150 LoC CUDA I8 Matmul That Beats CuBLAS Tensor Core FP16github.com1 point·carsonpoole··0 commentsOpen articleSaveView on HN