Significant improvement in prompt processing speed in Vulcan after updating to 1.101.1

[IntensivePorpoises]

I'm using a Radeon 6700XT on Win11, and previously: 12b models at more than q5KS were just unbearably slow. Now I can get whole responses from q6K quants in well under a minute. I don't know what you guys changed, but it really helped a lot, thanks!

[LostRuins]

Mostly due to upstream improvements in llama.cpp vulkan