Repository navigation
[Engine] the experimental CUDA 12 build runs on the Maxwell cards (sm_50) - #3595
Closed
msprengholz wants to merge 1 commit into
Closed
msprengholz wants to merge 1 commit into
msprengholz wants to merge 1 commit into
Conversation
…ards CUDA 13 dropped Maxwell, so those cards need the experimental CUDA 12 engine, and that engine refused anything below compute capability 6.0. A GTX TITAN X (sm_52) is a 12 GB card with enough memory for a small model, and the kernels compile for it, so the gate was the only thing in the way. The gate now admits 5.0, setup's card choice admits the Maxwell cards (and still refuses Kepler, sm_35, for which no engine code exists), and the engine's own check says 5.0 or newer in the experimental build. The release CUDA 13 engine is unchanged: it still needs 7.5. Measured on a GeForce GTX TITAN X (sm_52, 12 GB, driver 580.178.04, CUDA 12.9): the engine loads, serves and answers, and the Gyro-S profile in bench/results/2026-10-21-gyro-titanx is that card. Validation: tools/test_setup_choices.py (50 tests), tools/test_setup_older_gpus.py (19) and tools/test_setup_configs.py pass; cmake --build build-cuda12 with -DCMAKE_CUDA_ARCHITECTURES=50 -DSTRATA_EXPERIMENTAL_SM60=ON is clean. The HIP, SYCL and Metal engines are untouched by this change and were not built here (no ROCm, no Intel toolchain on this PC).
Author
|
Closing this one as well: the Maxwell (sm_50) support is not wanted upstream right now. |
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The problem
CUDA 13 dropped Maxwell, so a Maxwell card needs the experimental CUDA 12 engine. That engine refused every
card below compute capability 6.0, although the kernels compile for Maxwell and a GTX TITAN X (sm_52) is a
12 GB card with enough memory for a small model. The gate was the only thing in the way.
What this changes
CMakeLists.txt: the experimental build's lowest architecture is 5.0 (it was 6.0). The release CUDA 13engine still needs 7.5.
setup.py: setup's card choice admits the Maxwell cards (sm_50/52) for the CUDA 12 engine, and stillrefuses Kepler (sm_35), for which no engine code exists.
src/core/device.cu: the engine's own check says "5.0 or newer" in the experimental build.tools/test_setup_choices.py,tools/test_setup_older_gpus.py: the tests follow (50 and 19 tests).Nothing else changes: the release engine, the HIP engine, the SYCL engine and the Metal engine are untouched.
Validation
Environment: GeForce GTX TITAN X (sm_52, 12,288 MiB), driver 580.178.04, Linux 6.2 (CachyOS), CUDA 12.9 at
/opt/cuda, host compiler g++-14. Engine built with
-DCMAKE_CUDA_ARCHITECTURES=50 -DSTRATA_EXPERIMENTAL_SM60=ON(thebuild-cuda12folder).Passed
python tools/test_setup_choices.py- 50 tests, OK.python tools/test_setup_older_gpus.py- 19 tests, OK.python tools/test_setup_configs.py- OK.cmake --build build-cuda12 -j 8- clean, no warnings from the changed files.bench/results/2026-10-21-gyro-titanx, a real request, 201 tokens decoded, 77 s).Failed
Not run
Intel toolchain. The CUDA 13 release engine was not built here either (only CUDA 12.9 is installed).
The next pull request
The profile of this card is in the next PR: the GPU's kernels under Nsight Systems, the bottleneck analysis
and the one change that measured faster there (the dense projections' FP16 GEMMs as
cublasSgemmon FP32copies, opt-in
STRATA_DENSE_F32=1, 1.18-1.65x for the big shapes, bitwise on 5,201 chunk sizes).