NVIDIA CUDA Tile

1 min read Original article ↗

NVIDIA® CUDA® Tile is a tile-based GPU programming model that targets portability for NVIDIA Tensor Cores. CUDA Tile unlocks peak GPU performance with a programming model that simplifies the creation of optimized, tile-based kernels across NVIDIA platforms.


Get Started

CUDA Tile is based on the Tile IR specification and tools, including cuTile, which is the user-facing language support for CUDA Tile IR (Intermediate Representation) in Python (and, in the future, C++). The NVIDIA Python implementation of this tile-based programming model is cuTile Python.

CUDA Tile IR

Virtual Instruction Set for Tile Programming:

  • Enables native programming of GPUs within the structured high-performance context of the tile programming model

Get Started With CUDA Tile IR

cuTile Python

Python-Native, Tiled Kernel Development:

  • Seamless high-level Python expression of the CUDA Tile programming model

  • Built on the foundation of the Tile IR specification

  • Offers the ability to write, define, and optimize tiled GPU kernels using familiar Python syntax

Get Started With cuTile Python 


Learning Library


More Resources

Get started with CUDA Tile today.

Explore the Quick Start Guide