一. 更新显卡驱动: 参考博客:https://blog.youkuaiyun.com/gdymind/article/details/82153643
可以用nvidia-smi命令进行查看:
nvidia-smi
二. 安装cuda10,cudnn,可以参考博客:https://blog.youkuaiyun.com/qq_32408773/article/details/84112166,
https://blog.youkuaiyun.com/u010801439/article/details/80483036
1. NVIDIA 各个版本的cuda: https://developer.nvidia.com/cuda-toolkit-archive
2. cudnn下载:需要注册帐号才能下载:https://developer.nvidia.com/rdp/cudnn-download
3. 安装好后,可以用nvcc -V,查看cuda版本
nvcc -V
4. 查看显卡信息:https://blog.youkuaiyun.com/a784586/article/details/78688842
三. 安装Anaconda3,参考官网:https://www.anaconda.com/distribution/
1. 早期版本下载,可以参考网址:https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/?C=N&O=D
anaconda与python版本的对应:https://blog.youkuaiyun.com/weixin_40444270/article/details/83420020
2. 配置清华源,可以参考博客:https://blog.youkuaiyun.com/yangzhengzheng95/article/details/86222094
3. 安装好后,可以参考博客创建虚拟环境等:https://blog.youkuaiyun.com/yangzhengzheng95/article/details/86222094
四. 安装pytorch,参见pytorch官网:https://pytorch.org/,及博客:https://blog.youkuaiyun.com/yangzhengzheng95/article/details/86222094
pytorch中查看gpu的信息
import torch
torch.cuda.is_available() # cuda是否可用
torch.cuda.device_count() # gpu的数量
torch.cuda.get_device_name(0) # 0号gpu的名字
注:在ubuntu下可能存在可以通过命令行导入pytorch,但是在pycharm中不能导入,并且报如下错误:
ImportError: libcublas.so.9.0: cannot open shared object file: No such file or directory,解决方法,一般是cuda添加环境变量的问题,可以参考如下博客:
https://blog.youkuaiyun.com/qq_34374211/article/details/81018320
https://blog.youkuaiyun.com/SJTUzhou/article/details/83931884