-
Notifications
You must be signed in to change notification settings - Fork 538
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
Use CUDA math wheels #5966
Use CUDA math wheels #5966
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I have some questions for discussion / gaining understanding.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I have no reservations about this approach, aside from the open question on DLFW rpath modifications, so I’m approving. Before we merge we should figure out what timeline we want to use to roll this out across RAPIDS. 24.08? 24.10?
Given that 24.08 is now in burndown, I'm not sure it makes sense to make a large sweeping change like this so late, but I'm open to counter-arguments. |
I agree. I think we are going to need to push wheel improvements to 24.10 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Can you update the PR title and description to reflect the current state (remove DO NOT MERGE etc)? This approach looks solid to me now.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@KyleFromNVIDIA This is pretty clean now. Can you do a little writeup on the effect this PR has on wheel sizes, wheel file contents, and build times?
Comparing against 990db71 with Python 3.11, CUDA 12, x86_64... OldBuild Time: 10:41
NewBuild Time: 6:53
|
/merge |
Use CUDA math wheels to reduce wheel size by not statically linking CUDA math libraries.
Contributes to rapidsai/build-planning#35