KernelForge — Pure Rust GPU kernel template project

Write GPU compute kernels in stable Rust — no nightly or C++ required

Problem

GPU compute kernels are typically written in CUDA C++. Rust attempts often rely on nightly features, lacking stability and toolchain support. Developers want Rust's engineering advantages for kernels but lack a usable scaffold.

Solution

A GPU kernel development template built on stable Rust, integrating compilation, testing, and benchmarking workflows, supporting SPIR-V or PTX output, with example kernels and performance comparisons. Let developers focus on kernel logic, not toolchain setup.

Users

Systems and performance engineers interested in GPU compute, familiar with Rust but avoiding C++

MVP

Provide compilable example kernels; integrate cargo commands for one-click build and run; support at least one backend (e.g., Vulkan or CUDA); include simple benchmarks; document limitations and performance expectations

Revenue

Open-source project via GitHub Sponsors, or paid enterprise support and custom consulting

Why now

Rust continues to rise in systems programming, GPU compute demand expands from ML to general computing; stable Rust toolchains mature enough to make pure Rust GPU kernels feasible