Hi, I’m using the university cluster with CUDA 11.4 but I failed to install DGL. I’m wondering can we install DGL on the server with cuda 11.4?
I don’t think so. 11.3 the latest version DGL supports and so does pytorch.
Could you try dgl with cuda 11.3? It might work
Hi, I tried all the install command with pip, all doesn’t work
if so, you have to use 11.3. any reason to use 11.4 instead of 11.3? new features?
No specific reason. I’m running it on the cluster and the docker is provided by the college. It seems like I cannot change the path of CUDA driver.
Could you raise an issue at github repo? Thun we’ll track it there as a feature request
Sure! I will do that
NVIDIA is providing early access for DGL container with latest CUDA on NGC. Check this page Deep Learning Frameworks | NVIDIA Developer