Cuda and torch installation in python venv

1

Ubuntu 18.04, Python 3.6.9.
I created python environment and install cuda 10.2, then
pip3 install torch==1.11.0+cu102 torchvision==0.12.0+cu102 torchaudio==0.11.0 --extra-index-url whl/cu102
But then I discovered that
NVIDIA GeForce RTX 3060 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.
so I uninstalled cuda 10.2 and torch, and installed cuda 11.7.
And now pip3 install --no-cache-dir torch torchvision torchaudio, gives me torch for cuda 10.2
Collecting torch
Downloading files.pythonhosted.org/packages/a4/54/81b1c3c574a1ffde54b0c82ed2a37d81395709cdd5f50e59970aeed5d95e/torch-1.10.2-cp36-cp36m-manylinux1_x86_64.whl (881.9MB)
I tried other options, but without success
pip3 install torch==2.0.0+cu117 torchvision==0.15.1+cu117 torchaudio==2.0.1 --index-url cu117
Collecting torch==2.0.0+cu117
Could not find a version that satisfies the requirement torch==2.0.0+cu117 (from versions: )
No matching distribution found for torch==2.0.0+cu117
pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 torchaudio==0.13.1 --extra-index-url cu117
Collecting torch==1.13.1+cu117
Could not find a version that satisfies the requirement torch==1.13.1+cu117 (from versions: 1.0.0, 1.0.1, 1.0.1.post2, 1.1.0, 1.2.0, 1.3.0, 1.3.1, 1.4.0, 1.5.0, 1.5.1, 1.6.0, 1.7.0, 1.7.1, 1.8.0, 1.8.1, 1.9.0, 1.9.1, 1.10.0, 1.10.1, 1.10.2)
No matching distribution found for torch==1.13.1+cu117

Thanks in advance.

2

The error is expected since your Ampere GPU needs CUDA 11.x.
The posted install commands use an invalid --index-url since you are only specifying cu117 instead of the URL. Refer to the install commands from our website.

3

Yes, GPU needs CUDA 11.x. The main question is why pip3 install --no-cache-dir torch torchvision torchaudio, gives me torch for cuda 10.2, not torch for cuda 11.x?
I’m new to the site and have a limit on the number of links, so I removed some of them.

4

The current binaries are not built with CUDA 10.2 anymore, but your environment still picks it up.
I guess it’s because you are using a deprecated Python 3.6 release, which isn’t supported anymore since 23rd Dec 2021, so you might want to update Python.
The command also works for me properly:

pip install --no-cache-dir torch torchvision torchaudio
Looking in indexes: https://pypi.org/simple, https://pypi.ngc.nvidia.com
Collecting torch
  Downloading torch-2.0.1-cp310-cp310-manylinux1_x86_64.whl (619.9 MB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 619.9/619.9 MB 31.3 MB/s eta 0:00:00
Collecting torchvision
  Downloading torchvision-0.15.2-cp310-cp310-manylinux1_x86_64.whl (6.0 MB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 6.0/6.0 MB 27.8 MB/s eta 0:00:00
Collecting torchaudio
  Downloading torchaudio-2.0.2-cp310-cp310-manylinux1_x86_64.whl (4.4 MB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 4.4/4.4 MB 46.1 MB/s eta 0:00:00
Requirement already satisfied: filelock in /usr/local/lib/python3.10/dist-packages (from torch) (3.12.2)
Requirement already satisfied: typing-extensions in /usr/local/lib/python3.10/dist-packages (from torch) (4.6.3)
Requirement already satisfied: sympy in /usr/local/lib/python3.10/dist-packages (from torch) (1.12)
Requirement already satisfied: networkx in /usr/local/lib/python3.10/dist-packages (from torch) (2.6.3)
Requirement already satisfied: jinja2 in /usr/local/lib/python3.10/dist-packages (from torch) (3.1.2)
Collecting nvidia-cuda-nvrtc-cu11==11.7.99 (from torch)
  Downloading nvidia_cuda_nvrtc_cu11-11.7.99-2-py3-none-manylinux1_x86_64.whl (21.0 MB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 21.0/21.0 MB 52.2 MB/s eta 0:00:00
Collecting nvidia-cuda-runtime-cu11==11.7.99 (from torch)
  Downloading nvidia_cuda_runtime_cu11-11.7.99-py3-none-manylinux1_x86_64.whl (849 kB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 849.3/849.3 kB 65.4 MB/s eta 0:00:00
Collecting nvidia-cuda-cupti-cu11==11.7.101 (from torch)
  Downloading nvidia_cuda_cupti_cu11-11.7.101-py3-none-manylinux1_x86_64.whl (11.8 MB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 11.8/11.8 MB 63.0 MB/s eta 0:00:00
Collecting nvidia-cudnn-cu11==8.5.0.96 (from torch)
  Downloading nvidia_cudnn_cu11-8.5.0.96-2-py3-none-manylinux1_x86_64.whl (557.1 MB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 557.1/557.1 MB 116.2 MB/s eta 0:00:00
...
Successfully installed lit-16.0.6 nvidia-cublas-cu11-11.10.3.66 nvidia-cuda-cupti-cu11-11.7.101 nvidia-cuda-nvrtc-cu11-11.7.99 nvidia-cuda-runtime-cu11-11.7.99 nvidia-cudnn-cu11-8.5.0.96 nvidia-cufft-cu11-10.9.0.58 nvidia-curand-cu11-10.2.10.91 nvidia-cusolver-cu11-11.4.0.1 nvidia-cusparse-cu11-11.7.4.91 nvidia-nccl-cu11-2.14.3 nvidia-nvtx-cu11-11.7.91 torch-2.0.1 torchaudio-2.0.2 torchvision-0.15.2 triton-2.0.0```
1 Like
5

Yes, the problem was in Python 3.6.
I changed it to 3.8 and it helped.
Thank you!

1 Like