Skip to content

Commit adc8bac

Browse files
committed
vulkan: disable q8_0 mmvq on Nvidia
1 parent ab3ab2a commit adc8bac

File tree

1 file changed

+1
-1
lines changed

1 file changed

+1
-1
lines changed

ggml/src/ggml-vulkan/ggml-vulkan.cpp

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -5958,7 +5958,7 @@ static bool ggml_vk_should_use_mmvq(const vk_device& device, uint32_t m, uint32_
59585958
case VK_VENDOR_ID_NVIDIA:
59595959
switch (src0_type) {
59605960
case GGML_TYPE_Q8_0:
5961-
return k >= 3072 && m < k;
5961+
return false;
59625962
default:
59635963
return true;
59645964
}

0 commit comments

Comments
 (0)