Pytorch GPU via conda not working

1

I want to run pytorch on GPU (within conda) with the following settings but all attempts failed.

  • Ubuntu 24
  • GPU: RTX 4090
  • Driver Version: 550.120 (checked via nvidia-smi)
  • No system-wide cuda installation
  • pytorch install commands I tried:
    conda install pytorch torchvision torchaudio pytorch-cuda=12.4 -c pytorch -c nvidia
    conda install pytorch torchvision torchaudio pytorch-cuda=12.4 -c pytorch-nightly -c nvidia

What would you recommend?

2

Check if the right binary was installed via print(torch.version.cuda). If it’s showing 12.4, the binary might have trouble communicating with the NVIDIA driver.

3

Thank you but this returns None.
I further checked:

  • gcc (Ubuntu 13.2.0-23ubuntu4) 13.2.0
  • torch.version is ‘2.4.1’
  • conda list:
    cuda-cudart 12.4.127 0 nvidia
    cuda-cupti 12.4.127 0 nvidia
    cuda-libraries 12.4.1 0 nvidia
    cuda-nvrtc 12.4.127 0 nvidia
    cuda-nvtx 12.4.127 0 nvidia
    cuda-opencl 12.6.77 0 nvidia
    cuda-runtime 12.4.1 0 nvidia
    cuda-version 12.6 3 nvidia
    where I guess the 12.6 version might be causing the issue?
4

No, I don’t think so as the install command works for me in a new and empty conda environment:

conda install pytorch torchvision torchaudio pytorch-cuda=12.4 -c pytorch -c nvidia
...
The following packages will be downloaded:

    package                    |            build
    ---------------------------|-----------------
    blas-1.0                   |              mkl           6 KB
    brotli-python-1.0.9        |  py310h6a678d5_8         356 KB
    certifi-2024.8.30          |  py310h06a4308_0         162 KB
    cuda-cudart-12.4.127       |                0         198 KB  nvidia
    cuda-cupti-12.4.127        |                0        16.4 MB  nvidia
    cuda-libraries-12.4.1      |                0           2 KB  nvidia
    cuda-nvrtc-12.4.127        |                0        21.0 MB  nvidia
    cuda-nvtx-12.4.127         |                0          58 KB  nvidia
    cuda-opencl-12.6.77        |                0          25 KB  nvidia
    cuda-runtime-12.4.1        |                0           2 KB  nvidia
    cuda-version-12.6          |                3          16 KB  nvidia
...
    pytorch-2.5.1              |py3.10_cuda12.4_cudnn9.1.0_0        1.46 GB  pytorch
    pytorch-cuda-12.4          |       hc786d27_7           7 KB  pytorch
    pytorch-mutex-1.0          |             cuda           3 KB  pytorch
...

python -c "import torch; print(torch.__version__); print(torch.version.cuda); print(torch.cuda.is_available()); print(torch.randn(1).cuda())"
2.5.1                                                                                                                                                                                                                                                                             
12.4                                                                                                                                                                                                                                                                              
True                                                                                                                                                                                                                                                                              
tensor([0.3827], device='cuda:0')

However, you could also try to use the pip wheels if you are seeing issues with conda as conda binaries will be deprecated as mentioned here.

5

Thank you.
Resolved: creating a new (and empty!) conda env helped so the original
conda install pytorch torchvision torchaudio pytorch-cuda=12.4 -c pytorch -c nvidia
works now.