NVIDIA / cutlass
git clone https://github.com/NVIDIA/cutlass.git
v4.8.0 · 1mo ago
🔥 on GitHub Trending today
CUDA C++ templates and Python DSLs for building high-performance GEMM and other linear algebra kernels on NVIDIA GPUs.
Why you might care
Provides reusable, tunable building blocks for tensor-core matrix multiply across many data types, including FP8, FP4 and block-scaled formats, on Volta through Blackwell. CuTe DSL offers a Python interface to write kernels without deep C++ template expertise, with much faster compile times.
+6
stars, 24h · #5,417
+~33
stars, 7 days · #5,779
+~140
stars, 30 days · #5,382
11k
★ total · +0% this week
2.1k
forks · 790 open issues
Star history (as recorded here)
LanguageC++
License—
Created2017-11-30 (8y ago)
Last push2d ago
Watchers121
OwnerOrganization
Seen on GitHub Trending
- 2026-10-23 — #9 daily in c++ · +4 ★
GitHub topics
Links
- GitHub: github.com/NVIDIA/cutlass
- Homepage: https://docs.nvidia.com/cutlass/index.html
- Releases · Issues · Hacker News mentions
README
Loading README…
Blurb generated by claude-haiku-5-5 today. Star deltas marked ~ are extrapolated from partial history.