ci: linux CPU + CUDA 12.4 backends, consumer-only CUDA SASS - #56
Merged
Conversation
Target the consumer backbone and shrink the CUDA archives: - new linux-x64-cpu: pure CPU-only build (no GPU deps) so CPU users don't have to grab a GPU archive - new linux-x64-cuda-12.4: CUDA 12.4 works on older drivers (13.x needs a newer one) — the most-downloaded variant on Windows, now on Linux too - CUDA arch: 80;86;89;90;120 -> 75-virtual;86-real;89-real;120-real. Ship SASS only for consumer cards (RTX 30/40/50); drop server SASS (A100/H100/H200/B200) which JIT from the Turing PTX floor. Smaller archives, older consumer (Turing+) now covered via JIT too. - wire both into dev-build and release (needs/EXPECTED/notes); also folds the #55 notes wording fix (no em-dashes, 'AMD ROCm' not 'ROCm/HIP')
…ckwell) CUDA 12.4's nvcc fails with 'Unsupported gpu architecture compute_120' -- sm_120 (Blackwell) was added in CUDA 12.8. The 12.4 build is for older cards on older drivers anyway, so give it a wide old-card net instead: 61-virtual;75-real;86-real;89-real (Pascal PTX floor + Turing/Ampere/Ada SASS). The 13.3 job keeps sm_120 for RTX 50.
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Targets the consumer backbone and shrinks the CUDA archives (the 'hippo').
New backends
CUDA SASS trim
80;86;89;90;120→75-virtual;86-real;89-real;120-real:Also
New backends = feature → the next release is minor (b10018-1.3.0). This PR's dev-build run compiles both new jobs, so a green check = they build.
Not in scope (next batch): SYCL / OpenVINO / Windows ROCm — those ship without TurboQuant KV and need their own labeling decision.