llama.cpp
vulkan: use 4 rows for NVIDIA MUL_MAT_ID MMVQ except pre-Turing
#29274

Merged

vulkan: use 4 rows for NVIDIA MUL_MAT_ID MMVQ except pre-Turing #29274

ggerganov merged 1 commit into ggml-org:master from SG-Amadeus:sg-amadeus/vulkan-mmid-mmvq-rows4

SG-Amadeus requested a review 48 days ago

github-actions added Vulkan

github-actions added ggml

0cc4m commented on 2026-10-05

SG-Amadeus changed the title ~~vulkan: make MMVQ row policies device-specific and tune NVIDIA MUL_MAT_ID~~ vulkan: use 4 rows for NVIDIA MUL_MAT_ID MMVQ except pre-Turing 33 days ago

vulkan: use 4 rows for NVIDIA MUL_MAT_ID MMVQ

e13c77ad

SG-Amadeus force pushed to e13c77ad 33 days ago

0cc4m approved these changes on 2026-10-08

0cc4m added merge ready

ggerganov merged 404f557b into master 30 days ago

Reviewers

0cc4m

Assignees

No one assigned

Labels

Vulkan ggml merge ready

Milestone

No milestone

llama.cpp vulkan: use 4 rows for NVIDIA MUL_MAT_ID MMVQ except pre-Turing #29274 Merged

vulkan: use 4 rows for NVIDIA MUL_MAT_ID MMVQ except pre-Turing #29274

llama.cpp
vulkan: use 4 rows for NVIDIA MUL_MAT_ID MMVQ except pre-Turing
#29274

Merged