-
Notifications
You must be signed in to change notification settings - Fork 56
add CUDA 12.5 images #689
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
add CUDA 12.5 images #689
Conversation
jakirkham
left a comment
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.
Thanks James! 🙏
Am curious what you think about swapping where the x86_64 and aarch64 jobs are
For more context CUDA 12.2 added some interesting things like cuFile (for ARM) and HMM, which are helpful to test on CUDA 12.2 ARM specifically. Am not aware of a similar change in CUDA 12.5 (though could be missing it)
Given there are fewer ARM machines than x86_64, it may make sense to use x86_64 primarily in PRs and run ARM more on branches. That would mean less ARM jobs are needed during PR development
Anyways please let me know what you think 🙂
Co-authored-by: jakirkham <[email protected]>
|
Thanks @jakirkham ! That's helpful context I didn't have. I just applied your suggestions. |
|
Updating to pull in PR ( #690 ) now that is in |
|
See this This will be broken until rapidsai/cuspatial#1405 is merged. And that is blocked until CI in |
KyleFromNVIDIA
left a comment
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.
Just one quick question. Approved so it doesn't block merging.
| ARG LINUX_DISTRO=ubuntu | ||
| ARG LINUX_DISTRO_VER=22.04 |
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.
This need not be done in this PR. Just the discussion above ( #689 (comment) ) made me wonder
Should we do the same thing for the Linux version?
| ARG LINUX_DISTRO=ubuntu | |
| ARG LINUX_DISTRO_VER=22.04 | |
| ARG LINUX_DISTRO=unset | |
| ARG LINUX_DISTRO_VER=unset |
If this sounds reasonable, happy to send a 2nd PR with this change. Also ok not doing this if we don't think it makes sense
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.
Yes in my opinion, this is a pattern that should be used in any Dockerfile where it's unacceptable to silently fall back to default values if configuration is not passed.
I just didn't want to go any further with it in this PR and grow the diff here.
I'd support a 2nd PR with this change.
|
One of the 11.8.0 tests failed on CI. Restarted it That said, we are not changing 11.8.0 builds here and all other builds pass Edit: Actually that won't work since the images get deleted after |
|
Going to try restarting from scratch. If that doesn't work, we can discuss next steps on Monday |
|
/merge |
|
Thanks all! 🙏 |
Contributes to #832 Removes reliance on https://hub.docker.com/r/rapidsai/ci-conda in this repo. It was only being used in 2 places, both of which could be run in simpler environments: * in a `compute-matrix` job that just needs `bash`, `jq` and `yq` --> regular old `ubuntu-latest` GitHub Actions VM is fine * in a part of a multi-stage build that just needs `bash`, `git`, `pip`, and a Python interpreter --> `python:{version}` container is fine Other changes based on PR reviewers: * puts versions to be managed by renovate in a `versions.yaml` ([borrowing the approach from rapidsai/ci-imgs](https://github.com/rapidsai/ci-imgs/blob/28b8b5ac127bcf5cd6679cd8ee40a53ce7240421/ci/compute-build-args.sh#L36)) * updates shell code in Dockerfiles to match our preferred style (including the array format for package lists from rapidsai/ci-imgs#343) * sets all empty arg defaults to `notset` (ref: #689 (comment)) - *weakly prefer `notset` to `unset` because `unset` is a valid shell command* * alphabetizes build args in Dockerfiles ## Notes for Reviewers ### This is only the beginning To be clear, this project still heavily relies on `rapidsai/miniforge-cuda` (the base image for `rapidsai/ci-conda`. So there's more work to do for #832. But this at least means we no longer need to care about anything in these lines breaking the things in this repo 😁 : https://github.com/rapidsai/ci-imgs/blob/55b24734b7d594e768599771dcae2e718af52f9a/ci-conda.Dockerfile#L183-L313 Authors: - James Lamb (https://github.com/jameslamb) Approvers: - Bradley Dice (https://github.com/bdice) URL: #834
Contributes to rapidsai/build-planning#73
Proposes the following:
(cuda=12.5, arch=arm64)(cuda=12.5, arch=x86_64)(cuda=12.2, arch=arm64)context: rapidsai/build-planning#73 (comment)
Notes for Reviewers
Per offline discussion with @raydouglass , this would be accompanied by a deprecation notice in the RAPIDS 24.08 release stating that the CUDA 12.2 images will be removed in some future release (future release not yet determined).