CUTLASS: CUDA Templates and Python DSLs for High-Performance Linear Algebra

Fields

id

sources/src5-cutlass-github

URL

https://github.com/NVIDIA/cutlass

Publisher

NVIDIA

Kind

repo

Published

2017-01-01

Accessed

2026-10-04

notes

CUTLASS 4.8.0 (Sept 2026). Collection of abstractions for high-performance GEMM and related computations. Supports FP64, FP32, TF32, FP16, BF16, FP8 (e5m2/e4m3), block-scaled (NVFP4, MXFP4/6/8), narrow integer (4b, 8b), binary (1b). Covers Volta through Blackwell. v4 adds CuTe DSL (Python DSL). 10.5k stars. Essential for custom kernel development.

Notes

CUTLASS 4.8.0 (Sept 2026). Collection of abstractions for high-performance GEMM and related computations. Supports FP64, FP32, TF32, FP16, BF16, FP8 (e5m2/e4m3), block-scaled (NVFP4, MXFP4/6/8), narrow integer (4b, 8b), binary (1b). Covers Volta through Blackwell. v4 adds CuTe DSL (Python DSL). 10.5k stars. Essential for custom kernel development.

Cross-references

This record references no other records.