framework

cuTile Python

Python front end for writing tile-based CUDA kernels, which JIT-compiles and specializes a kernel on the constants and types it sees at call time.

Known aliases

CUDA Tile Python

Relationships

No evidence-backed relationships are recorded.

Current clusters

build1 publisher

Diffing Tile IR catches the kernel ports that pass tests and are still wrong

NVIDIA put all 24 public TileGym operators through an agent pipeline into cuTile Rust at 99.5% of cuTile Python performance on average, with a Tile IR diff deciding at each stage whether a conversion advanced.

Publishers:developer.nvidia.com

Reality

Evidence42
Adoption22
Hype gap+20
Incentives80
Confidence52