cuda9.2

cuda对系统有要求，cuda9.2需要ubuntu16

换上阿里云的ubuntu16的源
然后apt update
# "10-0" 表示安装CUDA10.0版本的 nvcc, 应根据镜像中cuda版本安装

1. 检查系统版本

   source /etc/os-release && echo $VERSION_ID

2. 导入apt仓库和其GPG key

# 第一步若输出 16.04 
curl -fsSL https://mirrors.aliyun.com/nvidia-cuda/ubuntu1604/x86_64/7fa2af80.pub | apt-key add -
echo "deb https://mirrors.aliyun.com/nvidia-cuda/ubuntu1604/x86_64/ /" > /etc/apt/sources.list.d/cuda.list

# 第一步若输出 18.04 
curl -fsSL https://mirrors.aliyun.com/nvidia-cuda/ubuntu1804/x86_64/7fa2af80.pub | apt-key add -
echo "deb https://mirrors.aliyun.com/nvidia-cuda/ubuntu1804/x86_64/ /" > /etc/apt/sources.list.d/cuda.list

3. 安装所需版本的 nvcc

apt install cuda-command-line-tools-9-2

 

https://matpool.com/supports/reference/faqs/#%E5%A6%82%E4%BD%95%E5%AE%89%E8%A3%85-nvcc%EF%BC%9F

然后用这个代码安装
apt-get update && apt-get install -y --no-install-recommends \
cuda-libraries-9-2=9.2.148-1 \
cuda-npp-9-2=9.2.148-1 \
cuda-nvtx-9-2=9.2.148-1 \
cuda-cublas-9-2=9.2.148.1-1 \
&& rm -rf /var/lib/apt/lists/*

 

最后还要软连接一下https://www.jianshu.com/p/fd0f84f858f8

进到/usr/local 删除cuda softlink rm -rf cuda and create another one ln -s cuda9.2 cuda

check nvcc --version

 

然后装一下cudnn：

cp cudnn-10.0-linux-x64-v7.3.1.20.solitairetheme8 cudnn-10.0-linux-x64-v7.3.1.20.tgz
tar -xvf cudnn-10.0-linux-x64-v7.3.1.20.tgz

cp cuda/include/cudnn.h /usr/local/cuda/include/

cp cuda/lib64/libcudnn* /usr/local/cuda/lib64/

chmod a+r /usr/local/cuda/include/cudnn.h

chmod a+r /usr/local/cuda/lib64/libcudnn*

查看版本：cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2

 

然后还要修改一下gcc：

#sudo apt-get install gcc-5.4
#sudo apt-get install g++-5.4

the webset of gcc http://ftp.gnu.org/gnu/gcc/

ls -l gcc*  发现lrwxrwxrwx 1 root root 7th May 16 18:16 /usr/bin/gcc -> gcc-7.3

然后我要sudo mv gcc gcc.bak #备份
　　　　sudo ln -s gcc-4.8 gcc #重新链接

g++如果有需要的话也是同理

show vision gcc -v g++ -v

 

有用的网站http://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/

https://github.com/CharlesShang/DCNv2/issues/11   cusparse.h从cuda9.0那里复制过来也没事，我就是这么做的

 

error:

 RuntimeWarning: numpy.dtype size changed, may indicate binary incompatibility. Expected 96, got 88
from .tslib import iNaT, NaT, Timestamp, Timedelta, OutOfBoundsDatetime

 

pip install --no-binary pandas -I pandas


https://blog.softhints.com/python-warning-numpy-dtype-size-changed-may-indicate-binary-incompatibility/