Phobos - Tiny scale-free kernel language for NVIDIA GPUs

Show HN: Phobos – A tiny scale-free kernel language with tile-DAG support

Phobos is a lightweight, scale-free kernel language inspired by Triton, designed to compile directly to PTX for execution on NVIDIA GPUs. Built with Rust and leveraging MLIR, it features first-class tensor types and native support for distributed tile-DAG designs. While developed as a personal research project to explore low-level GPU concepts, Phobos achieves impressive performance, reaching 76% of cuBLAS SGEMM GFLOP/s on an RTX 2080 SUPER. It offers developers a streamlined path to write optimized AI kernels without the complexity of traditional CUDA development.

"I strongly believe, in order to fully understand something, you really need to do the work and get your hands dirty."