CUDA: adding switch points per HW and quant type to tune the mvq->MMQ decode crossover - #26079

Merged

ORippler merged 7 commits into

ggml-org:masterfrom

praneshgo:pgonegandla/mmvq-mmq-crossover

Aug 20, 2026

Commits on Jul 29, 2026

CUDA: runtime GGML_CUDA_MMVQ_MAX to tune the mvq->MMQ decode crossover
praneshgo
committed
ff05098
Added Blackwell specific switch point, to reduce dependence on runtime env var.
praneshgo
committed
f2a3efa

Commits on Jul 30, 2026

Add per-HW switch point values for DGX Spark and removing runtime env var
praneshgo
committed
f2aace7

Commits on Jul 31, 2026

Adding switch points for Ada, tested on RTX 4090
praneshgo
committed
db1508d

Commits on Aug 3, 2026

Modifying DGX Spark numbers based on latest run and adding some comments and small functional changes relating to MoE
praneshgo
committed
9052a5c

Commits on Aug 19, 2026

Reverting an unnecessary conditional
praneshgo
committed
614d524
Update ggml/src/ggml-cuda/mmvq.cu
ORippler
authored
41a70d7