ggml-cuda: enable RDNA3.5 MMQ tile override#20
Merged
Conversation
DFlash verify batches are small (ne[1] ~= ddtree_budget+1, ~23 at the default budget=22); the stock 128x128 / 8-warp MMQ tile under-occupies on consumer RDNA. A 48x64 / 4-warp tile is faster there, output bit-identical. Qwen3.6-27B Q4_K_M, --ddtree-budget=22: gfx1100 (RX 7900 XTX): 56.78 -> 60.18 tok/s (+6.0%) gfx1201 (R9700): 54.65 -> 59.37 tok/s (+8.3%) Gated on RDNA3_0 + RDNA4; gfx1151/RDNA3.5 left on the default (not benchmarked). Opt out with -DLUCEBOX_RDNA_MMQ_TILE_OVERRIDE=0. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Depends on #18.
This is stacked on #18. Until #18 merges, GitHub will show the #18 commit plus
the follow-up commit in this PR; after #18 lands, this should reduce to the
single RDNA3.5 gate change.
This follow-up extends the RDNA MMQ tile override to RDNA3.5 / gfx1151
(Strix Halo). The existing PR deliberately kept gfx1151 on the default tile;
local testing on lucebox2 showed the 48x64 / 4-warp tile is also beneficial
there.
Measured on lucebox2, Qwen3.6-27B Q4_K_M, DFlash draft,
--ddtree-budget=22:For the 256-token run, the speculative decode shape stayed identical:
accepted=138/1040,target_forwards=119, accept rate 13.3%, so the gain isnot from different acceptance behavior.
Validation:
dflash_serveron lucebox2 withCMAKE_HIP_ARCHITECTURES=gfx1151.