Hi there, I am running an older version of the lambda stack which came with an nvidia driver supporting cuda 11.6. I can’t upgrade the stack itself right now, because it could break compatibility with currently running services.
I tried installing your docker image with a custom edited fork of your repo, where I changed the dummy dependency references to 11.6. But when I try to install the docker image, it keeps installing cuda 11.8, which is incompatible with the host system’s cuda 11.6 version:
sudo docker build -t lambda-stack:20.04-11.6 -f Dockerfile.focal https://github.com/Daryl149/lambda-stack-dockerfiles.git
So when running
sudo docker run --gpus all lambda-stack:20.04-11.6 python3 -c 'import torch; print(torch.cuda.device_count())'
I receive this error:
Error 804: forward compatibility was attempted on non supported HW (Triggered internally at ../c10/cuda/CUDAFunctions.cpp:109.)
I think it’s because 11.8 is set in this line in your repo:
echo "deb http://archive.lambdalabs.com/ubuntu focal main" > /etc/apt/sources.list.d/lambda.list
but I am unsure how to switch/browse to an older branch that references 11.6.