Does CUDA version matter?

[johnny-rodriguez]

I checked if Alphafold can use the GPU by running the following: docker run --rm --gpus all nvidia/cuda:11.0-base nvidia-smi

I get the following error: Unable to find image 'nvidia/cuda:11.0-base' locally docker: Error response from daemon: manifest for nvidia/cuda:11.0-base not found: manifest unknown: manifest unknown. See 'docker run --help'.

However, I know I have a newer version of cuda. If I run the following, everything looks great: docker run --rm --gpus all nvidia/cuda:11.6.2-base-ubuntu20.04 nvidia-smi

Will alphafold only recognize a specific version of CUDA, or is this good enough to run alphafold?

[johnny-rodriguez]

I discovered that nvidia/cuda:11.0-base does not exist anymore. As seen in the pull requests, I have updated to nvidia/cuda:11.0.3-base. Is there anything I need to change in the code to update this fix?

[dr-dvice]

I'm curious to know the answer to this as well, newer linux distros don't support the CUDA version shown in the installation instructions.

[rocketman8080]

+1 would really help to get the supported / recommended versions