Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Update check_nvlink_connectivity #1684

Merged
merged 3 commits into from
Aug 21, 2024
Merged

Conversation

sanderland
Copy link
Contributor

@sanderland sanderland commented Aug 20, 2024

Just a small fix to this function which was checking non-gpu rows.

Example topo that was failing
image

@rasbt
Copy link
Collaborator

rasbt commented Aug 21, 2024

Thanks for the PR! This seems like a reasonable update. I will take a closer look and try it out tomorrow.

@sanderland
Copy link
Contributor Author

Tests fail, but it looks like the mock output is fully connected (at least in the GPU-GPU block).

@rasbt
Copy link
Collaborator

rasbt commented Aug 21, 2024

Yes, this was a weird one. I think we had a different requirement back when we implemented this. But in my opinion, this test you mentioned should check "All GPUs are fully connected via NVLink." instead. That's because all GPUs are well-connected like you said (even though there are other accelerators that aren't).

I restructured the test and code a bit, so it should be fine now. Thanks again for the PR!

@rasbt rasbt enabled auto-merge (squash) August 21, 2024 14:40
@rasbt rasbt merged commit 68e69c7 into Lightning-AI:main Aug 21, 2024
8 of 9 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants