Glossary · Term

Triton

← all terms

Definition

Plain language

A programming language from OpenAI for writing fast GPU code more easily than raw CUDA.

As stated in the literature

A Python-like language and compiler for authoring GPU kernels at a higher level than CUDA; the language in which the expert-hand-tuned VeOmni baseline kernel was written before KernelPro beat it with from-scratch CUDA.

Why it matters: It lowers the barrier to writing high-performance GPU code, letting more developers tune kernels without mastering the most intricate hardware details.

For example, an engineer can write a fast GPU kernel in a Python-like style with Triton instead of wrestling with lower-level CUDA.

Heard on the show

“There's a kernel inside a real training system called VeOmni — a weight-gradient kernel for a mixture-of-experts model — that had been hand-tuned in Triton by expert engineers.”
Episode 177 — Why Raw Profiler Data Made an AI Worse at Writing GPU Code

Mentioned in 1 episode

  1. 177
    Why Raw Profiler Data Made an AI Worse at Writing GPU Code

Related terms