5.5x more CUDA performance with 5 minutes of work#2140

JohannesGaessler merged 1 commit intoggml-org:masterggml-org/llama.cpp:masterfrom

JohannesGaessler:cuda-mmvq-pascalJohannesGaessler/llama.cpp:cuda-mmvq-pascalCopy head branch name to clipboard