NVIDIA / cutlass

CUDA Templates and Python DSLs for High-Performance Linear Algebra

NVIDIA / nccl

Optimized primitives for collective multi-GPU communication