NVIDIA releases a preview of CUTLASS
Released a preview of CUTLASS — a CUDA C++ template library for GEMM
Why it matters
Broke kernels down into composable parts. FlashAttention, Transformer Engine and many modern attention kernels are built on top of it
Source — Primary sources first; secondary reports are labelled “Press”
Officialdeveloper.nvidia.comhttps://developer.nvidia.com/blog/cutlass-linear-algebra-cuda/Added in Week 35, 2026 · See that week