Unable to get CUDA to work on conda enviornment Ubuntu 22.04

Question

I am trying to install Cuda 11.3 (since i understand this is the most stable one and any version after leads to compatibility issues). I have Ubuntu 22.04 LTS and I followed these instructions: https://developer.nvidia.com/cuda-11-3-1-download-archive?target_os=Linux&target_arch=x86_64&Distribution=Ubuntu&target_version=20.04&target_type=deb_local

When I enter all the commands I do nvidia-smi and I get this:

| NVIDIA-SMI 520.61.05    Driver Version: 520.61.05    CUDA Version: 11.8     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA GeForce ...  On   | 00000000:01:00.0 Off |                  N/A |
|  0%   37C    P8     8W / 220W |     99MiB /  8192MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|    0   N/A  N/A      1245      G   /usr/lib/xorg/Xorg                 86MiB |
|    0   N/A  N/A      1498      G   /usr/bin/gnome-shell               10MiB |
+-----------------------------------------------------------------------------+

But when I do nvcc --version it returns Command 'nvcc' not found, but can be installed with: sudo apt install nvidia-cuda-toolkit. If I run the install command for nvidia-cuda-toolkit the nvidia-smi command stops working and when I test if cuda is being used with pytorch, I keep getting torch.cuda.is_available() as False.

I did sudo ubuntu-drivers autoinstall to install the drivers and nothing changed. The weird part is that in my base conda enviornment, torch.cuda.is_available() returns true. I would really appreciate it if anyone could help out.

Thanks!

Answer 1

I had the same issue after updating to Ubuntu 22.04. Following instructions from https://pytorch.org/ I went with cuda 11.6 even though my nvidia-smi gave the Cuda Version as 11.7

+-----------------------------------------------------------------------------+ | NVIDIA-SMI 515.76 Driver Version: 515.76 CUDA Version: 11.7 | |-------------------------------+----------------------+--------------------

NOTE: 'conda-forge' channel is required for cudatoolkit 11.6

conda install pytorch torchvision torchaudio cudatoolkit=11.6 -c pytorch -c conda-forge

python version for my conda environment is Python 3.10.6

Doesn't look like cuda driver version at operating system level has an impact on with cudatoolkit packaged with pytorch