出FP32、启用稀疏模式;在同类SM间求和,由对应每周期持续峰值乘流逝周期频率换算,表示每秒数学运算次数的上限 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.sum.per_cycle_elapsed |
0 次/cycle | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
sm__ops_path_tensor_src_tf32_dst_fp32_sparsity_on.sum.per_second |
0 次/s | Tensor路径的数学运算数,输入TF32、输出FP32、启用稀疏模式;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
sm__pipe_alu_cycles_active.avg.pct_of_peak_sustained_active |
11.9634 % | ALU管线有活动的周期数;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_alu_cycles_active.avg.pct_of_peak_sustained_elapsed |
11.8367 % | ALU管线有活动的周期数;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_alu_cycles_active.max.pct_of_peak_sustained_active |
12.4541 % | ALU管线有活动的周期数;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_alu_cycles_active.max.pct_of_peak_sustained_elapsed |
12.3222 % | ALU管线有活动的周期数;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_alu_cycles_active.min.pct_of_peak_sustained_active |
11.6158 % | ALU管线有活动的周期数;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_alu_cycles_active.min.pct_of_peak_sustained_elapsed |
11.4928 % | ALU管线有活动的周期数;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_alu_cycles_active.sum.pct_of_peak_sustained_active |
11.9634 % | ALU管线有活动的周期数;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_alu_cycles_active.sum.pct_of_peak_sustained_elapsed |
11.8367 % | ALU管线有活动的周期数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_fma_cycles_active.avg.pct_of_peak_sustained_active |
5.16628 % | FMA管线有活动的周期数;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_fma_cycles_active.avg.pct_of_peak_sustained_elapsed |
5.11157 % | FMA管线有活动的周期数;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_fma_cycles_active.max.pct_of_peak_sustained_active |
5.38919 % | FMA管线有活动的周期数;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_fma_cycles_active.max.pct_of_peak_sustained_elapsed |
5.33212 % | FMA管线有活动的周期数;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_fma_cycles_active.min.pct_of_peak_sustained_active |
5.04923 % | FMA管线有活动的周期数;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_fma_cycles_active.min.pct_of_peak_sustained_elapsed |
4.99576 % | FMA管线有活动的周期数;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_fma_cycles_active.sum.pct_of_peak_sustained_active |
5.16628 % | FMA管线有活动的周期数;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_fma_cycles_active.sum.pct_of_peak_sustained_elapsed |
5.11157 % | FMA管线有活动的周期数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_fmaheavy_cycles_active.avg.pct_of_peak_sustained_elapsed |
8.81983 % | FMA Heavy管线有活动的周期数;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_fmaheavy_cycles_active.max.pct_of_peak_sustained_elapsed |
9.20156 % | FMA Heavy管线有活动的周期数;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_fmaheavy_cycles_active.min.pct_of_peak_sustained_elapsed |
8.62745 % | FMA Heavy管线有活动的周期数;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_fmaheavy_cycles_active.sum.pct_of_peak_sustained_elapsed |
8.81983 % | FMA Heavy管线有活动的周期数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_fp64_cycles_active.avg.pct_of_peak_sustained_active |
0 % | FP64管线有活动的周期数;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_fp64_cycles_active.avg.pct_of_peak_sustained_elapsed |
0 % | FP64管线有活动的周期数;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_fp64_cycles_active.max.pct_of_peak_sustained_active |
0 % | FP64管线有活动的周期数;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_fp64_cycles_active.max.pct_of_peak_sustained_elapsed |
0 % | FP64管线有活动的周期数;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_fp64_cycles_active.min.pct_of_peak_sustained_active |
0 % | FP64管线有活动的周期数;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_fp64_cycles_active.min.pct_of_peak_sustained_elapsed |
0 % | FP64管线有活动的周期数;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_fp64_cycles_active.sum.pct_of_peak_sustained_active |
0 % | FP64管线有活动的周期数;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_fp64_cycles_active.sum.pct_of_peak_sustained_elapsed |
0 % | FP64管线有活动的周期数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_tensor_cycles_active.avg.pct_of_peak_sustained_active |
0 % | Tensor管线有活动的周期数;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_tensor_cycles_active.avg.pct_of_peak_sustained_elapsed |
0 % | Tensor管线有活动的周期数;在同类硬件单元间取平均,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_tensor_cycles_active.max.pct_of_peak_sustained_active |
0 % | Tensor管线有活动的周期数;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_tensor_cycles_active.max.pct_of_peak_sustained_elapsed |
0 % | Tensor管线有活动的周期数;在同类硬件单元间取最大值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_tensor_cycles_active.min.pct_of_peak_sustained_active |
0 % | Tensor管线有活动的周期数;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_tensor_cycles_active.min.pct_of_peak_sustained_elapsed |
0 % | Tensor管线有活动的周期数;在同类硬件单元间取最小值,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_tensor_cycles_active.sum.pct_of_peak_sustained_active |
0 % | Tensor管线有活动的周期数;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_tensor_cycles_active.sum.pct_of_peak_sustained_elapsed |
0 % | Tensor管线有活动的周期数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__pipe_tensor_op_hmma_cycles_active.avg.pct_of_peak_sustained_active |
0 % | Tensor管线中HMMA操作有活动的周期数;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_tensor_op_hmma_cycles_active.max.pct_of_peak_sustained_active |
0 % | Tensor管线中HMMA操作有活动的周期数;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_tensor_op_hmma_cycles_active.min.pct_of_peak_sustained_active |
0 % | Tensor管线中HMMA操作有活动的周期数;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_tensor_op_hmma_cycles_active.sum.pct_of_peak_sustained_active |
0 % | Tensor管线中HMMA操作有活动的周期数;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_tensor_op_imma_cycles_active.avg.pct_of_peak_sustained_active |
0 % | Tensor管线中IMMA/BMMA操作有活动的周期数;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_tensor_op_imma_cycles_active.max.pct_of_peak_sustained_active |
0 % | Tensor管线中IMMA/BMMA操作有活动的周期数;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_tensor_op_imma_cycles_active.min.pct_of_peak_sustained_active |
0 % | Tensor管线中IMMA/BMMA操作有活动的周期数;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__pipe_tensor_op_imma_cycles_active.sum.pct_of_peak_sustained_active |
0 % | Tensor管线中IMMA/BMMA操作有活动的周期数;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__sass_inst_executed_op_ldgsts_cache_access.sum |
0 inst | 经过L1缓存的全局到共享内存复制指令的warp级执行次数;在同类硬件单元间求和 |
sm__sass_inst_executed_op_ldgsts_cache_bypass.sum |
0 inst | 绕过L1缓存的全局到共享内存复制指令的warp级执行次数;在同类硬件单元间求和 |
sm__sass_l1tex_m_xbar2l1tex_read_bytes_mem_global_op_ldgsts_cache_bypass.sum |
0 byte | 从L2读入L1TEX访存阶段的字节数,用于绕过L1缓存的全局到共享内存复制;在同类硬件单元间求和 |
sm__sass_l1tex_m_xbar2l1tex_read_bytes_mem_global_op_ldgsts_cache_bypass.sum.pct_of_peak_sustained_elapsed |
0 % | 从L2读入L1TEX访存阶段的字节数,用于绕过L1缓存的全局到共享内存复制;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__sass_l1tex_m_xbar2l1tex_read_bytes_mem_global_op_ldgsts_cache_bypass.sum.per_second |
0 byte/s | 从L2读入L1TEX访存阶段的字节数,用于绕过L1缓存的全局到共享内存复制;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
sm__sass_l1tex_t_requests_pipe_lsu_mem_global_op_ldgsts.sum |
0 | L1TEX缓存查询阶段收到的全局到共享内存复制请求数;在同类硬件单元间求和 |
sm__sass_l1tex_t_requests_pipe_lsu_mem_global_op_ldgsts.sum.pct_of_peak_sustained_elapsed |
0 % | L1TEX缓存查询阶段收到的全局到共享内存复制请求数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__sass_l1tex_t_requests_pipe_lsu_mem_global_op_ldgsts_cache_access.sum |
0 | L1TEX缓存查询阶段收到的经过L1缓存的全局到共享内存复制请求数;在同类硬件单元间求和 |
sm__sass_l1tex_t_requests_pipe_lsu_mem_global_op_ldgsts_cache_bypass.sum |
0 | L1TEX缓存查询阶段收到的绕过L1缓存的全局到共享内存复制请求数;在同类硬件单元间求和 |
sm__sass_l1tex_t_sectors_pipe_lsu_mem_global_op_ldgsts_cache_access.sum |
0 sector | L1TEX缓存查询阶段收到的经过L1缓存的全局到共享内存复制sector数;在同类硬件单元间求和 |
sm__sass_l1tex_t_sectors_pipe_lsu_mem_global_op_ldgsts_cache_access.sum.pct_of_peak_sustained_elapsed |
0 % | L1TEX缓存查询阶段收到的经过L1缓存的全局到共享内存复制sector数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__sass_l1tex_t_sectors_pipe_lsu_mem_global_op_ldgsts_cache_access.sum.per_second |
0 sector/s | L1TEX缓存查询阶段收到的经过L1缓存的全局到共享内存复制sector数;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
sm__sass_l1tex_t_sectors_pipe_lsu_mem_global_op_ldgsts_cache_bypass.sum |
0 sector | L1TEX缓存查询阶段收到的绕过L1缓存的全局到共享内存复制sector数;在同类硬件单元间求和 |
sm__sass_l1tex_t_sectors_pipe_lsu_mem_global_op_ldgsts_cache_bypass.sum.pct_of_peak_sustained_elapsed |
0 % | L1TEX缓存查询阶段收到的绕过L1缓存的全局到共享内存复制sector数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
sm__sass_thread_inst_executed_op_dfma_pred_on.avg.peak_sustained |
2 inst/cycle | DFMA指令中满足执行条件的线程级执行次数的每周期持续峰值;在同类硬件单元间取平均 |
sm__sass_thread_inst_executed_op_dfma_pred_on.max.peak_sustained |
2 inst/cycle | DFMA指令中满足执行条件的线程级执行次数的每周期持续峰值;在同类硬件单元间取最大值 |
sm__sass_thread_inst_executed_op_dfma_pred_on.min.peak_sustained |
2 inst/cycle | DFMA指令中满足执行条件的线程级执行次数的每周期持续峰值;在同类硬件单元间取最小值 |
sm__sass_thread_inst_executed_op_dfma_pred_on.sum.peak_sustained |
164 inst/cycle | DFMA指令中满足执行条件的线程级执行次数的每周期持续峰值;在同类硬件单元间求和 |
sm__sass_thread_inst_executed_op_ffma_pred_on.avg.peak_sustained |
128 inst/cycle | FFMA指令中满足执行条件的线程级执行次数的每周期持续峰值;在同类硬件单元间取平均 |
sm__sass_thread_inst_executed_op_ffma_pred_on.max.peak_sustained |
128 inst/cycle | FFMA指令中满足执行条件的线程级执行次数的每周期持续峰值;在同类硬件单元间取最大值 |
sm__sass_thread_inst_executed_op_ffma_pred_on.min.peak_sustained |
128 inst/cycle | FFMA指令中满足执行条件的线程级执行次数的每周期持续峰值;在同类硬件单元间取最小值 |
sm__sass_thread_inst_executed_op_ffma_pred_on.sum.peak_sustained |
10,496 inst/cycle | FFMA指令中满足执行条件的线程级执行次数的每周期持续峰值;在同类硬件单元间求和 |
sm__sass_thread_inst_executed_op_hfma_pred_on.avg.peak_sustained |
64 inst/cycle | HFMA指令中满足执行条件的线程级执行次数的每周期持续峰值;在同类硬件单元间取平均 |
sm__sass_thread_inst_executed_op_hfma_pred_on.max.peak_sustained |
64 inst/cycle | HFMA指令中满足执行条件的线程级执行次数的每周期持续峰值;在同类硬件单元间取最大值 |
sm__sass_thread_inst_executed_op_hfma_pred_on.min.peak_sustained |
64 inst/cycle | HFMA指令中满足执行条件的线程级执行次数的每周期持续峰值;在同类硬件单元间取最小值 |
sm__sass_thread_inst_executed_op_hfma_pred_on.sum.peak_sustained |
5,248 inst/cycle | HFMA指令中满足执行条件的线程级执行次数的每周期持续峰值;在同类硬件单元间求和 |
sm__throughput.avg.pct_of_peak_sustained_elapsed |
15.9779 % | 假定SMSP间负载均衡时的SM执行资源吞吐汇总;各组成计数器先按avg取得同类单元的平均计数,再除以相同统计范围内全部流逝周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__throughput.max.pct_of_peak_sustained_elapsed |
16.6332 % | 假定SMSP间负载均衡时的SM执行资源吞吐汇总;各组成计数器先按max取得同类单元的最大计数,再除以相同统计范围内全部流逝周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__throughput.min.pct_of_peak_sustained_elapsed |
15.6736 % | 假定SMSP间负载均衡时的SM执行资源吞吐汇总;各组成计数器先按min取得同类单元的最小计数,再除以相同统计范围内全部流逝周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__throughput.sum.pct_of_peak_sustained_elapsed |
15.9779 % | 假定SMSP间负载均衡时的SM执行资源吞吐汇总;各组成计数器先按sum取得同类单元的合计计数,再除以相同统计范围内全部流逝周期对应的持续峰值总量并乘100;最终取各组成项百分比的最大值 |
sm__warps_active.avg.pct_of_peak_sustained_active |
95.407 % | 跨周期累计的在途warp数;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__warps_active.avg.per_cycle_active |
45.7953 warp | 跨周期累计的在途warp数;在同类硬件单元间取平均并除以对应单元活跃周期数,得到每活跃周期的处理量 |
sm__warps_active.max.pct_of_peak_sustained_active |
96.2929 % | 跨周期累计的在途warp数;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__warps_active.max.per_cycle_active |
46.2206 warp | 跨周期累计的在途warp数;在同类硬件单元间取最大值并除以对应单元活跃周期数,得到每活跃周期的处理量 |
sm__warps_active.min.pct_of_peak_sustained_active |
94.5705 % | 跨周期累计的在途warp数;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__warps_active.min.per_cycle_active |
45.3939 warp | 跨周期累计的在途warp数;在同类硬件单元间取最小值并除以对应单元活跃周期数,得到每活跃周期的处理量 |
sm__warps_active.sum.pct_of_peak_sustained_active |
95.407 % | 跨周期累计的在途warp数;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
sm__warps_active.sum.per_cycle_active |
3755.22 warp | 跨周期累计的在途warp数;在同类硬件单元间求和并除以对应单元活跃周期数,得到每活跃周期的处理量 |
SMSP调度、指令与停顿采样
名称以per_issue_active结尾的状态比例,用SMSP实际发射指令的周期数作分母。本设备每个SMSP每周期最多发射一条warp级指令,因此也可以把结果理解为每次发射摊到的warp周期数;这里保留指标名称对应的“发射周期”口径。
| 项 | 值 | 说明 |
|---|---|---|
smsp__average_warp_latency_per_inst_issued.ratio |
79.9999 warp周期/指令 | 累计warp驻留周期除以warp级发射指令数,得到每条发射指令摊到的warp周期数 |
smsp__average_warps_active_per_inst_executed.ratio |
80.0958 warp周期/指令 | 累计warp驻留周期除以warp级执行指令数,得到每条执行指令摊到的warp周期数 |
smsp__average_warps_issue_stalled_barrier_per_issue_active.ratio |
32.0652 warp周期/发射周期 | 统计warp处于“等待同一block的其它warp到达同步点”状态的累计周期数,除以SMSP发射指令的周期数,得到每个发射周期摊到的warp周期数 |
smsp__average_warps_issue_stalled_branch_resolving_per_issue_active.ratio |
0.104553 warp周期/发射周期 | 统计warp处于“等待分支目标与程序计数器更新”状态的累计周期数,除以SMSP发射指令的周期数,得到每个发射周期摊到的warp周期数 |
smsp__average_warps_issue_stalled_dispatch_stall_per_issue_active.ratio |
0.173602 warp周期/发射周期 | 统计warp处于“指令准备好后仍在派发环节等待”状态的累计周期数,除以SMSP发射指令的周期数,得到每个发射周期摊到的warp周期数 |
smsp__average_warps_issue_stalled_drain_per_issue_active.ratio |
0.233013 warp周期/发射周期 | 统计warp处于“执行EXIT后等待未完成访存结束”状态的累计周期数,除以SMSP发射指令的周期数,得到每个发射周期摊到的warp周期数 |
smsp__average_warps_issue_stalled_imc_miss_per_issue_active.ratio |
0.100616 warp周期/发射周期 | 统计warp处于“等待立即数常量缓存缺失处理”状态的累计周期数,除以SMSP发射指令的周期数,得到每个发射周期摊到的warp周期数 |
smsp__average_warps_issue_stalled_lg_throttle_per_issue_active.ratio |
2.57224 warp周期/发射周期 | 统计warp处于“等待本地或全局访存指令队列腾位”状态的累计周期数,除以SMSP发射指令的周期数,得到每个发射周期摊到的warp周期数 |
smsp__average_warps_issue_stalled_long_scoreboard_per_issue_active.ratio |
2.64909 warp周期/发射周期 | 统计warp处于“等待L1TEX访存结果依赖”状态的累计周期数,除以SMSP发射指令的周期数,得到每个发射周期摊到的warp周期数 |
smsp__average_warps_issue_stalled_math_pipe_throttle_per_issue_active.ratio |
0.180914 warp周期/发射周期 | 统计warp处于“等待执行管线可以接收指令”状态的累计周期数,除以SMSP发射指令的周期数,得到每个发射周期摊到的warp周期数 |
smsp__average_warps_issue_stalled_membar_per_issue_active.ratio |
0 warp周期/发射周期 | 统计warp处于“等待内存屏障完成”状态的累计周期数,除以SMSP发射指令的周期数,得到每个发射周期摊到的warp周期数 |
smsp__average_warps_issue_stalled_mio_throttle_per_issue_active.ratio |
28.1336 warp周期/发射周期 | 统计warp处于“等待MIO指令队列腾位”状态的累计周期数,除以SMSP发射指令的周期数,得到每个发射周期摊到的warp周期数 |
smsp__average_warps_issue_stalled_misc_per_issue_active.ratio |
0.000174526 warp周期/发射周期 | 统计warp处于“其它硬件等待”状态的累计周期数,除以SMSP发射指令的周期数,得到每个发射周期摊到的warp周期数 |
smsp__average_warps_issue_stalled_no_instruction_per_issue_active.ratio |
0.129986 warp周期/发射周期 | 统计warp处于“等待取指或指令缓存缺失处理”状态的累计周期数,除以SMSP发射指令的周期数,得到每个发射周期摊到的warp周期数 |
smsp__average_warps_issue_stalled_not_selected_per_issue_active.ratio |
0.948209 warp周期/发射周期 | 统计warp处于“已经准备好,但调度器选中了其它warp”状态的累计周期数,除以SMSP发射指令的周期数,得到每个发射周期摊到的warp周期数 |
smsp__average_warps_issue_stalled_selected_per_issue_active.ratio |
0.999997 warp周期/发射周期 | 统计warp处于“被调度器选中并发出指令”状态的累计周期数,除以SMSP发射指令的周期数,得到每个发射周期摊到的warp周期数 |
smsp__average_warps_issue_stalled_short_scoreboard_per_issue_active.ratio |
8.34797 warp周期/发射周期 | 统计warp处于“等待共享内存等MIO操作的结果依赖”状态的累计周期数,除以SMSP发射指令的周期数,得到每个发射周期摊到的warp周期数 |
smsp__average_warps_issue_stalled_sleeping_per_issue_active.ratio |
0 warp周期/发射周期 | 统计warp处于“等待睡眠指令结束”状态的累计周期数,除以SMSP发射指令的周期数,得到每个发射周期摊到的warp周期数 |
smsp__average_warps_issue_stalled_tex_throttle_per_issue_active.ratio |
0.000668755 warp周期/发射周期 | 统计warp处于“等待纹理指令队列腾位”状态的累计周期数,除以SMSP发射指令的周期数,得到每个发射周期摊到的warp周期数 |
smsp__average_warps_issue_stalled_wait_per_issue_active.ratio |
3.22942 warp周期/发射周期 | 统计warp处于“等待固定延迟指令的结果依赖”状态的累计周期数,除以SMSP发射指令的周期数,得到每个发射周期摊到的warp周期数 |
smsp__branch_targets_threads_divergent |
0 {2} 次分支 | 按分支指令位置保存的线程分支目标分歧计数 |
smsp__cycles_active.avg |
438044 cycle | SMSP至少有一个warp驻留的周期数;在同类硬件单元间取平均 |
smsp__cycles_active.max |
441,146 cycle | SMSP至少有一个warp驻留的周期数;在同类硬件单元间取最大值 |
smsp__cycles_active.min |
435,339 cycle | SMSP至少有一个warp驻留的周期数;在同类硬件单元间取最小值 |
smsp__cycles_active.sum |
143,678,364 cycle | SMSP至少有一个warp驻留的周期数;在同类硬件单元间求和 |
smsp__cycles_elapsed.avg |
442680 cycle | SMSP自身时钟的流逝周期数;在同类硬件单元间取平均 |
smsp__cycles_elapsed.avg.per_second |
1.39019e+09 Hz | SMSP单元的流逝周期数按同类单元取平均值,再除以kernel执行时长,得到对应的时钟频率 |
smsp__cycles_elapsed.max |
442,842 cycle | SMSP自身时钟的流逝周期数;在同类硬件单元间取最大值 |
smsp__cycles_elapsed.max.per_second |
1.3907e+09 Hz | SMSP单元的流逝周期数按同类单元取最大值,再除以kernel执行时长,得到对应的时钟频率 |
smsp__cycles_elapsed.min |
442,486 cycle | SMSP自身时钟的流逝周期数;在同类硬件单元间取最小值 |
smsp__cycles_elapsed.min.per_second |
1.38958e+09 Hz | SMSP单元的流逝周期数按同类单元取最小值,再除以kernel执行时长,得到对应的时钟频率 |
smsp__cycles_elapsed.sum |
145,199,112 cycle | SMSP自身时钟的流逝周期数;在同类硬件单元间求和 |
smsp__cycles_elapsed.sum.per_second |
4.55982e+11 Hz | 全部SMSP单元的流逝周期数相加,再除以kernel执行时长,得到各单元时钟频率的合计值 |
smsp__inst_executed.avg |
62,464 inst | warp级执行指令数;在同类硬件单元间取平均 |
smsp__inst_executed.max |
65,026 inst | warp级执行指令数;在同类硬件单元间取最大值 |
smsp__inst_executed.min |
61,272 inst | warp级执行指令数;在同类硬件单元间取最小值 |
smsp__inst_executed.sum |
20,488,192 inst | warp级执行指令数;在同类硬件单元间求和 |
smsp__inst_executed_op_branch.avg |
2023.02 inst | CBU指令(排除BMOV与BSSY)的warp级执行次数;在同类硬件单元间取平均 |
smsp__inst_executed_op_branch.max |
2,106 inst | CBU指令(排除BMOV与BSSY)的warp级执行次数;在同类硬件单元间取最大值 |
smsp__inst_executed_op_branch.min |
1,984 inst | CBU指令(排除BMOV与BSSY)的warp级执行次数;在同类硬件单元间取最小值 |
smsp__inst_executed_op_branch.sum |
663,552 inst | CBU指令(排除BMOV与BSSY)的warp级执行次数;在同类硬件单元间求和 |
smsp__inst_executed_op_generic_atom_dot_alu.sum |
0 inst | 全局原子算术或逻辑指令的warp级执行次数;在同类硬件单元间求和 |
smsp__inst_executed_op_generic_atom_dot_alu.sum.pct_of_peak_sustained_elapsed |
0 % | 全局原子算术或逻辑指令的warp级执行次数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
smsp__inst_executed_op_generic_atom_dot_cas.sum |
0 inst | 全局原子比较交换指令的warp级执行次数;在同类硬件单元间求和 |
smsp__inst_executed_op_generic_atom_dot_cas.sum.pct_of_peak_sustained_elapsed |
0 % | 全局原子比较交换指令的warp级执行次数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
smsp__inst_executed_op_global_red.sum |
0 inst | 全局内存归约指令的warp级执行次数;在同类硬件单元间求和 |
smsp__inst_executed_op_global_red.sum.pct_of_peak_sustained_elapsed |
0 % | 全局内存归约指令的warp级执行次数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
smsp__inst_executed_op_ldgsts.sum |
0 inst | 全局到共享内存复制指令的warp级执行次数;在同类硬件单元间求和 |
smsp__inst_executed_op_ldgsts.sum.pct_of_peak_sustained_elapsed |
0 % | 全局到共享内存复制指令的warp级执行次数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
smsp__inst_executed_op_ldsm.sum |
0 inst | 共享内存矩阵加载指令的warp级执行次数;在同类硬件单元间求和 |
smsp__inst_executed_op_ldsm.sum.pct_of_peak_sustained_elapsed |
0 % | 共享内存矩阵加载指令的warp级执行次数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
smsp__inst_executed_op_shared_atom.sum |
0 inst | 共享内存原子指令的warp级执行次数;在同类硬件单元间求和 |
smsp__inst_executed_op_shared_atom.sum.pct_of_peak_sustained_elapsed |
0 % | 共享内存原子指令的warp级执行次数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
smsp__inst_executed_op_surface_atom_dot_alu.sum |
0 inst | surface原子算术或逻辑指令的warp级执行次数;在同类硬件单元间求和 |
smsp__inst_executed_op_surface_atom_dot_alu.sum.pct_of_peak_sustained_elapsed |
0 % | surface原子算术或逻辑指令的warp级执行次数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
smsp__inst_executed_op_surface_atom_dot_cas.sum |
0 inst | surface原子比较交换指令的warp级执行次数;在同类硬件单元间求和 |
smsp__inst_executed_op_surface_atom_dot_cas.sum.pct_of_peak_sustained_elapsed |
0 % | surface原子比较交换指令的warp级执行次数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
smsp__inst_executed_op_surface_ld.sum |
0 inst | surface加载指令的warp级执行次数;在同类硬件单元间求和 |
smsp__inst_executed_op_surface_ld.sum.pct_of_peak_sustained_elapsed |
0 % | surface加载指令的warp级执行次数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
smsp__inst_executed_op_surface_red.sum |
0 inst | surface归约指令的warp级执行次数;在同类硬件单元间求和 |
smsp__inst_executed_op_surface_red.sum.pct_of_peak_sustained_elapsed |
0 % | surface归约指令的warp级执行次数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
smsp__inst_executed_op_surface_st.sum |
0 inst | surface存储指令的warp级执行次数;在同类硬件单元间求和 |
smsp__inst_executed_op_surface_st.sum.pct_of_peak_sustained_elapsed |
0 % | surface存储指令的warp级执行次数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
smsp__inst_executed_op_texture.sum |
0 inst | 纹理指令的warp级执行次数;在同类硬件单元间求和 |
smsp__inst_executed_op_texture.sum.pct_of_peak_sustained_elapsed |
0 % | 纹理指令的warp级执行次数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
smsp__inst_issued.avg |
62538.9 inst | warp级发射指令数;在同类硬件单元间取平均 |
smsp__inst_issued.max |
65,104 inst | warp级发射指令数;在同类硬件单元间取最大值 |
smsp__inst_issued.min |
60,720 inst | warp级发射指令数;在同类硬件单元间取最小值 |
smsp__inst_issued.sum |
20,512,745 inst | warp级发射指令数;在同类硬件单元间求和 |
smsp__issue_active.avg.pct_of_peak_sustained_active |
14.2769 % | 发射了一条指令的周期数;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
smsp__issue_active.avg.per_cycle_active |
0.142769 | 发射了一条指令的周期数;在同类硬件单元间取平均并除以对应单元活跃周期数,得到每活跃周期的处理量 |
smsp__issue_active.max.pct_of_peak_sustained_active |
14.8624 % | 发射了一条指令的周期数;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
smsp__issue_active.max.per_cycle_active |
0.148624 | 发射了一条指令的周期数;在同类硬件单元间取最大值并除以对应单元活跃周期数,得到每活跃周期的处理量 |
smsp__issue_active.min.pct_of_peak_sustained_active |
13.8616 % | 发射了一条指令的周期数;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
smsp__issue_active.min.per_cycle_active |
0.138616 | 发射了一条指令的周期数;在同类硬件单元间取最小值并除以对应单元活跃周期数,得到每活跃周期的处理量 |
smsp__issue_active.sum.pct_of_peak_sustained_active |
14.2769 % | 发射了一条指令的周期数;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
smsp__issue_active.sum.per_cycle_active |
46.8281 | 发射了一条指令的周期数;在同类硬件单元间求和并除以对应单元活跃周期数,得到每活跃周期的处理量 |
smsp__issue_inst0.avg.pct_of_peak_sustained_active |
85.7231 % | 没有发射任何指令的活跃周期数;在同类硬件单元间取平均,以对应单元活跃周期内的持续峰值归一化为百分比 |
smsp__issue_inst0.max.pct_of_peak_sustained_active |
85.8458 % | 没有发射任何指令的活跃周期数;在同类硬件单元间取最大值,以对应单元活跃周期内的持续峰值归一化为百分比 |
smsp__issue_inst0.min.pct_of_peak_sustained_active |
85.5209 % | 没有发射任何指令的活跃周期数;在同类硬件单元间取最小值,以对应单元活跃周期内的持续峰值归一化为百分比 |
smsp__issue_inst0.sum.pct_of_peak_sustained_active |
85.7231 % | 没有发射任何指令的活跃周期数;在同类硬件单元间求和,以对应单元活跃周期内的持续峰值归一化为百分比 |
smsp__maximum_warps_avg_per_active_cycle |
12 warp | 按本次启动资源限制计算的每SMSP理论最大驻留warp数 |
smsp__pcsamp_aggregated_passes |
2 次 | 统计warp停顿采样所用的采集次数 |
smsp__pcsamp_buffer_size_bytes |
33,554,432 byte | warp停顿采样缓冲区字节数 |
smsp__pcsamp_dropped_bytes |
0 byte | warp停顿采样因缓冲区不足而丢弃的字节数 |
smsp__pcsamp_interval |
6 | warp停顿采样间隔的配置编号 |
smsp__pcsamp_interval_cycles |
2,048 cycle | 两次warp停顿采样间隔的时钟周期数 |
smsp__pcsamp_sample_count |
17,547 {112} sample | warp停顿采样总样本数,实例按指令位置区分 |
smsp__pcsamp_warps_issue_stalled_barrier |
6,602 {112} sample | warp停顿采样的样本数,记录等待同一block的其它warp到达同步点;按全部采样周期口径统计 |
smsp__pcsamp_warps_issue_stalled_barrier_not_issued |
5,714 {112} sample | warp停顿采样的样本数,记录等待同一block的其它warp到达同步点;仅统计调度器未发射指令的周期中取得的样本 |
smsp__pcsamp_warps_issue_stalled_branch_resolving |
29 {112} sample | warp停顿采样的样本数,记录等待分支目标与程序计数器更新;按全部采样周期口径统计 |
smsp__pcsamp_warps_issue_stalled_branch_resolving_not_issued |
26 {112} sample | warp停顿采样的样本数,记录等待分支目标与程序计数器更新;仅统计调度器未发射指令的周期中取得的样本 |
smsp__pcsamp_warps_issue_stalled_dispatch_stall |
41 {112} sample | warp停顿采样的样本数,记录指令准备好后仍在派发环节等待;按全部采样周期口径统计 |
smsp__pcsamp_warps_issue_stalled_dispatch_stall_not_issued |
20 {112} sample | warp停顿采样的样本数,记录指令准备好后仍在派发环节等待;仅统计调度器未发射指令的周期中取得的样本 |
smsp__pcsamp_warps_issue_stalled_drain |
44 {112} sample | warp停顿采样的样本数,记录执行EXIT后等待未完成访存结束;按全部采样周期口径统计 |
smsp__pcsamp_warps_issue_stalled_drain_not_issued |
38 {112} sample | warp停顿采样的样本数,记录执行EXIT后等待未完成访存结束;仅统计调度器未发射指令的周期中取得的样本 |
smsp__pcsamp_warps_issue_stalled_imc_miss |
10 {112} sample | warp停顿采样的样本数,记录等待立即数常量缓存缺失处理;按全部采样周期口径统计 |
smsp__pcsamp_warps_issue_stalled_imc_miss_not_issued |
10 {112} sample | warp停顿采样的样本数,记录等待立即数常量缓存缺失处理;仅统计调度器未发射指令的周期中取得的样本 |
smsp__pcsamp_warps_issue_stalled_lg_throttle |
552 {112} sample | warp停顿采样的样本数,记录等待本地或全局访存指令队列腾位;按全部采样周期口径统计 |
smsp__pcsamp_warps_issue_stalled_lg_throttle_not_issued |
479 {112} sample | warp停顿采样的样本数,记录等待本地或全局访存指令队列腾位;仅统计调度器未发射指令的周期中取得的样本 |
smsp__pcsamp_warps_issue_stalled_long_scoreboard |
690 {112} sample | warp停顿采样的样本数,记录等待L1TEX访存结果依赖;按全部采样周期口径统计 |
smsp__pcsamp_warps_issue_stalled_long_scoreboard_not_issued |
573 {112} sample | warp停顿采样的样本数,记录等待L1TEX访存结果依赖;仅统计调度器未发射指令的周期中取得的样本 |
smsp__pcsamp_warps_issue_stalled_math_pipe_throttle |
48 {112} sample | warp停顿采样的样本数,记录等待执行管线可以接收指令;按全部采样周期口径统计 |
smsp__pcsamp_warps_issue_stalled_math_pipe_throttle_not_issued |
18 {112} sample | warp停顿采样的样本数,记录等待执行管线可以接收指令;仅统计调度器未发射指令的周期中取得的样本 |
smsp__pcsamp_warps_issue_stalled_membar |
0 {112} sample | warp停顿采样的样本数,记录等待内存屏障完成;按全部采样周期口径统计 |
smsp__pcsamp_warps_issue_stalled_membar_not_issued |
0 {112} sample | warp停顿采样的样本数,记录等待内存屏障完成;仅统计调度器未发射指令的周期中取得的样本 |
smsp__pcsamp_warps_issue_stalled_mio_throttle |
6,225 {112} sample | warp停顿采样的样本数,记录等待MIO指令队列腾位;按全部采样周期口径统计 |
smsp__pcsamp_warps_issue_stalled_mio_throttle_not_issued |
5,688 {112} sample | warp停顿采样的样本数,记录等待MIO指令队列腾位;仅统计调度器未发射指令的周期中取得的样本 |
smsp__pcsamp_warps_issue_stalled_misc |
0 {112} sample | warp停顿采样的样本数,记录其它硬件等待;按全部采样周期口径统计 |
smsp__pcsamp_warps_issue_stalled_misc_not_issued |
0 {112} sample | warp停顿采样的样本数,记录其它硬件等待;仅统计调度器未发射指令的周期中取得的样本 |
smsp__pcsamp_warps_issue_stalled_no_instructions |
25 {112} sample | warp停顿采样的样本数,记录等待取指或指令缓存缺失处理;按全部采样周期口径统计 |
smsp__pcsamp_warps_issue_stalled_no_instructions_not_issued |
14 {112} sample | warp停顿采样的样本数,记录等待取指或指令缓存缺失处理;仅统计调度器未发射指令的周期中取得的样本 |
smsp__pcsamp_warps_issue_stalled_not_selected |
207 {112} sample | warp停顿采样的样本数,记录已经准备好,但调度器选中了其它warp;按全部采样周期口径统计 |
smsp__pcsamp_warps_issue_stalled_not_selected_not_issued |
0 {112} sample | warp停顿采样的样本数,记录已经准备好,但调度器选中了其它warp;仅统计调度器未发射指令的周期中取得的样本 |
smsp__pcsamp_warps_issue_stalled_selected |
256 {112} sample | warp停顿采样的样本数,记录被调度器选中并发出指令;按全部采样周期口径统计 |
smsp__pcsamp_warps_issue_stalled_selected_not_issued |
0 {112} sample | warp停顿采样的样本数,记录被调度器选中并发出指令;仅统计调度器未发射指令的周期中取得的样本 |
smsp__pcsamp_warps_issue_stalled_short_scoreboard |
2,015 {112} sample | warp停顿采样的样本数,记录等待共享内存等MIO操作的结果依赖;按全部采样周期口径统计 |
smsp__pcsamp_warps_issue_stalled_short_scoreboard_not_issued |
1,693 {112} sample | warp停顿采样的样本数,记录等待共享内存等MIO操作的结果依赖;仅统计调度器未发射指令的周期中取得的样本 |
smsp__pcsamp_warps_issue_stalled_sleeping |
0 {112} sample | warp停顿采样的样本数,记录等待睡眠指令结束;按全部采样周期口径统计 |
smsp__pcsamp_warps_issue_stalled_sleeping_not_issued |
0 {112} sample | warp停顿采样的样本数,记录等待睡眠指令结束;仅统计调度器未发射指令的周期中取得的样本 |
smsp__pcsamp_warps_issue_stalled_tex_throttle |
0 {112} sample | warp停顿采样的样本数,记录等待纹理指令队列腾位;按全部采样周期口径统计 |
smsp__pcsamp_warps_issue_stalled_tex_throttle_not_issued |
0 {112} sample | warp停顿采样的样本数,记录等待纹理指令队列腾位;仅统计调度器未发射指令的周期中取得的样本 |
smsp__pcsamp_warps_issue_stalled_wait |
803 {112} sample | warp停顿采样的样本数,记录等待固定延迟指令的结果依赖;按全部采样周期口径统计 |
smsp__pcsamp_warps_issue_stalled_wait_not_issued |
527 {112} sample | warp停顿采样的样本数,记录等待固定延迟指令的结果依赖;仅统计调度器未发射指令的周期中取得的样本 |
smsp__sass_average_branch_targets_threads_uniform.pct |
100 % | 所有活跃线程选择相同分支目标的比例,以百分比表示;按该指标的汇总口径统计 |
smsp__sass_average_data_bytes_per_sector_mem_global_op_ld.max_rate |
32 byte/sector | 全局加载每个sector平均使用的字节数的理论上限;32 byte/sector表示一个32字节sector全部得到利用 |
smsp__sass_average_data_bytes_per_sector_mem_global_op_ld.ratio |
4 byte/sector | 全局加载中线程实际请求的字节数除以对应sector数,得到每个32字节sector平均使用的字节数 |
smsp__sass_average_data_bytes_per_sector_mem_global_op_st.max_rate |
32 byte/sector | 全局存储每个sector平均使用的字节数的理论上限;32 byte/sector表示一个32字节sector全部得到利用 |
smsp__sass_average_data_bytes_per_sector_mem_global_op_st.ratio |
4 byte/sector | 全局存储中线程实际请求的字节数除以对应sector数,得到每个32字节sector平均使用的字节数 |
smsp__sass_average_data_bytes_per_sector_mem_local_op_ld.max_rate |
32 byte/sector | 线程本地内存加载每个sector平均使用的字节数的理论上限;32 byte/sector表示一个32字节sector全部得到利用 |
smsp__sass_average_data_bytes_per_sector_mem_local_op_ld.ratio |
0 byte/sector | 线程本地内存加载中线程实际请求的字节数除以对应sector数,得到每个32字节sector平均使用的字节数 |
smsp__sass_average_data_bytes_per_sector_mem_local_op_st.max_rate |
32 byte/sector | 线程本地内存存储每个sector平均使用的字节数的理论上限;32 byte/sector表示一个32字节sector全部得到利用 |
smsp__sass_average_data_bytes_per_sector_mem_local_op_st.ratio |
0 byte/sector | 线程本地内存存储中线程实际请求的字节数除以对应sector数,得到每个32字节sector平均使用的字节数 |
smsp__sass_branch_targets_threads_divergent.avg |
0 | 同一warp有至少两个活跃线程选择不同分支目标时的分歧计数;在同类硬件单元间取平均 |
smsp__sass_branch_targets_threads_divergent.max |
0 | 同一warp有至少两个活跃线程选择不同分支目标时的分歧计数;在同类硬件单元间取最大值 |
smsp__sass_branch_targets_threads_divergent.min |
0 | 同一warp有至少两个活跃线程选择不同分支目标时的分歧计数;在同类硬件单元间取最小值 |
smsp__sass_branch_targets_threads_divergent.sum |
0 | 同一warp有至少两个活跃线程选择不同分支目标时的分歧计数;在同类硬件单元间求和 |
smsp__sass_inst_executed_memdesc_explicit_evict_type |
0 | 逐访存指令的显式缓存淘汰类型编码,按实例读取 |
smsp__sass_inst_executed_memdesc_explicit_hitprop_evict_first |
0 {9} inst | 显式指定访存描述符淘汰策略的warp级执行指令数,命中时策略为EVICT_FIRST;按该指标的汇总口径统计 |
smsp__sass_inst_executed_memdesc_explicit_hitprop_evict_last |
0 {9} inst | 显式指定访存描述符淘汰策略的warp级执行指令数,命中时策略为EVICT_LAST;按该指标的汇总口径统计 |
smsp__sass_inst_executed_memdesc_explicit_hitprop_evict_normal |
0 {9} inst | 显式指定访存描述符淘汰策略的warp级执行指令数,命中时策略为EVICT_NORMAL;按该指标的汇总口径统计 |
smsp__sass_inst_executed_memdesc_explicit_hitprop_evict_normal_demote |
0 {9} inst | 显式指定访存描述符淘汰策略的warp级执行指令数,命中时策略为EVICT_NORMAL_DEMOTE;按该指标的汇总口径统计 |
smsp__sass_inst_executed_memdesc_explicit_missprop_evict_first |
0 {9} inst | 显式指定访存描述符淘汰策略的warp级执行指令数,未命中时策略为EVICT_FIRST;按该指标的汇总口径统计 |
smsp__sass_inst_executed_memdesc_explicit_missprop_evict_normal |
0 {9} inst | 显式指定访存描述符淘汰策略的warp级执行指令数,未命中时策略为EVICT_NORMAL;按该指标的汇总口径统计 |
smsp__sass_inst_executed_op_global_ld.sum |
524,288 inst | 全局内存加载指令的warp级执行次数;在同类硬件单元间求和 |
smsp__sass_inst_executed_op_global_st.sum |
8,192 inst | 全局内存存储指令的warp级执行次数;在同类硬件单元间求和 |
smsp__sass_inst_executed_op_local_ld.sum |
0 inst | 线程本地内存加载指令的warp级执行次数;在同类硬件单元间求和 |
smsp__sass_inst_executed_op_local_st.sum |
0 inst | 线程本地内存存储指令的warp级执行次数;在同类硬件单元间求和 |
smsp__sass_inst_executed_op_shared_ld.sum |
909,312 inst | 共享内存加载指令的warp级执行次数;在同类硬件单元间求和 |
smsp__sass_inst_executed_op_shared_st.sum |
974,848 inst | 共享内存存储指令的warp级执行次数;在同类硬件单元间求和 |
smsp__sass_l1tex_data_bank_conflicts_pipe_lsu_mem_shared_op_ldgsts.sum |
0 | 全局到共享内存复制产生的共享内存bank冲突数;在同类硬件单元间求和 |
smsp__sass_l1tex_data_bank_conflicts_pipe_lsu_mem_shared_op_st.sum |
0 | 共享内存存储产生的共享内存bank冲突数;在同类硬件单元间求和 |
smsp__sass_l1tex_data_pipe_lsu_wavefronts_mem_shared_op_ldgsts.sum |
0 | 全局到共享内存复制在L1TEX数据阶段处理的共享内存wavefront数;在同类硬件单元间求和 |
smsp__sass_l1tex_data_pipe_lsu_wavefronts_mem_shared_op_ldgsts.sum.pct_of_peak_sustained_elapsed |
0 % | 全局到共享内存复制在L1TEX数据阶段处理的共享内存wavefront数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
smsp__sass_l1tex_data_pipe_lsu_wavefronts_mem_shared_op_ldgsts_cache_access.sum |
0 | 全局到共享内存复制在L1TEX数据阶段处理的共享内存wavefront数;在同类硬件单元间求和 |
smsp__sass_l1tex_data_pipe_lsu_wavefronts_mem_shared_op_ldgsts_cache_access.sum.pct_of_peak_sustained_elapsed |
0 % | 全局到共享内存复制在L1TEX数据阶段处理的共享内存wavefront数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
smsp__sass_l1tex_data_pipe_lsu_wavefronts_mem_shared_op_st.sum |
974,848 | 共享内存存储在L1TEX数据阶段处理的共享内存wavefront数;在同类硬件单元间求和 |
smsp__sass_l1tex_data_pipe_lsu_wavefronts_mem_shared_op_st.sum.pct_of_peak_sustained_elapsed |
0.671387 % | 共享内存存储在L1TEX数据阶段处理的共享内存wavefront数;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
smsp__sass_l1tex_m_xbar2l1tex_read_sectors_mem_global_op_ldgsts_cache_bypass.sum |
0 sector | 从L2读入L1TEX访存阶段的sector数,用于绕过L1缓存的全局到共享内存复制;在同类硬件单元间求和 |
smsp__sass_l1tex_m_xbar2l1tex_read_sectors_mem_global_op_ldgsts_cache_bypass.sum.pct_of_peak_sustained_elapsed |
0 % | 从L2读入L1TEX访存阶段的sector数,用于绕过L1缓存的全局到共享内存复制;在同类硬件单元间求和,以对应单元全部流逝周期内的持续峰值归一化为百分比 |
smsp__sass_thread_inst_executed_op_dadd_pred_on.avg.per_cycle_elapsed |
0 inst/cycle | DADD指令中满足执行条件的线程级执行次数;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_dadd_pred_on.max.per_cycle_elapsed |
0 inst/cycle | DADD指令中满足执行条件的线程级执行次数;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_dadd_pred_on.min.per_cycle_elapsed |
0 inst/cycle | DADD指令中满足执行条件的线程级执行次数;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_dadd_pred_on.sum.per_cycle_elapsed |
0 inst/cycle | DADD指令中满足执行条件的线程级执行次数;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_dfma_pred_on.avg.per_cycle_elapsed |
0 inst/cycle | DFMA指令中满足执行条件的线程级执行次数;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_dfma_pred_on.max.per_cycle_elapsed |
0 inst/cycle | DFMA指令中满足执行条件的线程级执行次数;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_dfma_pred_on.min.per_cycle_elapsed |
0 inst/cycle | DFMA指令中满足执行条件的线程级执行次数;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_dfma_pred_on.sum.per_cycle_elapsed |
0 inst/cycle | DFMA指令中满足执行条件的线程级执行次数;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_dmul_pred_on.avg.per_cycle_elapsed |
0 inst/cycle | DMUL指令中满足执行条件的线程级执行次数;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_dmul_pred_on.max.per_cycle_elapsed |
0 inst/cycle | DMUL指令中满足执行条件的线程级执行次数;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_dmul_pred_on.min.per_cycle_elapsed |
0 inst/cycle | DMUL指令中满足执行条件的线程级执行次数;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_dmul_pred_on.sum.per_cycle_elapsed |
0 inst/cycle | DMUL指令中满足执行条件的线程级执行次数;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_fadd_pred_on.avg.per_cycle_elapsed |
0.129933 inst/cycle | FADD指令中满足执行条件的线程级执行次数;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_fadd_pred_on.max.per_cycle_elapsed |
0.15738 inst/cycle | FADD指令中满足执行条件的线程级执行次数;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_fadd_pred_on.min.per_cycle_elapsed |
0.110547 inst/cycle | FADD指令中满足执行条件的线程级执行次数;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_fadd_pred_on.sum.per_cycle_elapsed |
42.6181 inst/cycle | FADD指令中满足执行条件的线程级执行次数;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_ffma_pred_on.avg.per_cycle_elapsed |
0 inst/cycle | FFMA指令中满足执行条件的线程级执行次数;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_ffma_pred_on.max.per_cycle_elapsed |
0 inst/cycle | FFMA指令中满足执行条件的线程级执行次数;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_ffma_pred_on.min.per_cycle_elapsed |
0 inst/cycle | FFMA指令中满足执行条件的线程级执行次数;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_ffma_pred_on.sum.per_cycle_elapsed |
0 inst/cycle | FFMA指令中满足执行条件的线程级执行次数;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_fmul_pred_on.avg.per_cycle_elapsed |
0 inst/cycle | FMUL指令中满足执行条件的线程级执行次数;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_fmul_pred_on.max.per_cycle_elapsed |
0 inst/cycle | FMUL指令中满足执行条件的线程级执行次数;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_fmul_pred_on.min.per_cycle_elapsed |
0 inst/cycle | FMUL指令中满足执行条件的线程级执行次数;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_fmul_pred_on.sum.per_cycle_elapsed |
0 inst/cycle | FMUL指令中满足执行条件的线程级执行次数;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_hfma_pred_on.avg.per_cycle_elapsed |
0 inst/cycle | HFMA指令中满足执行条件的线程级执行次数;在同类硬件单元间取平均并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_hfma_pred_on.max.per_cycle_elapsed |
0 inst/cycle | HFMA指令中满足执行条件的线程级执行次数;在同类硬件单元间取最大值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_hfma_pred_on.min.per_cycle_elapsed |
0 inst/cycle | HFMA指令中满足执行条件的线程级执行次数;在同类硬件单元间取最小值并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__sass_thread_inst_executed_op_hfma_pred_on.sum.per_cycle_elapsed |
0 inst/cycle | HFMA指令中满足执行条件的线程级执行次数;在同类硬件单元间求和并除以对应单元全部流逝周期数,得到每流逝周期的处理量 |
smsp__thread_inst_executed_per_inst_executed.ratio |
31.938 | 每条warp级执行指令平均对应的活跃线程数 |
smsp__thread_inst_executed_pred_on_per_inst_executed.ratio |
25.7905 | 每条warp级执行指令平均实际参与操作的活跃线程数 |
smsp__warps_active.avg.peak_sustained |
12 warp | 跨周期累计的活跃warp数,即warp驻留周期累计量的每周期持续峰值;在同类硬件单元间取平均 |
smsp__warps_active.avg.per_cycle_active |
11.4215 warp | 跨周期累计的活跃warp数,即warp驻留周期累计量;在同类硬件单元间取平均并除以对应单元活跃周期数,得到每活跃周期的处理量 |
smsp__warps_active.max.peak_sustained |
12 warp | 跨周期累计的活跃warp数,即warp驻留周期累计量的每周期持续峰值;在同类硬件单元间取最大值 |
smsp__warps_active.max.per_cycle_active |
11.5033 warp | 跨周期累计的活跃warp数,即warp驻留周期累计量;在同类硬件单元间取最大值并除以对应单元活跃周期数,得到每活跃周期的处理量 |
smsp__warps_active.min.peak_sustained |
12 warp | 跨周期累计的活跃warp数,即warp驻留周期累计量的每周期持续峰值;在同类硬件单元间取最小值 |
smsp__warps_active.min.per_cycle_active |
11.3449 warp | 跨周期累计的活跃warp数,即warp驻留周期累计量;在同类硬件单元间取最小值并除以对应单元活跃周期数,得到每活跃周期的处理量 |
smsp__warps_active.sum.peak_sustained |
3,936 warp | 跨周期累计的活跃warp数,即warp驻留周期累计量的每周期持续峰值;在同类硬件单元间求和 |
smsp__warps_active.sum.per_cycle_active |
3746.24 warp | 跨周期累计的活跃warp数,即warp驻留周期累计量;在同类硬件单元间求和并除以对应单元活跃周期数,得到每活跃周期的处理量 |
smsp__warps_eligible.avg.per_cycle_active |
0.278029 warp | 跨周期累计的合格warp数,即warp具备发射条件的周期累计量;在同类硬件单元间取平均并除以对应单元活跃周期数,得到每活跃周期的处理量 |
smsp__warps_eligible.max.per_cycle_active |
0.301486 warp | 跨周期累计的合格warp数,即warp具备发射条件的周期累计量;在同类硬件单元间取最大值并除以对应单元活跃周期数,得到每活跃周期的处理量 |
smsp__warps_eligible.min.per_cycle_active |
0.260755 warp | 跨周期累计的合格warp数,即warp具备发射条件的周期累计量;在同类硬件单元间取最小值并除以对应单元活跃周期数,得到每活跃周期的处理量 |
smsp__warps_eligible.sum.per_cycle_active |
91.1937 warp | 跨周期累计的合格warp数,即warp具备发射条件的周期累计量;在同类硬件单元间求和并除以对应单元活跃周期数,得到每活跃周期的处理量 |
Source指令关联与派生值
| 项 | 值 | 说明 |
|---|---|---|
derived__avg_thread_executed |
3,717 {121} thread | 逐指令的活跃线程指令数除以warp级执行次数,得到每次执行平均活跃线程数;实例按指令区分,汇总值由多个指令实例聚合 |
derived__avg_thread_executed_true |
3,442 {121} thread | 逐指令满足执行条件的线程指令数除以warp级执行次数,得到每次执行平均实际参与线程数;实例按指令区分,汇总值由多个指令实例聚合 |
derived__l1tex__lsu_writeback_bytes_mem_lg.sum.peak_sustained |
10,496 byte/cycle | L1TEX本地或全局数据返回通路的每周期持续峰值乘128字节,换算为byte/周期的理论带宽 |
derived__l1tex__lsu_writeback_bytes_mem_lg.sum.per_second |
2.57548e+12 byte/s | L1TEX本地或全局数据返回通路的每秒实际活跃计数乘128字节,换算为byte/s的实际带宽 |
derived__lts__lts2xbar_bytes.sum.peak_sustained |
1,536 byte/cycle | L2向交叉开关通路的每周期持续峰值乘32字节,换算为byte/周期的理论带宽 |
derived__lts__lts2xbar_bytes.sum.per_second |
1.6856e+12 byte/s | L2向交叉开关通路的每秒实际活跃计数乘32字节,换算为byte/s的实际带宽 |
derived__memory_l1_conflicts_shared_nway |
26 | 逐指令的共享内存wavefront数除以执行次数,得到平均每次执行需要多少wavefront;默认汇总值包含多个指令实例 |
derived__memory_l1_wavefronts_shared_excessive |
524,288 {15} wavefront | 实际共享内存wavefront数减去理想wavefront数,得到多余wavefront总数,单位wavefront |
derived__memory_l2_theoretical_sectors_global_excessive |
14,680,064 {9} sector | 全局访存理论sector数减去理想sector数,得到多余sector总数,单位sector |
derived__pct_occupancy_per_block_size |
4,754 {65} % | 各参数取值下的理论驻留warp数除以每SM最大warp数,再乘100,得到occupancy曲线;按实例读取各点,汇总值包含多个点 |
derived__pct_occupancy_per_register_count |
10,252 {257} % | 各参数取值下的理论驻留warp数除以每SM最大warp数,再乘100,得到occupancy曲线;按实例读取各点,汇总值包含多个点 |
derived__pct_occupancy_per_shared_mem_size |
4,732 {200} % | 各参数取值下的理论驻留warp数除以每SM最大warp数,再乘100,得到occupancy曲线;按实例读取各点,汇总值包含多个点 |
derived__sm__sass_thread_inst_executed_op_dfma_pred_on_x2 |
328 数学运算/周期(峰值) | DFMA线程操作的理论每周期持续峰值乘2,换算成数学运算量;一次双精度乘加包含一次乘法和一次加法,共2次数学运算 |
derived__sm__sass_thread_inst_executed_op_ffma_pred_on_x2 |
20,992 数学运算/周期(峰值) | FFMA线程操作的理论每周期持续峰值乘2,换算成数学运算量;一次单精度乘加包含一次乘法和一次加法,共2次数学运算 |
derived__sm__sass_thread_inst_executed_op_hfma_pred_on_x4 |
20,992 数学运算/周期(峰值) | HFMA线程操作的理论每周期持续峰值乘4,换算成数学运算量;打包半精度指令处理两个元素,每个元素各做一次乘法和一次加法,共4次数学运算 |
derived__smsp__inst_executed_op_branch_pct |
0.032387 比例 | 分支指令数除以全部执行指令数得到比例;本字段存0到1的比例,换成百分比还需乘100 |
derived__smsp__sass_thread_inst_executed_op_dfma_pred_on_x2 |
0 数学运算/流逝周期 | DFMA线程操作的实际每流逝周期执行速率乘2,换算成数学运算量;一次双精度乘加包含一次乘法和一次加法,共2次数学运算 |
derived__smsp__sass_thread_inst_executed_op_ffma_pred_on_x2 |
0 数学运算/流逝周期 | FFMA线程操作的实际每流逝周期执行速率乘2,换算成数学运算量;一次单精度乘加包含一次乘法和一次加法,共2次数学运算 |
derived__smsp__sass_thread_inst_executed_op_hadd_pred_on_x2 |
N/A 数学运算/流逝周期 | HADD线程操作的实际每流逝周期执行速率乘2,换算成数学运算量;打包半精度指令处理两个元素,每个元素各做一次加法,共2次数学运算 |
derived__smsp__sass_thread_inst_executed_op_hfma_pred_on_x4 |
0 数学运算/流逝周期 | HFMA线程操作的实际每流逝周期执行速率乘4,换算成数学运算量;打包半精度指令处理两个元素,每个元素各做一次乘法和一次加法,共4次数学运算 |
derived__smsp__sass_thread_inst_executed_op_hmul_pred_on_x2 |
N/A 数学运算/流逝周期 | HMUL线程操作的实际每流逝周期执行速率乘2,换算成数学运算量;打包半精度指令处理两个元素,每个元素各做一次乘法,共2次数学运算 |
derived_tempMetric0 |
228,800 | 工具生成的临时作图字段,包含200个实例;本报告未给出公开含义,具体算法不作推定 |
inst_executed |
20,488,192 {121} inst | 按SASS指令地址保存的warp级执行次数 |
memory_access_size_type |
0 | 逐指令访存宽度类型编码,具体类型按各实例读取 |
memory_access_type |
0 | 逐指令读、写等访存操作类型编码,具体类型按各实例读取 |
memory_l1_tag_requests_global |
16,785,408 {9} sector | 全局访存指令产生的L1缓存标记查询量,按指令记录;本报告该字段单位为sector |
memory_l1_wavefronts_shared |
2,801,664 {15} wavefront | 共享内存访问的实际wavefront数,按指令记录 |
memory_l1_wavefronts_shared_ideal |
2,277,376 {15} wavefront | 共享内存访问理想情况下需要的wavefront数,按指令记录 |
memory_l2_theoretical_sectors_global |
16,785,408 {9} sector | 按指令访问地址计算的全局访存理论sector数 |
memory_l2_theoretical_sectors_global_ideal |
2,105,344 {9} sector | 全局访存理想合并访问所需的sector数 |
memory_l2_theoretical_sectors_local |
0 sector | 按指令地址计算的线程本地访存理论sector数 |
memory_type |
0 | 逐指令地址空间类型编码,具体地址空间按各实例读取 |
sass__inst_executed_global_loads |
524,288 inst | 全局内存加载的动态访存指令计数,按实际形成请求的执行口径统计 |
sass__inst_executed_global_stores |
8,192 inst | 全局内存存储的动态访存指令计数,按实际形成请求的执行口径统计 |
sass__inst_executed_local_loads |
0 inst | 线程本地内存加载的动态访存指令计数,按实际形成请求的执行口径统计 |
sass__inst_executed_local_stores |
0 inst | 线程本地内存存储的动态访存指令计数,按实际形成请求的执行口径统计 |
sass__inst_executed_per_opcode |
20,488,192 | warp级执行指令数,按操作码分类保存 |
sass__inst_executed_per_opcode_with_modifier_all |
20,488,192 | warp级执行指令数,按操作码及全部修饰符分类保存 |
sass__inst_executed_per_opcode_with_modifier_selective |
20,488,192 | warp级执行指令数,按操作码及选定修饰符分类保存 |
sass__inst_executed_shared_loads |
909,312 inst | 共享内存加载的动态访存指令计数,按实际形成请求的执行口径统计 |
sass__inst_executed_shared_stores |
974,848 inst | 共享内存存储的动态访存指令计数,按实际形成请求的执行口径统计 |
sass__thread_inst_executed_true_per_opcode |
528,400,384 | 满足执行条件的线程级执行次数,按操作码分类保存 |
sass__thread_inst_executed_true_per_opcode_with_modifier_all |
528,400,384 | 满足执行条件的线程级执行次数,按操作码及全部修饰符分类保存 |
sass__thread_inst_executed_true_per_opcode_with_modifier_selective |
528,400,384 | 满足执行条件的线程级执行次数,按操作码及选定修饰符分类保存 |
thread_inst_executed |
654,352,384 {121} inst | 按SASS指令地址保存的活跃线程指令数 |
thread_inst_executed_true |
528,400,384 {121} inst | 按SASS指令地址保存的满足执行条件的线程指令数 |
PM时间序列
| 项 | 值 | 说明 |
|---|---|---|
FBSP.TriageSCG.dramc__read_throughput.avg.pct_of_peak_sustained_elapsed |
N/A {466} % | PM时间序列:显存读方向相对持续峰值的速率;该项没有可用汇总值,各采样窗口的数值按实例读取 |
FBSP.TriageSCG.dramc__throughput.avg.pct_of_peak_sustained_elapsed |
N/A {466} % | PM时间序列:显存吞吐;该项没有可用汇总值,各采样窗口的数值按实例读取 |
FBSP.TriageSCG.dramc__write_throughput.avg.pct_of_peak_sustained_elapsed |
N/A {466} % | PM时间序列:显存写方向相对持续峰值的速率;该项没有可用汇总值,各采样窗口的数值按实例读取 |
FE_B.TriageAC.gr__ctas_launched_queue_sync.sum |
8,192 {466} block | PM时间序列:新启动的block数;默认值跨多个采样窗口聚合,各窗口数值按实例读取 |
LTS.TriageSCG.lts__average_t_sector_hit_rate_realtime.pct |
N/A {458} % | PM时间序列:L2 sector命中率;该项没有可用汇总值,各采样窗口的数值按实例读取 |
LTS.TriageSCG.lts__throughput.avg.pct_of_peak_sustained_elapsed |
N/A {458} % | PM时间序列:L2吞吐;该项没有可用汇总值,各采样窗口的数值按实例读取 |
SM_A.TriageAC.l1tex__data_pipe_lsu_wavefronts.avg |
248512 | PM时间序列:平均每L1TEX处理的wavefront数;默认值跨多个采样窗口聚合,各窗口数值按实例读取 |
SM_A.TriageAC.l1tex__lsu_writeback_active.avg |
94447.3 {458} cycle | PM时间序列:平均每L1TEX数据返回通路的活跃周期数;默认值跨多个采样窗口聚合,各窗口数值按实例读取 |
SM_A.TriageAC.sm__cycles_active.avg |
438812 {466} cycle | PM时间序列:平均每SM有warp驻留的周期数;默认值跨多个采样窗口聚合,各窗口数值按实例读取 |
SM_A.TriageAC.sm__inst_executed_realtime.avg.per_cycle_active |
198.568 {466} inst/cycle | PM时间序列:每SM每活跃周期执行的warp级指令数;默认值跨多个采样窗口聚合,各窗口数值按实例读取 |
SM_A.TriageSCG.l1tex__throughput.avg.pct_of_peak_sustained_elapsed |
N/A {458} % | PM时间序列:L1TEX吞吐;该项没有可用汇总值,各采样窗口的数值按实例读取 |
SM_A.TriageSCG.sm__inst_executed_pipe_alu_realtime.avg.pct_of_peak_sustained_elapsed |
8.06764 {466} % | PM时间序列:ALU指令吞吐相对持续峰值的比例;默认值跨多个采样窗口聚合,各窗口数值按实例读取 |
SM_B.TriageAC.l1tex__t_sector_hit_rate.pct |
N/A {458} % | PM时间序列:L1TEX sector命中率;该项没有可用汇总值,各采样窗口的数值按实例读取 |
SM_C.TriageSCG.smsp__inst_executed_pipe_fmaheavy.avg.pct_of_peak_sustained_elapsed |
4.52701 {466} % | PM时间序列:FMA Heavy指令吞吐相对持续峰值的比例;默认值跨多个采样窗口聚合,各窗口数值按实例读取 |
SM_C.TriageSCG.smsp__inst_executed_pipe_fmalite.avg.pct_of_peak_sustained_elapsed |
0.956771 {466} % | PM时间序列:FMA Light指令吞吐相对持续峰值的比例;默认值跨多个采样窗口聚合,各窗口数值按实例读取 |
TriageAC.tpc__warps_active_realtime.avg.per_cycle_active |
14407.8 {466} warp | PM时间序列:折算到每SM的平均活跃warp数;默认值跨多个采样窗口聚合,各窗口数值按实例读取 |
TriageAC.tpc__warps_active_realtime.sum.per_cycle_active |
1.18144e+06 {466} warp | PM时间序列:整卡活跃warp数;默认值跨多个采样窗口聚合,各窗口数值按实例读取 |
TriageSCG.sm__throughput.avg.pct_of_peak_sustained_elapsed |
N/A {466} % | PM时间序列:SM吞吐;该项没有可用汇总值,各采样窗口的数值按实例读取 |
pmsampling:sm__pipe_tensor_cycles_active_realtime.avg.pct_of_peak_sustained_elapsed |
0 {466} % | PM时间序列:Tensor管线活跃周期相对持续峰值的比例;默认值跨多个采样窗口聚合,各窗口数值按实例读取 |
采集器与重放记录
| 项 | 值 | 说明 |
|---|---|---|
profiler__perfworks_session_reuse |
0 | 是否复用PerfWorks采集会话 |
profiler__pmsampler_buffer_size_bytes |
25,165,824 {2} byte | PM采样每组的缓冲区字节数,实例分别记录各组 |
profiler__pmsampler_ctxsw_0 |
1 | 指定PM采样组的GPU上下文切换轨迹;实例对应时间戳,值1表示目标上下文活跃 |
profiler__pmsampler_ctxsw_1 |
1 | 指定PM采样组的GPU上下文切换轨迹;实例对应时间戳,值1表示目标上下文活跃 |
profiler__pmsampler_dropped_samples |
0 {2} sample | PM采样因缓冲区不足而丢弃的样本数 |
profiler__pmsampler_interval_time |
1 {2} us | PM采样间隔,原记录以纳秒计,本表换算为微秒 |
profiler__pmsampler_merged_samples |
853 {2} sample | PM采样传出遇到硬件反压时合并的样本数,实例分别记录各组 |
profiler__pmsampler_pass_groups |
2 | PM采样组数,实例保存各组对应的指标列表 |
profiler__replayer_bytes_mem_accessible.avg |
75,530,576 byte | 重放采集记录中工作负载可访问的内存字节数;在各次记录间取平均值 |
profiler__replayer_bytes_mem_accessible.max |
75,530,576 byte | 重放采集记录中工作负载可访问的内存字节数;在各次记录间取最大值 |
profiler__replayer_bytes_mem_accessible.min |
75,530,576 byte | 重放采集记录中工作负载可访问的内存字节数;在各次记录间取最小值 |
profiler__replayer_bytes_mem_accessible.sum |
3,021,223,040 byte | 重放采集记录中工作负载可访问的内存字节数;在各次记录间求和,得到累计字节数 |
profiler__replayer_bytes_mem_backed_up.avg |
75,530,576 byte | 重放采集记录中工具备份的内存字节数;在各次记录间取平均值 |
profiler__replayer_bytes_mem_backed_up.max |
75,530,576 byte | 重放采集记录中工具备份的内存字节数;在各次记录间取最大值 |
profiler__replayer_bytes_mem_backed_up.min |
75,530,576 byte | 重放采集记录中工具备份的内存字节数;在各次记录间取最小值 |
profiler__replayer_bytes_mem_backed_up.sum |
3,021,223,040 byte | 重放采集记录中工具备份的内存字节数;在各次记录间求和,得到累计字节数 |
profiler__replayer_passes |
40 次 | 本次指标采集总重放次数 |
profiler__replayer_passes_type_warmup |
0 次 | 专门用于预热的重放次数 |
profiler__timestamp_workload_end_0 |
已隐去 | 指定采样组的kernel结束时间戳标记;时间戳保存在实例关联标识中 |
profiler__timestamp_workload_end_1 |
已隐去 | 指定采样组的kernel结束时间戳标记;时间戳保存在实例关联标识中 |
profiler__timestamp_workload_start_0 |
已隐去 | 指定采样组的kernel开始时间戳标记;时间戳保存在实例关联标识中 |
profiler__timestamp_workload_start_1 |
已隐去 | 指定采样组的kernel开始时间戳标记;时间戳保存在实例关联标识中 |
互联与NUMA
| 项 | 值 | 说明 |
|---|---|---|
c2clink__enabled_mask |
0 | 已启用互联的位掩码 |
c2clink__present |
0 | 是否存在C2C设备互联 |
numa__cpu_affinity |
0-31 | 逐GPU保存的亲和逻辑CPU编号范围,默认汇总值表示记录条目 |
numa__dev_display_name_all |
NVIDIA GeForce RTX 3090 | NUMA表中的GPU名称列表,默认汇总值表示记录条目 |
numa__id_cpu |
0 | 逐GPU保存的CPU亲和NUMA节点编号,实际编号按实例读取 |
numa__id_memory |
0 | 逐GPU保存的内存亲和NUMA节点编号,实际编号按实例读取 |
nvlink__bandwidth |
0 | NVLink链路带宽记录,报告未标出单位 |
nvlink__count_logical |
0 | 逻辑NVLink连接数 |
nvlink__count_physical |
0 | 物理NVLink连接数 |
nvlink__destination_ports |
0 | NVLink目标端口列表 |
nvlink__dev0Id |
0 | NVLink一端的设备编号 |
nvlink__dev0type |
0 | NVLink一端的设备类型编码 |
nvlink__dev1Id |
0 | NVLink另一端的设备编号 |
nvlink__dev1type |
0 | NVLink另一端的设备类型编码 |
nvlink__dev_display_name_all |
NVIDIA GeForce RTX 3090 | NVLink拓扑中的GPU名称列表 |
nvlink__enabled_mask |
0 | 已启用互联的位掩码 |
nvlink__is_direct_link |
0 | 是否为设备间直接连接 |
nvlink__is_nvswitch_connected |
0 | 是否经NVSwitch连接 |
nvlink__max_count |
4 | 该物理NVLink设备报告的最大链路数,实例按设备ID映射;4是设备的链路数量上限,本报告实际发现的逻辑、物理连接数均为0,启用掩码也为0,不能据此数桥接器或已连接GPU |
nvlink__peer_access |
0 | 是否具有对端访问能力 |
nvlink__peer_atomic |
0 | 是否具有对端原子操作能力 |
nvlink__source_ports |
0 | NVLink源端口列表 |
nvlink__system_access |
0 | 是否具有系统内存访问能力 |
nvlink__system_atomic |
0 | 是否具有系统内存原子操作能力 |
nvlrx__bytes.sum.per_second |
0 byte/s | NVLink接收的字节数,计数粒度为16字节;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
nvltx__bytes.sum.per_second |
0 byte/s | NVLink发送的字节数,计数粒度为16字节;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
pcie__read_bytes.avg.per_second |
0 byte/s | PCIe读取数据的字节数,计数粒度为512字节;在同类硬件单元间取平均并除以运行时长,得到每秒处理量 |
pcie__read_bytes.max.per_second |
0 byte/s | PCIe读取数据的字节数,计数粒度为512字节;在同类硬件单元间取最大值并除以运行时长,得到每秒处理量 |
pcie__read_bytes.min.per_second |
0 byte/s | PCIe读取数据的字节数,计数粒度为512字节;在同类硬件单元间取最小值并除以运行时长,得到每秒处理量 |
pcie__read_bytes.sum.per_second |
0 byte/s | PCIe读取数据的字节数,计数粒度为512字节;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
pcie__write_bytes.avg.per_second |
5.62758e+07 byte/s | PCIe写入数据的字节数,计数粒度为512字节;在同类硬件单元间取平均并除以运行时长,得到每秒处理量 |
pcie__write_bytes.max.per_second |
5.62758e+07 byte/s | PCIe写入数据的字节数,计数粒度为512字节;在同类硬件单元间取最大值并除以运行时长,得到每秒处理量 |
pcie__write_bytes.min.per_second |
5.62758e+07 byte/s | PCIe写入数据的字节数,计数粒度为512字节;在同类硬件单元间取最小值并除以运行时长,得到每秒处理量 |
pcie__write_bytes.sum.per_second |
5.62758e+07 byte/s | PCIe写入数据的字节数,计数粒度为512字节;在同类硬件单元间求和并除以运行时长,得到每秒处理量 |
界面分组与吞吐分解记录
| 项 | 值 | 说明 |
|---|---|---|
breakdown:gpu__compute_memory_throughput.avg.pct_of_peak_sustained_elapsed |
dram__cycles_active.avg.pct_of_peak_sustained_elapsed fbpa__dram_sectors.avg.pct_of_peak_sustained_elapsed gpu__compute_memory_access_throughput_internal_activity.avg.pct_of_peak_sustained_elapsed gpu__compute_memory_request_throughput_internal_activity.avg.pct_of_peak_sustained_elapsed l1tex__data_bank_reads.avg.pct_of_peak_sustained_elapsed l1tex__data_bank_writes.avg.pct_of_peak_sustained_elapsed l1tex__data_pipe_lsu_wavefronts.avg.pct_of_peak_sustained_elapsed l1tex__data_pipe_tex_wavefronts.avg.pct_of_peak_sustained_elapsed l1tex__f_wavefronts.avg.pct_of_peak_sustained_elapsed l1tex__lsu_writeback_active.avg.pct_of_peak_sustained_elapsed l1tex__lsuin_requests.avg.pct_of_peak_sustained_elapsed l1tex__m_l1tex2xbar_req_cycles_active.avg.pct_of_peak_sustained_elapsed l1tex__m_xbar2l1tex_read_sectors.avg.pct_of_peak_sustained_elapsed l1tex__tex_writeback_active.avg.pct_of_peak_sustained_elapsed l1tex__texin_sm2tex_req_cycles_active.avg.pct_of_peak_sustained_elapsed lts__d_atomic_input_cycles_active.avg.pct_of_peak_sustained_elapsed lts__d_sectors.avg.pct_of_peak_sustained_elapsed lts__d_sectors_fill_device.avg.pct_of_peak_sustained_elapsed lts__d_sectors_fill_sysmem.avg.pct_of_peak_sustained_elapsed lts__lts2xbar_cycles_active.avg.pct_of_peak_sustained_elapsed lts__t_sectors.avg.pct_of_peak_sustained_elapsed lts__t_tag_requests.avg.pct_of_peak_sustained_elapsed lts__xbar2lts_cycles_active.avg.pct_of_peak_sustained_elapsed % |
吞吐汇总的组成指标列表;值中的各名称指向对应指标,列表本身用于组织显示 |
breakdown:gpu__compute_memory_throughput.max.pct_of_peak_sustained_elapsed |
dram__cycles_active.max.pct_of_peak_sustained_elapsed fbpa__dram_sectors.max.pct_of_peak_sustained_elapsed gpu__compute_memory_access_throughput_internal_activity.max.pct_of_peak_sustained_elapsed gpu__compute_memory_request_throughput_internal_activity.max.pct_of_peak_sustained_elapsed l1tex__data_bank_reads.max.pct_of_peak_sustained_elapsed l1tex__data_bank_writes.max.pct_of_peak_sustained_elapsed l1tex__data_pipe_lsu_wavefronts.max.pct_of_peak_sustained_elapsed l1tex__data_pipe_tex_wavefronts.max.pct_of_peak_sustained_elapsed l1tex__f_wavefronts.max.pct_of_peak_sustained_elapsed l1tex__lsu_writeback_active.max.pct_of_peak_sustained_elapsed l1tex__lsuin_requests.max.pct_of_peak_sustained_elapsed l1tex__m_l1tex2xbar_req_cycles_active.max.pct_of_peak_sustained_elapsed l1tex__m_xbar2l1tex_read_sectors.max.pct_of_peak_sustained_elapsed l1tex__tex_writeback_active.max.pct_of_peak_sustained_elapsed l1tex__texin_sm2tex_req_cycles_active.max.pct_of_peak_sustained_elapsed lts__d_atomic_input_cycles_active.max.pct_of_peak_sustained_elapsed lts__d_sectors.max.pct_of_peak_sustained_elapsed lts__d_sectors_fill_device.max.pct_of_peak_sustained_elapsed lts__d_sectors_fill_sysmem.max.pct_of_peak_sustained_elapsed lts__lts2xbar_cycles_active.max.pct_of_peak_sustained_elapsed lts__t_sectors.max.pct_of_peak_sustained_elapsed lts__t_tag_requests.max.pct_of_peak_sustained_elapsed lts__xbar2lts_cycles_active.max.pct_of_peak_sustained_elapsed % |
吞吐汇总的组成指标列表;值中的各名称指向对应指标,列表本身用于组织显示 |
breakdown:gpu__compute_memory_throughput.min.pct_of_peak_sustained_elapsed |
dram__cycles_active.min.pct_of_peak_sustained_elapsed fbpa__dram_sectors.min.pct_of_peak_sustained_elapsed gpu__compute_memory_access_throughput_internal_activity.min.pct_of_peak_sustained_elapsed gpu__compute_memory_request_throughput_internal_activity.min.pct_of_peak_sustained_elapsed l1tex__data_bank_reads.min.pct_of_peak_sustained_elapsed l1tex__data_bank_writes.min.pct_of_peak_sustained_elapsed l1tex__data_pipe_lsu_wavefronts.min.pct_of_peak_sustained_elapsed l1tex__data_pipe_tex_wavefronts.min.pct_of_peak_sustained_elapsed l1tex__f_wavefronts.min.pct_of_peak_sustained_elapsed l1tex__lsu_writeback_active.min.pct_of_peak_sustained_elapsed l1tex__lsuin_requests.min.pct_of_peak_sustained_elapsed l1tex__m_l1tex2xbar_req_cycles_active.min.pct_of_peak_sustained_elapsed l1tex__m_xbar2l1tex_read_sectors.min.pct_of_peak_sustained_elapsed l1tex__tex_writeback_active.min.pct_of_peak_sustained_elapsed l1tex__texin_sm2tex_req_cycles_active.min.pct_of_peak_sustained_elapsed lts__d_atomic_input_cycles_active.min.pct_of_peak_sustained_elapsed lts__d_sectors.min.pct_of_peak_sustained_elapsed lts__d_sectors_fill_device.min.pct_of_peak_sustained_elapsed lts__d_sectors_fill_sysmem.min.pct_of_peak_sustained_elapsed lts__lts2xbar_cycles_active.min.pct_of_peak_sustained_elapsed lts__t_sectors.min.pct_of_peak_sustained_elapsed lts__t_tag_requests.min.pct_of_peak_sustained_elapsed lts__xbar2lts_cycles_active.min.pct_of_peak_sustained_elapsed % |
吞吐汇总的组成指标列表;值中的各名称指向对应指标,列表本身用于组织显示 |
breakdown:gpu__compute_memory_throughput.sum.pct_of_peak_sustained_elapsed |
dram__cycles_active.sum.pct_of_peak_sustained_elapsed fbpa__dram_sectors.sum.pct_of_peak_sustained_elapsed gpu__compute_memory_access_throughput_internal_activity.sum.pct_of_peak_sustained_elapsed gpu__compute_memory_request_throughput_internal_activity.sum.pct_of_peak_sustained_elapsed l1tex__data_bank_reads.sum.pct_of_peak_sustained_elapsed l1tex__data_bank_writes.sum.pct_of_peak_sustained_elapsed l1tex__data_pipe_lsu_wavefronts.sum.pct_of_peak_sustained_elapsed l1tex__data_pipe_tex_wavefronts.sum.pct_of_peak_sustained_elapsed l1tex__f_wavefronts.sum.pct_of_peak_sustained_elapsed l1tex__lsu_writeback_active.sum.pct_of_peak_sustained_elapsed l1tex__lsuin_requests.sum.pct_of_peak_sustained_elapsed l1tex__m_l1tex2xbar_req_cycles_active.sum.pct_of_peak_sustained_elapsed l1tex__m_xbar2l1tex_read_sectors.sum.pct_of_peak_sustained_elapsed l1tex__tex_writeback_active.sum.pct_of_peak_sustained_elapsed l1tex__texin_sm2tex_req_cycles_active.sum.pct_of_peak_sustained_elapsed lts__d_atomic_input_cycles_active.sum.pct_of_peak_sustained_elapsed lts__d_sectors.sum.pct_of_peak_sustained_elapsed lts__d_sectors_fill_device.sum.pct_of_peak_sustained_elapsed lts__d_sectors_fill_sysmem.sum.pct_of_peak_sustained_elapsed lts__lts2xbar_cycles_active.sum.pct_of_peak_sustained_elapsed lts__t_sectors.sum.pct_of_peak_sustained_elapsed lts__t_tag_requests.sum.pct_of_peak_sustained_elapsed lts__xbar2lts_cycles_active.sum.pct_of_peak_sustained_elapsed % |
吞吐汇总的组成指标列表;值中的各名称指向对应指标,列表本身用于组织显示 |
breakdown:sm__throughput.avg.pct_of_peak_sustained_elapsed |
idc__request_cycles_active.avg.pct_of_peak_sustained_elapsed sm__inst_executed.avg.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_adu.avg.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_cbu_pred_on_any.avg.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_ipa.avg.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_lsu.avg.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_tex.avg.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_uniform.avg.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_xu.avg.pct_of_peak_sustained_elapsed sm__instruction_throughput_internal_activity.avg.pct_of_peak_sustained_elapsed sm__issue_active.avg.pct_of_peak_sustained_elapsed sm__memory_throughput_internal_activity.avg.pct_of_peak_sustained_elapsed sm__mio2rf_writeback_active.avg.pct_of_peak_sustained_elapsed sm__mio_inst_issued.avg.pct_of_peak_sustained_elapsed sm__mio_pq_read_cycles_active.avg.pct_of_peak_sustained_elapsed sm__mio_pq_write_cycles_active.avg.pct_of_peak_sustained_elapsed sm__pipe_alu_cycles_active.avg.pct_of_peak_sustained_elapsed sm__pipe_fma_cycles_active.avg.pct_of_peak_sustained_elapsed sm__pipe_fmaheavy_cycles_active.avg.pct_of_peak_sustained_elapsed sm__pipe_fp64_cycles_active.avg.pct_of_peak_sustained_elapsed sm__pipe_tensor_cycles_active.avg.pct_of_peak_sustained_elapsed % |
吞吐汇总的组成指标列表;值中的各名称指向对应指标,列表本身用于组织显示 |
breakdown:sm__throughput.max.pct_of_peak_sustained_elapsed |
idc__request_cycles_active.max.pct_of_peak_sustained_elapsed sm__inst_executed.max.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_adu.max.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_cbu_pred_on_any.max.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_ipa.max.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_lsu.max.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_tex.max.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_uniform.max.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_xu.max.pct_of_peak_sustained_elapsed sm__instruction_throughput_internal_activity.max.pct_of_peak_sustained_elapsed sm__issue_active.max.pct_of_peak_sustained_elapsed sm__memory_throughput_internal_activity.max.pct_of_peak_sustained_elapsed sm__mio2rf_writeback_active.max.pct_of_peak_sustained_elapsed sm__mio_inst_issued.max.pct_of_peak_sustained_elapsed sm__mio_pq_read_cycles_active.max.pct_of_peak_sustained_elapsed sm__mio_pq_write_cycles_active.max.pct_of_peak_sustained_elapsed sm__pipe_alu_cycles_active.max.pct_of_peak_sustained_elapsed sm__pipe_fma_cycles_active.max.pct_of_peak_sustained_elapsed sm__pipe_fmaheavy_cycles_active.max.pct_of_peak_sustained_elapsed sm__pipe_fp64_cycles_active.max.pct_of_peak_sustained_elapsed sm__pipe_tensor_cycles_active.max.pct_of_peak_sustained_elapsed % |
吞吐汇总的组成指标列表;值中的各名称指向对应指标,列表本身用于组织显示 |
breakdown:sm__throughput.min.pct_of_peak_sustained_elapsed |
idc__request_cycles_active.min.pct_of_peak_sustained_elapsed sm__inst_executed.min.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_adu.min.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_cbu_pred_on_any.min.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_ipa.min.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_lsu.min.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_tex.min.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_uniform.min.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_xu.min.pct_of_peak_sustained_elapsed sm__instruction_throughput_internal_activity.min.pct_of_peak_sustained_elapsed sm__issue_active.min.pct_of_peak_sustained_elapsed sm__memory_throughput_internal_activity.min.pct_of_peak_sustained_elapsed sm__mio2rf_writeback_active.min.pct_of_peak_sustained_elapsed sm__mio_inst_issued.min.pct_of_peak_sustained_elapsed sm__mio_pq_read_cycles_active.min.pct_of_peak_sustained_elapsed sm__mio_pq_write_cycles_active.min.pct_of_peak_sustained_elapsed sm__pipe_alu_cycles_active.min.pct_of_peak_sustained_elapsed sm__pipe_fma_cycles_active.min.pct_of_peak_sustained_elapsed sm__pipe_fmaheavy_cycles_active.min.pct_of_peak_sustained_elapsed sm__pipe_fp64_cycles_active.min.pct_of_peak_sustained_elapsed sm__pipe_tensor_cycles_active.min.pct_of_peak_sustained_elapsed % |
吞吐汇总的组成指标列表;值中的各名称指向对应指标,列表本身用于组织显示 |
breakdown:sm__throughput.sum.pct_of_peak_sustained_elapsed |
idc__request_cycles_active.sum.pct_of_peak_sustained_elapsed sm__inst_executed.sum.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_adu.sum.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_cbu_pred_on_any.sum.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_ipa.sum.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_lsu.sum.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_tex.sum.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_uniform.sum.pct_of_peak_sustained_elapsed sm__inst_executed_pipe_xu.sum.pct_of_peak_sustained_elapsed sm__instruction_throughput_internal_activity.sum.pct_of_peak_sustained_elapsed sm__issue_active.sum.pct_of_peak_sustained_elapsed sm__memory_throughput_internal_activity.sum.pct_of_peak_sustained_elapsed sm__mio2rf_writeback_active.sum.pct_of_peak_sustained_elapsed sm__mio_inst_issued.sum.pct_of_peak_sustained_elapsed sm__mio_pq_read_cycles_active.sum.pct_of_peak_sustained_elapsed sm__mio_pq_write_cycles_active.sum.pct_of_peak_sustained_elapsed sm__pipe_alu_cycles_active.sum.pct_of_peak_sustained_elapsed sm__pipe_fma_cycles_active.sum.pct_of_peak_sustained_elapsed sm__pipe_fmaheavy_cycles_active.sum.pct_of_peak_sustained_elapsed sm__pipe_fp64_cycles_active.sum.pct_of_peak_sustained_elapsed sm__pipe_tensor_cycles_active.sum.pct_of_peak_sustained_elapsed % |
吞吐汇总的组成指标列表;值中的各名称指向对应指标,列表本身用于组织显示 |
group:memory__chart |
dram__bytes_read.sum dram__bytes_read.sum.pct_of_peak_sustained_elapsed dram__bytes_read.sum.per_second dram__bytes_write.sum dram__bytes_write.sum.pct_of_peak_sustained_elapsed dram__bytes_write.sum.per_second l1tex__data_pipe_lsu_wavefronts_mem_shared.sum.pct_of_peak_sustained_elapsed l1tex__m_l1tex2xbar_write_bytes.sum l1tex__m_l1tex2xbar_write_bytes.sum.pct_of_peak_sustained_elapsed l1tex__m_l1tex2xbar_write_bytes.sum.per_second l1tex__m_xbar2l1tex_read_bytes.sum l1tex__m_xbar2l1tex_read_bytes.sum.pct_of_peak_sustained_elapsed l1tex__m_xbar2l1tex_read_bytes.sum.per_second l1tex__t_requests_pipe_lsu_mem_global_op_atom.sum l1tex__t_requests_pipe_lsu_mem_global_op_atom.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_lsu_mem_global_op_ld.sum l1tex__t_requests_pipe_lsu_mem_global_op_ld.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_lsu_mem_global_op_red.sum l1tex__t_requests_pipe_lsu_mem_global_op_red.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_lsu_mem_global_op_st.sum l1tex__t_requests_pipe_lsu_mem_global_op_st.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_lsu_mem_local_op_ld.sum l1tex__t_requests_pipe_lsu_mem_local_op_ld.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_lsu_mem_local_op_st.sum l1tex__t_requests_pipe_lsu_mem_local_op_st.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_tex_mem_surface_op_atom.sum l1tex__t_requests_pipe_tex_mem_surface_op_atom.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_tex_mem_surface_op_ld.sum l1tex__t_requests_pipe_tex_mem_surface_op_ld.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_tex_mem_surface_op_red.sum l1tex__t_requests_pipe_tex_mem_surface_op_red.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_tex_mem_surface_op_st.sum l1tex__t_requests_pipe_tex_mem_surface_op_st.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_tex_mem_texture.sum l1tex__t_requests_pipe_tex_mem_texture.sum.pct_of_peak_sustained_elapsed l1tex__t_sector_hit_rate.pct lts__average_gcomp_input_sector_success_rate.pct lts__average_gcomp_output_sector_compression_achieved_rate.ratio lts__gcomp_input_sectors.sum lts__gcomp_input_sectors.sum.pct_of_peak_sustained_elapsed lts__gcomp_input_sectors.sum.per_second lts__gcomp_output_sectors.sum lts__gcomp_output_sectors.sum.pct_of_peak_sustained_elapsed lts__gcomp_output_sectors.sum.per_second lts__t_sector_hit_rate.pct lts__t_sectors_srcunit_tex_aperture_peer_op_atom.sum lts__t_sectors_srcunit_tex_aperture_peer_op_atom.sum.pct_of_peak_sustained_elapsed lts__t_sectors_srcunit_tex_aperture_peer_op_atom.sum.per_second lts__t_sectors_srcunit_tex_aperture_peer_op_read_lookup_miss.sum lts__t_sectors_srcunit_tex_aperture_peer_op_read_lookup_miss.sum.pct_of_peak_sustained_elapsed lts__t_sectors_srcunit_tex_aperture_peer_op_read_lookup_miss.sum.per_second lts__t_sectors_srcunit_tex_aperture_peer_op_red.sum lts__t_sectors_srcunit_tex_aperture_peer_op_red.sum.pct_of_peak_sustained_elapsed lts__t_sectors_srcunit_tex_aperture_peer_op_red.sum.per_second lts__t_sectors_srcunit_tex_aperture_peer_op_write.sum lts__t_sectors_srcunit_tex_aperture_peer_op_write.sum.pct_of_peak_sustained_elapsed lts__t_sectors_srcunit_tex_aperture_peer_op_write.sum.per_second lts__t_sectors_srcunit_tex_aperture_sysmem_op_atom.sum lts__t_sectors_srcunit_tex_aperture_sysmem_op_atom.sum.pct_of_peak_sustained_elapsed lts__t_sectors_srcunit_tex_aperture_sysmem_op_atom.sum.per_second lts__t_sectors_srcunit_tex_aperture_sysmem_op_read_lookup_miss.sum lts__t_sectors_srcunit_tex_aperture_sysmem_op_read_lookup_miss.sum.pct_of_peak_sustained_elapsed lts__t_sectors_srcunit_tex_aperture_sysmem_op_read_lookup_miss.sum.per_second lts__t_sectors_srcunit_tex_aperture_sysmem_op_red.sum lts__t_sectors_srcunit_tex_aperture_sysmem_op_red.sum.pct_of_peak_sustained_elapsed lts__t_sectors_srcunit_tex_aperture_sysmem_op_red.sum.per_second lts__t_sectors_srcunit_tex_aperture_sysmem_op_write.sum lts__t_sectors_srcunit_tex_aperture_sysmem_op_write.sum.pct_of_peak_sustained_elapsed lts__t_sectors_srcunit_tex_aperture_sysmem_op_write.sum.per_second lts__t_sectors_srcunit_tex_op_read.avg.pct_of_peak_sustained_elapsed lts__t_sectors_srcunit_tex_op_read.sum lts__t_sectors_srcunit_tex_op_write.avg.pct_of_peak_sustained_elapsed lts__t_sectors_srcunit_tex_op_write.sum sm__sass_l1tex_m_xbar2l1tex_read_bytes_mem_global_op_ldgsts_cache_bypass.sum sm__sass_l1tex_m_xbar2l1tex_read_bytes_mem_global_op_ldgsts_cache_bypass.sum.pct_of_peak_sustained_elapsed sm__sass_l1tex_m_xbar2l1tex_read_bytes_mem_global_op_ldgsts_cache_bypass.sum.per_second sm__sass_l1tex_t_requests_pipe_lsu_mem_global_op_ldgsts.sum sm__sass_l1tex_t_requests_pipe_lsu_mem_global_op_ldgsts.sum.pct_of_peak_sustained_elapsed sm__sass_l1tex_t_sectors_pipe_lsu_mem_global_op_ldgsts_cache_access.sum sm__sass_l1tex_t_sectors_pipe_lsu_mem_global_op_ldgsts_cache_access.sum.pct_of_peak_sustained_elapsed sm__sass_l1tex_t_sectors_pipe_lsu_mem_global_op_ldgsts_cache_access.sum.per_second smsp__cycles_elapsed.sum smsp__inst_executed_op_generic_atom_dot_alu.sum smsp__inst_executed_op_generic_atom_dot_alu.sum.pct_of_peak_sustained_elapsed smsp__inst_executed_op_generic_atom_dot_cas.sum smsp__inst_executed_op_generic_atom_dot_cas.sum.pct_of_peak_sustained_elapsed smsp__inst_executed_op_global_red.sum smsp__inst_executed_op_global_red.sum.pct_of_peak_sustained_elapsed smsp__inst_executed_op_ldgsts.sum smsp__inst_executed_op_ldgsts.sum.pct_of_peak_sustained_elapsed smsp__inst_executed_op_ldsm.sum smsp__inst_executed_op_ldsm.sum.pct_of_peak_sustained_elapsed smsp__inst_executed_op_shared_atom.sum smsp__inst_executed_op_shared_atom.sum.pct_of_peak_sustained_elapsed smsp__inst_executed_op_surface_atom_dot_alu.sum smsp__inst_executed_op_surface_atom_dot_alu.sum.pct_of_peak_sustained_elapsed smsp__inst_executed_op_surface_atom_dot_cas.sum smsp__inst_executed_op_surface_atom_dot_cas.sum.pct_of_peak_sustained_elapsed smsp__inst_executed_op_surface_ld.sum smsp__inst_executed_op_surface_ld.sum.pct_of_peak_sustained_elapsed smsp__inst_executed_op_surface_red.sum smsp__inst_executed_op_surface_red.sum.pct_of_peak_sustained_elapsed smsp__inst_executed_op_surface_st.sum smsp__inst_executed_op_surface_st.sum.pct_of_peak_sustained_elapsed smsp__inst_executed_op_texture.sum smsp__inst_executed_op_texture.sum.pct_of_peak_sustained_elapsed smsp__sass_inst_executed_op_global_ld.sum smsp__sass_inst_executed_op_global_st.sum smsp__sass_inst_executed_op_local_ld.sum smsp__sass_inst_executed_op_local_st.sum smsp__sass_inst_executed_op_shared_ld.sum smsp__sass_inst_executed_op_shared_st.sum smsp__sass_l1tex_data_pipe_lsu_wavefronts_mem_shared_op_ldgsts.sum smsp__sass_l1tex_data_pipe_lsu_wavefronts_mem_shared_op_ldgsts.sum.pct_of_peak_sustained_elapsed |
界面所用的指标分组列表;值中的各名称指向对应指标,列表本身用于组织显示 |
group:memory__dram_table |
dram__bytes_read.sum dram__bytes_read.sum.pct_of_peak_sustained_elapsed dram__bytes_read.sum.per_second dram__bytes_write.sum dram__bytes_write.sum.pct_of_peak_sustained_elapsed dram__bytes_write.sum.per_second dram__sectors_read.sum dram__sectors_write.sum |
界面所用的指标分组列表;值中的各名称指向对应指标,列表本身用于组织显示 |
group:memory__first_level_cache_table |
l1tex__lsu_writeback_active_mem_lg.sum l1tex__lsu_writeback_active_mem_lg.sum.pct_of_peak_sustained_elapsed l1tex__m_l1tex2xbar_write_sectors_mem_global_op_atom.sum l1tex__m_l1tex2xbar_write_sectors_mem_global_op_atom.sum.pct_of_peak_sustained_elapsed l1tex__m_l1tex2xbar_write_sectors_mem_global_op_red.sum l1tex__m_l1tex2xbar_write_sectors_mem_global_op_red.sum.pct_of_peak_sustained_elapsed l1tex__m_l1tex2xbar_write_sectors_mem_lg_op_st.sum l1tex__m_l1tex2xbar_write_sectors_mem_lg_op_st.sum.pct_of_peak_sustained_elapsed l1tex__m_l1tex2xbar_write_sectors_mem_surface_op_atom.sum l1tex__m_l1tex2xbar_write_sectors_mem_surface_op_atom.sum.pct_of_peak_sustained_elapsed l1tex__m_l1tex2xbar_write_sectors_mem_surface_op_red.sum l1tex__m_l1tex2xbar_write_sectors_mem_surface_op_red.sum.pct_of_peak_sustained_elapsed l1tex__m_l1tex2xbar_write_sectors_mem_surface_op_st.sum l1tex__m_l1tex2xbar_write_sectors_mem_surface_op_st.sum.pct_of_peak_sustained_elapsed l1tex__m_xbar2l1tex_read_sectors_mem_global_op_atom.sum l1tex__m_xbar2l1tex_read_sectors_mem_global_op_atom.sum.pct_of_peak_sustained_elapsed l1tex__m_xbar2l1tex_read_sectors_mem_lg_op_ld.sum l1tex__m_xbar2l1tex_read_sectors_mem_lg_op_ld.sum.pct_of_peak_sustained_elapsed l1tex__m_xbar2l1tex_read_sectors_mem_surface_op_atom.sum l1tex__m_xbar2l1tex_read_sectors_mem_surface_op_atom.sum.pct_of_peak_sustained_elapsed l1tex__m_xbar2l1tex_read_sectors_mem_surface_op_ld.sum l1tex__m_xbar2l1tex_read_sectors_mem_surface_op_ld.sum.pct_of_peak_sustained_elapsed l1tex__m_xbar2l1tex_read_sectors_mem_texture.sum l1tex__m_xbar2l1tex_read_sectors_mem_texture.sum.pct_of_peak_sustained_elapsed l1tex__t_output_wavefronts_pipe_lsu_mem_global_op_atom.sum l1tex__t_output_wavefronts_pipe_lsu_mem_global_op_atom.sum.pct_of_peak_sustained_elapsed l1tex__t_output_wavefronts_pipe_lsu_mem_global_op_ld.sum l1tex__t_output_wavefronts_pipe_lsu_mem_global_op_ld.sum.pct_of_peak_sustained_elapsed l1tex__t_output_wavefronts_pipe_lsu_mem_global_op_red.sum l1tex__t_output_wavefronts_pipe_lsu_mem_global_op_red.sum.pct_of_peak_sustained_elapsed l1tex__t_output_wavefronts_pipe_lsu_mem_global_op_st.sum l1tex__t_output_wavefronts_pipe_lsu_mem_global_op_st.sum.pct_of_peak_sustained_elapsed l1tex__t_output_wavefronts_pipe_lsu_mem_local_op_ld.sum l1tex__t_output_wavefronts_pipe_lsu_mem_local_op_ld.sum.pct_of_peak_sustained_elapsed l1tex__t_output_wavefronts_pipe_lsu_mem_local_op_st.sum l1tex__t_output_wavefronts_pipe_lsu_mem_local_op_st.sum.pct_of_peak_sustained_elapsed l1tex__t_output_wavefronts_pipe_tex_mem_surface_op_atom.sum l1tex__t_output_wavefronts_pipe_tex_mem_surface_op_atom.sum.pct_of_peak_sustained_elapsed l1tex__t_output_wavefronts_pipe_tex_mem_surface_op_ld.sum l1tex__t_output_wavefronts_pipe_tex_mem_surface_op_ld.sum.pct_of_peak_sustained_elapsed l1tex__t_output_wavefronts_pipe_tex_mem_surface_op_red.sum l1tex__t_output_wavefronts_pipe_tex_mem_surface_op_red.sum.pct_of_peak_sustained_elapsed l1tex__t_output_wavefronts_pipe_tex_mem_surface_op_st.sum l1tex__t_output_wavefronts_pipe_tex_mem_surface_op_st.sum.pct_of_peak_sustained_elapsed l1tex__t_output_wavefronts_pipe_tex_mem_texture.sum l1tex__t_output_wavefronts_pipe_tex_mem_texture.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_lsu_mem_global_op_atom.sum l1tex__t_requests_pipe_lsu_mem_global_op_atom.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_lsu_mem_global_op_ld.sum l1tex__t_requests_pipe_lsu_mem_global_op_ld.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_lsu_mem_global_op_red.sum l1tex__t_requests_pipe_lsu_mem_global_op_red.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_lsu_mem_global_op_st.sum l1tex__t_requests_pipe_lsu_mem_global_op_st.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_lsu_mem_local_op_ld.sum l1tex__t_requests_pipe_lsu_mem_local_op_ld.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_lsu_mem_local_op_st.sum l1tex__t_requests_pipe_lsu_mem_local_op_st.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_tex_mem_surface_op_atom.sum l1tex__t_requests_pipe_tex_mem_surface_op_atom.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_tex_mem_surface_op_ld.sum l1tex__t_requests_pipe_tex_mem_surface_op_ld.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_tex_mem_surface_op_red.sum l1tex__t_requests_pipe_tex_mem_surface_op_red.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_tex_mem_surface_op_st.sum l1tex__t_requests_pipe_tex_mem_surface_op_st.sum.pct_of_peak_sustained_elapsed l1tex__t_requests_pipe_tex_mem_texture.sum l1tex__t_requests_pipe_tex_mem_texture.sum.pct_of_peak_sustained_elapsed l1tex__t_sector_pipe_lsu_mem_global_op_atom_hit_rate.pct l1tex__t_sector_pipe_lsu_mem_global_op_ld_hit_rate.pct l1tex__t_sector_pipe_lsu_mem_global_op_red_hit_rate.pct l1tex__t_sector_pipe_lsu_mem_global_op_st_hit_rate.pct l1tex__t_sector_pipe_lsu_mem_local_op_ld_hit_rate.pct l1tex__t_sector_pipe_lsu_mem_local_op_st_hit_rate.pct l1tex__t_sector_pipe_tex_mem_surface_op_atom_hit_rate.pct l1tex__t_sector_pipe_tex_mem_surface_op_ld_hit_rate.pct l1tex__t_sector_pipe_tex_mem_surface_op_red_hit_rate.pct l1tex__t_sector_pipe_tex_mem_surface_op_st_hit_rate.pct l1tex__t_sector_pipe_tex_mem_texture_op_tex_hit_rate.pct l1tex__t_sectors_pipe_lsu_mem_global_op_atom.sum l1tex__t_sectors_pipe_lsu_mem_global_op_atom_lookup_hit.sum l1tex__t_sectors_pipe_lsu_mem_global_op_atom_lookup_miss.sum l1tex__t_sectors_pipe_lsu_mem_global_op_ld.sum l1tex__t_sectors_pipe_lsu_mem_global_op_ld_lookup_hit.sum l1tex__t_sectors_pipe_lsu_mem_global_op_ld_lookup_miss.sum l1tex__t_sectors_pipe_lsu_mem_global_op_red.sum l1tex__t_sectors_pipe_lsu_mem_global_op_red_lookup_hit.sum l1tex__t_sectors_pipe_lsu_mem_global_op_red_lookup_miss.sum l1tex__t_sectors_pipe_lsu_mem_global_op_st.sum l1tex__t_sectors_pipe_lsu_mem_global_op_st_lookup_hit.sum l1tex__t_sectors_pipe_lsu_mem_global_op_st_lookup_miss.sum l1tex__t_sectors_pipe_lsu_mem_local_op_ld.sum l1tex__t_sectors_pipe_lsu_mem_local_op_ld_lookup_hit.sum l1tex__t_sectors_pipe_lsu_mem_local_op_ld_lookup_miss.sum l1tex__t_sectors_pipe_lsu_mem_local_op_st.sum l1tex__t_sectors_pipe_lsu_mem_local_op_st_lookup_hit.sum l1tex__t_sectors_pipe_lsu_mem_local_op_st_lookup_miss.sum l1tex__t_sectors_pipe_tex_mem_surface_op_atom.sum l1tex__t_sectors_pipe_tex_mem_surface_op_atom_lookup_hit.sum l1tex__t_sectors_pipe_tex_mem_surface_op_atom_lookup_miss.sum l1tex__t_sectors_pipe_tex_mem_surface_op_ld.sum l1tex__t_sectors_pipe_tex_mem_surface_op_ld_lookup_hit.sum l1tex__t_sectors_pipe_tex_mem_surface_op_ld_lookup_miss.sum l1tex__t_sectors_pipe_tex_mem_surface_op_red.sum l1tex__t_sectors_pipe_tex_mem_surface_op_red_lookup_hit.sum l1tex__t_sectors_pipe_tex_mem_surface_op_red_lookup_miss.sum l1tex__t_sectors_pipe_tex_mem_surface_op_st.sum l1tex__t_sectors_pipe_tex_mem_surface_op_st_lookup_hit.sum l1tex__t_sectors_pipe_tex_mem_surface_op_st_lookup_miss.sum l1tex__t_sectors_pipe_tex_mem_texture.sum l1tex__t_sectors_pipe_tex_mem_texture_lookup_hit.sum l1tex__t_sectors_pipe_tex_mem_texture_lookup_miss.sum l1tex__tex_writeback_active.sum l1tex__tex_writeback_active.sum.pct_of_peak_sustained_elapsed sm__sass_inst_executed_op_ldgsts_cache_access.sum sm__sass_inst_executed_op_ldgsts_cache_bypass.sum sm__sass_l1tex_t_requests_pipe_lsu_mem_global_op_ldgsts_cache_access.sum sm__sass_l1tex_t_requests_pipe_lsu_mem_global_op_ldgsts_cache_bypass.sum sm__sass_l1tex_t_sectors_pipe_lsu_mem_global_op_ldgsts_cache_access.sum sm__sass_l1tex_t_sectors_pipe_lsu_mem_global_op_ldgsts_cache_access.sum.pct_of_peak_sustained_elapsed sm__sass_l1tex_t_sectors_pipe_lsu_mem_global_op_ldgsts_cache_bypass.sum sm__sass_l1tex_t_sectors_pipe_lsu_mem_global_op_ldgsts_cache_bypass.sum.pct_of_peak_sustained_elapsed smsp__inst_executed_op_generic_atom_dot_alu.sum smsp__inst_executed_op_generic_atom_dot_cas.sum smsp__inst_executed_op_global_red.sum smsp__inst_executed_op_surface_atom_dot_alu.sum smsp__inst_executed_op_surface_atom_dot_cas.sum smsp__inst_executed_op_surface_ld.sum smsp__inst_executed_op_surface_red.sum smsp__inst_executed_op_surface_st.sum smsp__inst_executed_op_texture.sum smsp__sass_inst_executed_op_global_ld.sum smsp__sass_inst_executed_op_global_st.sum smsp__sass_inst_executed_op_local_ld.sum smsp__sass_inst_executed_op_local_st.sum smsp__sass_l1tex_data_pipe_lsu_wavefronts_mem_shared_op_ldgsts_cache_access.sum smsp__sass_l1tex_data_pipe_lsu_wavefronts_mem_shared_op_ldgsts_cache_access.sum.pct_of_peak_sustained_elapsed smsp__sass_l1tex_m_xbar2l1tex_read_sectors_mem_global_op_ldgsts_cache_bypass.sum smsp__sass_l1tex_m_xbar2l1tex_read_sectors_mem_global_op_ldgsts_cache_bypass.sum.pct_of_peak_sustained_elapsed |
界面所用的指标分组列表;值中的各名称指向对应指标,列表本身用于组织显示 |
group:memory__l2_cache_evict_policy_table |
lts__t_sectors_evict_first_lookup_hit.sum lts__t_sectors_evict_first_lookup_miss.sum lts__t_sectors_evict_last_lookup_hit.sum lts__t_sectors_evict_last_lookup_miss.sum lts__t_sectors_evict_normal_demote_lookup_hit.sum lts__t_sectors_evict_normal_demote_lookup_miss.sum lts__t_sectors_evict_normal_lookup_hit.sum lts__t_sectors_evict_normal_lookup_miss.sum lts__t_sectors_srcunit_tex_evict_first_lookup_hit.sum lts__t_sectors_srcunit_tex_evict_first_lookup_miss.sum lts__t_sectors_srcunit_tex_evict_last_lookup_hit.sum lts__t_sectors_srcunit_tex_evict_last_lookup_miss.sum lts__t_sectors_srcunit_tex_evict_normal_demote_lookup_hit.sum lts__t_sectors_srcunit_tex_evict_normal_demote_lookup_miss.sum lts__t_sectors_srcunit_tex_evict_normal_lookup_hit.sum lts__t_sectors_srcunit_tex_evict_normal_lookup_miss.sum lts__t_sectors_srcunit_tex_op_atom_evict_first_lookup_hit.sum lts__t_sectors_srcunit_tex_op_atom_evict_first_lookup_miss.sum lts__t_sectors_srcunit_tex_op_atom_evict_last_lookup_hit.sum lts__t_sectors_srcunit_tex_op_atom_evict_last_lookup_miss.sum lts__t_sectors_srcunit_tex_op_atom_evict_normal_lookup_hit.sum lts__t_sectors_srcunit_tex_op_atom_evict_normal_lookup_miss.sum lts__t_sectors_srcunit_tex_op_read_evict_first_lookup_hit.sum lts__t_sectors_srcunit_tex_op_read_evict_first_lookup_miss.sum lts__t_sectors_srcunit_tex_op_read_evict_last_lookup_hit.sum lts__t_sectors_srcunit_tex_op_read_evict_last_lookup_miss.sum lts__t_sectors_srcunit_tex_op_read_evict_normal_demote_lookup_hit.sum lts__t_sectors_srcunit_tex_op_read_evict_normal_demote_lookup_miss.sum lts__t_sectors_srcunit_tex_op_read_evict_normal_lookup_hit.sum lts__t_sectors_srcunit_tex_op_read_evict_normal_lookup_miss.sum lts__t_sectors_srcunit_tex_op_write_evict_first_lookup_hit.sum lts__t_sectors_srcunit_tex_op_write_evict_first_lookup_miss.sum lts__t_sectors_srcunit_tex_op_write_evict_last_lookup_hit.sum lts__t_sectors_srcunit_tex_op_write_evict_last_lookup_miss.sum lts__t_sectors_srcunit_tex_op_write_evict_normal_demote_lookup_hit.sum lts__t_sectors_srcunit_tex_op_write_evict_normal_demote_lookup_miss.sum lts__t_sectors_srcunit_tex_op_write_evict_normal_lookup_hit.sum lts__t_sectors_srcunit_tex_op_write_evict_normal_lookup_miss.sum |
界面所用的指标分组列表;值中的各名称指向对应指标,列表本身用于组织显示 |
group:memory__l2_cache_table |
lts__t_requests.sum lts__t_requests_srcunit_tex.sum lts__t_requests_srcunit_tex_op_atom_dot_alu.sum lts__t_requests_srcunit_tex_op_atom_dot_cas.sum lts__t_requests_srcunit_tex_op_read.sum lts__t_requests_srcunit_tex_op_red.sum lts__t_requests_srcunit_tex_op_write.sum lts__t_sectors.avg.pct_of_peak_sustained_elapsed lts__t_sectors.avg.peak_sustained lts__t_sectors.avg.per_cycle_elapsed lts__t_sectors.sum lts__t_sectors.sum.per_second lts__t_sectors_aperture_device_lookup_miss.sum lts__t_sectors_aperture_peer_lookup_miss.sum lts__t_sectors_aperture_sysmem_lookup_miss.sum lts__t_sectors_data_ecc.avg.pct_of_peak_sustained_elapsed lts__t_sectors_data_ecc.avg.peak_sustained lts__t_sectors_data_ecc.avg.per_cycle_elapsed lts__t_sectors_data_ecc.sum lts__t_sectors_data_ecc.sum.per_second lts__t_sectors_lookup_hit.sum lts__t_sectors_lookup_miss.sum lts__t_sectors_srcunit_tex.avg.pct_of_peak_sustained_elapsed lts__t_sectors_srcunit_tex.avg.peak_sustained lts__t_sectors_srcunit_tex.avg.per_cycle_elapsed lts__t_sectors_srcunit_tex.sum lts__t_sectors_srcunit_tex.sum.pct_of_peak_sustained_elapsed lts__t_sectors_srcunit_tex.sum.per_second lts__t_sectors_srcunit_tex_aperture_device_lookup_miss.sum lts__t_sectors_srcunit_tex_aperture_device_op_atom_dot_alu_lookup_miss.sum lts__t_sectors_srcunit_tex_aperture_device_op_atom_dot_cas_lookup_miss.sum lts__t_sectors_srcunit_tex_aperture_device_op_read_lookup_miss.sum lts__t_sectors_srcunit_tex_aperture_device_op_red_lookup_miss.sum lts__t_sectors_srcunit_tex_aperture_device_op_write_lookup_miss.sum lts__t_sectors_srcunit_tex_aperture_peer_lookup_miss.sum lts__t_sectors_srcunit_tex_aperture_peer_op_atom_dot_alu_lookup_miss.sum lts__t_sectors_srcunit_tex_aperture_peer_op_atom_dot_cas_lookup_miss.sum lts__t_sectors_srcunit_tex_aperture_peer_op_read_lookup_miss.sum lts__t_sectors_srcunit_tex_aperture_peer_op_red_lookup_miss.sum lts__t_sectors_srcunit_tex_aperture_peer_op_write_lookup_miss.sum lts__t_sectors_srcunit_tex_aperture_sysmem_lookup_miss.sum lts__t_sectors_srcunit_tex_aperture_sysmem_op_atom_dot_alu_lookup_miss.sum lts__t_sectors_srcunit_tex_aperture_sysmem_op_atom_dot_cas_lookup_miss.sum lts__t_sectors_srcunit_tex_aperture_sysmem_op_read_lookup_miss.sum lts__t_sectors_srcunit_tex_aperture_sysmem_op_red_lookup_miss.sum lts__t_sectors_srcunit_tex_aperture_sysmem_op_write_lookup_miss.sum lts__t_sectors_srcunit_tex_lookup_hit.sum lts__t_sectors_srcunit_tex_lookup_miss.sum lts__t_sectors_srcunit_tex_op_atom_dot_alu.avg.pct_of_peak_sustained_elapsed lts__t_sectors_srcunit_tex_op_atom_dot_alu.avg.peak_sustained lts__t_sectors_srcunit_tex_op_atom_dot_alu.avg.per_cycle_elapsed lts__t_sectors_srcunit_tex_op_atom_dot_alu.sum lts__t_sectors_srcunit_tex_op_atom_dot_alu.sum.per_second lts__t_sectors_srcunit_tex_op_atom_dot_alu_lookup_hit.sum lts__t_sectors_srcunit_tex_op_atom_dot_alu_lookup_miss.sum lts__t_sectors_srcunit_tex_op_atom_dot_cas.avg.pct_of_peak_sustained_elapsed lts__t_sectors_srcunit_tex_op_atom_dot_cas.avg.peak_sustained lts__t_sectors_srcunit_tex_op_atom_dot_cas.avg.per_cycle_elapsed lts__t_sectors_srcunit_tex_op_atom_dot_cas.sum lts__t_sectors_srcunit_tex_op_atom_dot_cas.sum.per_second lts__t_sectors_srcunit_tex_op_atom_dot_cas_lookup_hit.sum lts__t_sectors_srcunit_tex_op_atom_dot_cas_lookup_miss.sum lts__t_sectors_srcunit_tex_op_read.avg.pct_of_peak_sustained_elapsed lts__t_sectors_srcunit_tex_op_read.avg.peak_sustained lts__t_sectors_srcunit_tex_op_read.avg.per_cycle_elapsed lts__t_sectors_srcunit_tex_op_read.sum lts__t_sectors_srcunit_tex_op_read.sum.per_second lts__t_sectors_srcunit_tex_op_read_lookup_hit.sum lts__t_sectors_srcunit_tex_op_read_lookup_miss.sum lts__t_sectors_srcunit_tex_op_red.avg.pct_of_peak_sustained_elapsed lts__t_sectors_srcunit_tex_op_red.avg.peak_sustained lts__t_sectors_srcunit_tex_op_red.avg.per_cycle_elapsed lts__t_sectors_srcunit_tex_op_red.sum lts__t_sectors_srcunit_tex_op_red.sum.per_second lts__t_sectors_srcunit_tex_op_red_lookup_hit.sum lts__t_sectors_srcunit_tex_op_red_lookup_miss.sum lts__t_sectors_srcunit_tex_op_write.avg.pct_of_peak_sustained_elapsed lts__t_sectors_srcunit_tex_op_write.avg.peak_sustained lts__t_sectors_srcunit_tex_op_write.avg.per_cycle_elapsed lts__t_sectors_srcunit_tex_op_write.sum lts__t_sectors_srcunit_tex_op_write.sum.per_second lts__t_sectors_srcunit_tex_op_write_lookup_hit.sum lts__t_sectors_srcunit_tex_op_write_lookup_miss.sum |
界面所用的指标分组列表;值中的各名称指向对应指标,列表本身用于组织显示 |
group:memory__shared_table |
l1tex__data_bank_conflicts_pipe_lsu_mem_shared.sum l1tex__data_bank_conflicts_pipe_lsu_mem_shared_op_atom.sum l1tex__data_bank_conflicts_pipe_lsu_mem_shared_op_ld.sum l1tex__data_bank_conflicts_pipe_lsu_mem_shared_op_st.sum l1tex__data_pipe_lsu_wavefronts_mem_shared.sum l1tex__data_pipe_lsu_wavefronts_mem_shared.sum.pct_of_peak_sustained_elapsed l1tex__data_pipe_lsu_wavefronts_mem_shared_op_atom.sum l1tex__data_pipe_lsu_wavefronts_mem_shared_op_atom.sum.pct_of_peak_sustained_elapsed l1tex__data_pipe_lsu_wavefronts_mem_shared_op_ld.sum l1tex__data_pipe_lsu_wavefronts_mem_shared_op_ld.sum.pct_of_peak_sustained_elapsed l1tex__data_pipe_lsu_wavefronts_mem_shared_op_st.sum l1tex__data_pipe_lsu_wavefronts_mem_shared_op_st.sum.pct_of_peak_sustained_elapsed smsp__inst_executed_op_ldgsts.sum smsp__inst_executed_op_ldsm.sum smsp__inst_executed_op_shared_atom.sum smsp__sass_inst_executed_op_shared_ld.sum smsp__sass_inst_executed_op_shared_st.sum smsp__sass_l1tex_data_bank_conflicts_pipe_lsu_mem_shared_op_ldgsts.sum smsp__sass_l1tex_data_bank_conflicts_pipe_lsu_mem_shared_op_st.sum smsp__sass_l1tex_data_pipe_lsu_wavefronts_mem_shared_op_ldgsts.sum smsp__sass_l1tex_data_pipe_lsu_wavefronts_mem_shared_op_ldgsts.sum.pct_of_peak_sustained_elapsed smsp__sass_l1tex_data_pipe_lsu_wavefronts_mem_shared_op_st.sum smsp__sass_l1tex_data_pipe_lsu_wavefronts_mem_shared_op_st.sum.pct_of_peak_sustained_elapsed |
界面所用的指标分组列表;值中的各名称指向对应指标,列表本身用于组织显示 |
group:smsp__pcsamp_warp_stall_reasons |
smsp__pcsamp_warps_issue_stalled_barrier smsp__pcsamp_warps_issue_stalled_branch_resolving smsp__pcsamp_warps_issue_stalled_dispatch_stall smsp__pcsamp_warps_issue_stalled_drain smsp__pcsamp_warps_issue_stalled_imc_miss smsp__pcsamp_warps_issue_stalled_lg_throttle smsp__pcsamp_warps_issue_stalled_long_scoreboard smsp__pcsamp_warps_issue_stalled_math_pipe_throttle smsp__pcsamp_warps_issue_stalled_membar smsp__pcsamp_warps_issue_stalled_mio_throttle smsp__pcsamp_warps_issue_stalled_misc smsp__pcsamp_warps_issue_stalled_no_instructions smsp__pcsamp_warps_issue_stalled_not_selected smsp__pcsamp_warps_issue_stalled_selected smsp__pcsamp_warps_issue_stalled_short_scoreboard smsp__pcsamp_warps_issue_stalled_sleeping smsp__pcsamp_warps_issue_stalled_tex_throttle smsp__pcsamp_warps_issue_stalled_wait |
界面所用的指标分组列表;值中的各名称指向对应指标,列表本身用于组织显示 |
group:smsp__pcsamp_warp_stall_reasons_not_issued |
smsp__pcsamp_warps_issue_stalled_barrier_not_issued smsp__pcsamp_warps_issue_stalled_branch_resolving_not_issued smsp__pcsamp_warps_issue_stalled_dispatch_stall_not_issued smsp__pcsamp_warps_issue_stalled_drain_not_issued smsp__pcsamp_warps_issue_stalled_imc_miss_not_issued smsp__pcsamp_warps_issue_stalled_lg_throttle_not_issued smsp__pcsamp_warps_issue_stalled_long_scoreboard_not_issued smsp__pcsamp_warps_issue_stalled_math_pipe_throttle_not_issued smsp__pcsamp_warps_issue_stalled_membar_not_issued smsp__pcsamp_warps_issue_stalled_mio_throttle_not_issued smsp__pcsamp_warps_issue_stalled_misc_not_issued smsp__pcsamp_warps_issue_stalled_no_instructions_not_issued smsp__pcsamp_warps_issue_stalled_not_selected_not_issued smsp__pcsamp_warps_issue_stalled_selected_not_issued smsp__pcsamp_warps_issue_stalled_short_scoreboard_not_issued smsp__pcsamp_warps_issue_stalled_sleeping_not_issued smsp__pcsamp_warps_issue_stalled_tex_throttle_not_issued smsp__pcsamp_warps_issue_stalled_wait_not_issued |
界面所用的指标分组列表;值中的各名称指向对应指标,列表本身用于组织显示 |

浙公网安备 33010602011771号