cuTile is a programming model for writing parallel kernels for NVIDIA GPUs
-
Updated
Sep 24, 2026 - Python
cuTile is a programming model for writing parallel kernels for NVIDIA GPUs
Self-hosted functional language with explicit effects. Compiles to JVM bytecode, C, and NVIDIA GPU kernels via cuTile.
Rust + cuTile research prototype for paged latent-cache LLM decode attention, validated on an RTX 4060.
A complete GPT built entirely on NVIDIA cuTile: declarative GPU kernels in ~20 lines instead of ~150 hand-tuned CUDA, matching PyTorch speed in a ~10MB footprint
Machine Learning Accelerators
Weighted Jacobi stencil in Rust with cuTile CUDA
CPU virtual machine for CUDA Tile IR bytecode - run cuTile kernels without a GPU
Microsecond-scale limit order book inference with custom CUDA/cuTile kernels, TensorRT Plugin V3, CUDA Graphs, and p99 latency benchmarking.
Accelerate LLM inference with TurboQuant KV cache compression on NVIDIA cuTile, using custom GPU kernels for 5x smaller caches and unbiased attention
Experimental named-axis Rust ML research framework on NVIDIA cuTile, built around executable experiment contracts.
To associate your repository with the cutile topic, visit your repo's landing page and select "manage topics."