home/categories/framework-internals/xpu-forces-mojo-opset-mojo-opset-backends-ttx-kernels-npu-triton-npu-kernel-opt-skill-md
framework-internalsdevelopment

triton-npu-kernel-opt

Triton kernel optimization guide for Ascend NPU (910B/910C). Use when writing or optimizing Triton kernels targeting NPU backend, including GEMM, attention, normalization, or any compute-intensive kernel. Covers hardware constraints, proven optimization patterns, compiler flags, tile tuning, and known pitfalls specific to the ttx backend with Ascend SoCs.

XPU-Forces
maintainer
XPU-Forces
Updated 3/31/2026
Stars
17
Forks
28
quick start

Installation and usage

Triton kernel optimization guide for Ascend NPU (910B/910C). Use when writing or optimizing Triton kernels targeting NPU backend, including GEMM, attention, normalization, or any compute-intensive kernel. Covers hardware constraints, proven optimization patterns, compiler flags, tile tuning, and known pitfalls specific to the ttx backend with Ascend SoCs.

Installation
$ install --globalskills.sh
Usage

Once installed, you can use this skill by running the following command in your terminal:

skills use triton-npu-kernel-opt