---
title: "Triton Programming Model"
description: "Triton, the Python GPU kernel language: one program per tile, the kernel and its launch grid, compiling on first call, and where such kernels run."
canonical_url: "https://fanout.sh/inference-eng/curriculum/13-02"
md_url: "https://fanout.sh/inference-eng/curriculum/13-02.md"
access: "pro-preview"
---

# Triton Programming Model

Triton, the Python GPU kernel language: one program per tile, the kernel and its launch grid, compiling on first call, and where such kernels run.

- Access: Fanout Pro (public preview)

## Public preview

The opening page from the Triton and kernel engineering module ended with kernels described and none written: work grouped into grids, blocks, and warps of thirty-two, and a promise that the rest of the module writes them. The GPU threads and warps lesson built that picture from NVIDIA's documents; this page opens the tool the writing happens in. Triton is…

## In this lesson

- One program per tile
- The kernel itself
- The wrapper and the grid
- Compilation on the first call
- Where these kernels already run

## Continue learning

Pro unlocks the complete lesson, video, practice, and code where available.

---
This representation contains public Fanout content only. Protected Pro lessons, account data, billing, checkout, and pricing are not included.

Browse the public content map: https://fanout.sh/sitemap.md