Add ROCm CDNA5 (gfx1250) support - #2023
Merged
matthewdouglas merged 1 commit intoJul 28, 2026
Merged
Conversation
- Include gfx1250 in the ROCm 7.14+ build architecture list. - Add IS_CDNA5 macro (gfx1250/gfx1251). CDNA5 runs in Wave32 mode and is kept out of IS_CDNA (Wave64) so the default warp size stays 32. - Route CDNA5 into the fp32-FMA 4-bit GEMM path (IS_CDNA || IS_CDNA5) rather than the RDNA v_dot2 path, which CDNA5 does not implement. - Route gfx125x through the existing gfx12 SIMT-vs-dequant crossover (M<=8); dedicated calibration left as follow-up. - Document gfx1250 in README and installation docs.
zjin-lcf
force-pushed
the
feat/gfx1250-cdna5-support
branch
from
July 27, 2026 22:40
8fc8337 to
092e9bf
Compare
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
matthewdouglas
approved these changes
Jul 28, 2026
Member
|
LGTM, thanks! |
matthewdouglas
merged commit Jul 28, 2026
4e007ec
into
bitsandbytes-foundation:main
86 checks passed
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Adds ROCm CDNA5 (gfx1250) support to bitsandbytes.
gfx1250in the ROCm 7.14+ build architecture list (.github/scripts/build-rocm.sh).IS_CDNA5macro (gfx1250/gfx1251) incsrc/common.cuh. CDNA5 runs in Wave32 mode and is deliberately kept out ofIS_CDNA(Wave64), so the default warp size stays 32 (unchanged from upstream).IS_CDNA || IS_CDNA5) rather than the RDNAv_dot2path, which CDNA5 does not implement (nodot12-insts/v_dot2_f32_*builtins — they fail to compile for gfx1250).gfx125xthrough the existing gfx12 SIMT-vs-dequant crossover (M<=8) inbitsandbytes/backends/cuda/ops.py; dedicated calibration is left as a follow-up.gfx1250inREADME.mdanddocs/source/installation.mdx.Rationale
Despite its
gfx125xtarget number, CDNA5 behaves like CDNA (Wave32 but no SIMT dot instructions), so it needs the fp32-FMA math path for correct/accurate bf16/fp16 4-bit GEMM rather than the RDNAv_dot2path.Test plan
gfx1250in the arch list.