Correspondence between rtx3090 and pytorch versions

stay RTX 3090 Judge , The current version of torch Whether the version can be used , Generally, the following methods are required ：

1. conda activate torch1.8.1 ( Activate the relevant virtual environment )
2. python Get into python Environmental Science ,　import torch Import torch Installation package ;
3. test 　torch.cuda.is_available(),
4. test torch.zero(1).cuda()

until , Step 4 complete , To describe the current version of cuda　 You can call the current version of pytorch;

Key points of the problem ：

1. install pytorch In the process ,　 There are two points to note ,　 One is currently installed pytorch edition ,　 The pytroch The official version website specifies which versions are included cuda edition ;
2. Use pip install torch＝＝1.8.1 Mode of installation , The default is torch edition ＋　 On the current host cuda edition
3. Possible problems ,　 Current host cuda edition 　 Incompatible with this torch Several official releases in the version cuda　 edition ;

1. Problem phenomenon

>>> torch.zeros(1).cuda()
/home/respecting/anaconda3/envs/torch1.8.1/lib/python3.7/site-packages/torch/cuda/__init__.py:104: UserWarning: 
NVIDIA GeForce RTX 3090 with CUDA capability sm_86 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_70.
If you want to use the NVIDIA GeForce RTX 3090 GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/

1.1 Problem analysis

Indicates the currently installed pytorch The version does not match the appropriate cuda, 　 At present pytorch Version of cuda The version does not correspond to its own host , Installed cuda　 edition ,

pytorch Installed in cuda edition ,　 There are two conditions that need to be met ：

1. At present pytorch Version of computing power support 　 The computing power of the graphics card on the current machine ;
2. pytorch 　 Medium cuda edition 　 It cannot be higher than the one already installed on the current machine cuda　 edition ;

Specifically ,　 The same pytorch edition , such as pytorch 1.8.1 Will correspond to different versions of cuda

# ROCM 4.0.1 (Linux only)
pip install torch==1.8.1+rocm4.0.1 torchvision==0.9.1+rocm4.0.1 torchaudio==0.8.1 -f https://download.pytorch.org/whl/torch_stable.html

# ROCM 3.10 (Linux only)
pip install torch==1.8.1+rocm3.10 torchvision==0.9.1+rocm3.10 torchaudio==0.8.1 -f https://download.pytorch.org/whl/torch_stable.html

# CUDA 11.1
pip install torch==1.8.1+cu111 torchvision==0.9.1+cu111 torchaudio==0.8.1 -f https://download.pytorch.org/whl/torch_stable.html

# CUDA 10.2
pip install torch==1.8.1+cu102 torchvision==0.9.1+cu102 torchaudio==0.8.1 -f https://download.pytorch.org/whl/torch_stable.html

# CUDA 10.1
pip install torch==1.8.1+cu101 torchvision==0.9.1+cu101 torchaudio==0.8.1 -f https://download.pytorch.org/whl/torch_stable.html

# CPU only
pip install torch==1.8.1+cpu torchvision==0.9.1+cpu torchaudio==0.8.1 -f https://download.pytorch.org/whl/torch_stable.html

1.2　 solve the problem

pytorch Installed in cuda edition ,　 There are two conditions that need to be met ：

1. At present pytorch Version of computing power support 　 The computing power of the graphics card on the current machine ;
2. pytorch Medium cuda The version cannot be higher than the one already installed on the current machine cuda　 edition ;

Knowing the cause of the problem ,　 We can solve the problem ：

1. RTX3090 Need at least cuda 11.1 edition , To drive the device , So we can install cuda11.1 Above version
2. So when you want to install pytorch In the version , Find greater than 　 cuda11.1 <= pytorch-cuda --version <= 　 Installed on the current machine cuda --version

Because the author installed on the machine is cuda11.2 ,　 and 3090 Corresponding cuda Version must be greater than or equal to cuda11.1,
Therefore, installation pytorch 1.8.1 Medium 　cuda11.1 edition , 　 Uninstall and reinstall the corresponding version ;

 pip install -i https://pypi.douban.com/simple torch-1.8.1+cu111-cp37-cp37m-linux_x86_64.whl 
Looking in indexes: https://pypi.douban.com/simple
Processing ./torch-1.8.1+cu111-cp37-cp37m-linux_x86_64.whl
Requirement already satisfied: numpy in /home/respecting/anaconda3/envs/torch1.8.1/lib/python3.7/site-packages (from torch==1.8.1+cu111) (1.21.6)
Requirement already satisfied: typing-extensions in /home/respecting/anaconda3/envs/torch1.8.1/lib/python3.7/site-packages (from torch==1.8.1+cu111) (4.2.0)
Installing collected packages: torch
  Attempting uninstall: torch
    Found existing installation: torch 1.8.1
    Uninstalling torch-1.8.1:
      Successfully uninstalled torch-1.8.1
Successfully installed torch-1.8.1+cu111

torchvision install

(torch1.8.1) [email protected]:/media/respecting/Ubuntu 18.0/June18$  pip install -i https://pypi.douban.com/simple torch-1.8.1+cu111-cp37-cp37m-linux_x86_64.whl 
Looking in indexes: https://pypi.douban.com/simple
Processing ./torch-1.8.1+cu111-cp37-cp37m-linux_x86_64.whl
Requirement already satisfied: numpy in /home/respecting/anaconda3/envs/torch1.8.1/lib/python3.7/site-packages (from torch==1.8.1+cu111) (1.21.6)
Requirement already satisfied: typing-extensions in /home/respecting/anaconda3/envs/torch1.8.1/lib/python3.7/site-packages (from torch==1.8.1+cu111) (4.2.0)
Installing collected packages: torch
  Attempting uninstall: torch
    Found existing installation: torch 1.8.1
    Uninstalling torch-1.8.1:
      Successfully uninstalled torch-1.8.1
Successfully installed torch-1.8.1+cu111

2. To test

(torch1.8.1) [email protected]:~$ python
Python 3.7.13 (default, Mar 29 2022, 02:18:16) 
[GCC 7.5.0] :: Anaconda, Inc. on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> torch.zeros(1).cuda()
tensor([0.], device='cuda:0')
>>>