Commit c7518dbec for llama.cpp
commit c7518dbec59281c430ca5278ab4e60eadbc88b13
Author: Beinsezii <39478211+Beinsezii@users.noreply.github.com>
Date: Mon Nov 16 08:56:20 2026 -0800
MMVQ: Tune ne11 Gate for RDNA3_5 (#32308)
diff --git a/ggml/src/ggml-cuda/mmvq.cu b/ggml/src/ggml-cuda/mmvq.cu
index c6d895db4..9118fbd55 100644
--- a/ggml/src/ggml-cuda/mmvq.cu
+++ b/ggml/src/ggml-cuda/mmvq.cu
@@ -378,6 +378,14 @@ bool ggml_cuda_should_use_mmvq(enum ggml_type type, int cc, int64_t ne11) {
return ne11 <= MMVQ_MAX_BATCH_SIZE;
}
}
+ if (GGML_CUDA_CC_IS_RDNA3_5(cc)) {
+ switch (type) { // tuned on Radeon 8060S (gfx1151)
+ case GGML_TYPE_Q8_0:
+ return ne11 <= 6;
+ default:
+ return ne11 <= MMVQ_MAX_BATCH_SIZE;
+ }
+ }
if (GGML_CUDA_CC_IS_CDNA(cc)) {
if (GGML_CUDA_CC_IS_CDNA1(cc)) {
switch (type) {