ubuntu18.04 cuda11.0 cudnn8.0.4 tensorflow2.4.0

腾讯云创建实例,选择GPU nvidia-V100,腾讯云服务器没有阿里云做的方便,创建实例请不要勾选系统自动安装cuda

ubuntu 18.04 python3

https://developer.nvidia.com/cuda-toolkit-archive

版本:CUDA Toolkit 11.0

下载文件到本地:

http://developer.download.nvidia.com/compute/cuda/11.0.2/local_installers/cuda_11.0.2_450.51.05_linux.run

不需要另外安装驱动,这个包里有驱动程序安装包,完全安装即可

安装:

chmod +x cuda_11.0.2_450.51.05_linux.run
./cuda_11.0.2_450.51.05_linux.run

全部安装,安装时间大概10分钟

以下步骤校验安装是否成功

cd /usr/local/cuda-10.1/samples/1_Utilities/deviceQuery
make
./deviceQuery

cuda安装完成后重启系统

reboot

重启后查看版本号

nvidia-smi

返回结果

Fri Feb 26 20:11:42 2021       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 450.51.05    Driver Version: 450.51.05    CUDA Version: 11.0     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  Tesla V100-SXM2...  On   | 00000000:00:08.0 Off |                    0 |
| N/A   37C    P0    55W / 300W |  30590MiB / 32510MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
|   1  Tesla V100-SXM2...  On   | 00000000:00:09.0 Off |                    0 |
| N/A   37C    P0    56W / 300W |  30590MiB / 32510MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
|   2  Tesla V100-SXM2...  On   | 00000000:00:0A.0 Off |                    0 |
| N/A   35C    P0    52W / 300W |  30590MiB / 32510MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
|   3  Tesla V100-SXM2...  On   | 00000000:00:0B.0 Off |                    0 |
| N/A   34C    P0    51W / 300W |  30590MiB / 32510MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

下载cudnn

https://developer.nvidia.com/rdp/cudnn-archive

版本:Download cuDNN v8.0.4 (September 28th, 2020), for CUDA 11.0

cuDNN Library for Linux (x86_64)

需要登录

下载后将后缀名修改为:tgz

解压:解压后将 cuda->include cuda->lib64 2个目录下的所有文件拷贝到

/usr/local/cuda-11.0->include

/usr/local/cuda-11.0->lib64

配置环境变量

vi ~/.bashrc
export CUDA_HOME=/usr/local/cuda
export PATH=$PATH:$CUDA_HOME/bin
export LD_LIBRARY_PATH=/usr/local/cuda-11.0/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
source ~/.bashrc

查看是否生效

cat /proc/driver/nvidia/version
nvcc -V

tensorflo2.4.0 自动识别是否支持GPU

先升级pip,ubuntu18.04 系统自带python3.6版本

pip3 install --upgrade pip
pip3 install tensorflow

最后检查 tensorflow 是否有可用GPU

import tensorflow as tf
print(tf.test.is_gpu_available())

返回结果中包含如下信息,表示成功

...
2021-02-26 20:34:32.784852: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1267]      0 1 2 3 
2021-02-26 20:34:32.784862: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1280] 0:   N Y Y Y 
2021-02-26 20:34:32.784869: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1280] 1:   Y N Y Y 
2021-02-26 20:34:32.784876: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1280] 2:   Y Y N Y 
2021-02-26 20:34:32.784882: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1280] 3:   Y Y Y N 
...
True