cudaPackages_12: 12.4 -> 12.8 #390885
Merged
cudaPackages_12: 12.4 -> 12.8 #390885
+68
−29
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
github-actions
bot
added
6.topic: cuda
prusnak
force-pushed
the
cuda-12.8
branch
2 times, most recently
from
5085730 to
ed2719b
Compare
|
I successfully built |
|
@SomeoneSerge i think something went wrong with your review - most of the packages that failed in your review build for me (for example |
prusnak
requested review from
ConnorBaker,
GaetanLepage,
SomeoneSerge and
samuela
prusnak
force-pushed
the
cuda-12.8
branch
3 times, most recently
from
809ae9c to
68baed4
Compare
|
I don't have the computing power to run I guess that it's up to @SomeoneSerge and/or @ConnorBaker to make a call on this. |
|
One of your commits is missing a "c" at the beginning ( |
13 tasks
Yes, I think the machine must've run out of storage...
I'll just voice my more general stance on this, for the future: if people become unresponsive it's perfectly reasonable to move stuff in lack of their feedback, assuming due diligence. Gaetan pointed out on matrix that this update is becoming a blocker for other work such as fixing mistral-rs. We've previously already gone by the route of "update and face the fallout", I think we must do so again until we have resources to do more than minimum maintenance |
Comment on lines
201
to
207
| # NVIDIA B100 Accelerated | ||
| archName = "Blackwell"; | ||
| computeCapability = "10.0a"; | ||
| isJetson = false; | ||
| minCudaVersion = "12.8"; | ||
| dontDefaultAfter = null; | ||
| maxCudaVersion = null; |
There was a problem hiding this comment.
Instead of removing metadata let's add some boolean for filtering it out in default capabilities computation.
There was a problem hiding this comment.
So you are suggesting setting let's say dontDefaultAfter = "12.0" with a comment?
The |
Not torch. It builds with older CUDA with no issues. |
|
Okay, will update the PR later today and merge. I will add myself to the CUDA maintainers team in nixpkgs, if you don't mind, so I can catch the fallout if any. |
github-actions
bot
added
the
6.topic: teams
…12.0a to save space
RTX 50{7,8,9}0 is Compute Capability 12.0
GaetanLepage
approved these changes
Mar 25, 2025
Give me few hours, I am running the latest build, I will merge then. |
wegank
added
the
12.approvals: 1
Welcome on the team! Consider opening a PR to https://github.com/NixOS/nixos-homepage/ if you haven't yet |
|
So it seems the fallout was small. The new build failures are listed here: https://hydra.nix-community.org/eval/356757
and all are caused by the failure in the What puzzles me is that the same failure already happened earlier with older CUDA and was reported here: #348386 Will investigate and try to fix the issue in the upcoming days. |
13 tasks
It was easier than I thought - GCC14 compatibility - fix in #393413 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
I finally managed to fix onnxruntime build with CUDA 12.8 which was AFAIK the only blocker for changing the default.
The fix is to turn off LTO when linking with CUDA, because otherwise the linking takes too much time and then ends up in infinite loop anyway.
Also in the past I enthusiastically added support for Compute Capability 10.0 and 10.1 GPUs but this is causing linking errors for onnxruntime (resulting shared library >2 GB, similar to what we face with magma), so I removed them and only kept CC 12.0 to save space (e.g. ollama upstream also does not build for CC 10.x but only for 12.0) and this makes it possible for onnxruntime shared library to link again.
Things done
nix.conf? (See Nix manual)sandbox = relaxedsandbox = truenix-shell -p nixpkgs-review --run "nixpkgs-review rev HEAD". Note: all changes have to be committed, also see nixpkgs-review usage./result/bin/)Add a 👍 reaction to pull requests you find important.