Rust Threads on the GPU

Article

• Vectorware maps Rust std::thread semantics onto GPU warps for GPU execution
• Aims to let Rust developers write GPU code without learning CUDA/WGSL kernel models
• Each spawned closure maps to one warp, preventing lane divergence by construction

Discussion

• Main critique: mapping threads to warps makes GPU act like a slow CPU — defeats the point
• No public repo found; several commenters asked where to try it
• Concern that GPU-unaware code patterns won’t benefit from parallelism without full redesign
• Suggested pairing with Burn/CubeCL for a more complete Rust GPU stack

Discuss on HN