CUTLASS
CUTLASS is NVIDIA’s collection of building blocks and examples for constructing high-performance matrix and related GPU kernels.
How it works
It exposes choices such as tiling, memory movement, matrix instructions and epilogues. This makes specialized fusion possible but also places responsibility on the author to choose supported layouts and resource budgets. A custom configuration still needs a baseline and correctness tests.
Work through an example
A GEMM followed by a custom bias and activation may benefit from an epilogue that writes the final result once. The benefit depends on removed traffic and the effect on the main kernel.
Where the term can mislead
CUTLASS is not a single universal kernel that beats cuBLAS for every shape. Compare the exact operation and hardware.
Connect the idea
Take it into practice
Continue with the existing concept, apply it to an interview problem, or follow a course lesson. Access requirements are shown before you open a page.
New to GPUs? Start with the free execution-model concept.
Sources and scope
Examples here are teaching calculations, not measured GPU performance. Consult the linked documentation for the exact device, instruction and tool version. Source review: 25 October 2026.