Making GPUs Feel Native in Rust
Talkat RustConf 2026 · on Thursday, 2026-09-10· at 11:00· Room 511 | Palais des Congrès de Montréal, Palais des Congrès de Montréal·
summary
Claims a Rust-first design that preserves ownership, borrowing and lifetimes on GPUs by treating std as a stable facade and falling back to the CPU via host calls for missing OS features. Implements threading by having non-main warps spin while a warp-zero main controls spawn/join via flags so Rust thread APIs work unchanged; Tokio runs on the GPU and some Rust patterns hit the theoretical “speed of light” performance equal to handwritten CUDA. Trade-offs: extra power from spinning warps, compiler optimizations still immature, and concurrency limited by hardware warps.
AI-written summary · about
chapters (11)
- 0:00Introduction to GPUs in Rust
- 3:14GPU programming challenges today
- 7:27Rust-first GPU design approach
- 10:00Host calls and OS abstraction
- 13:16Concurrency model mismatch
- 21:48Solving GPU concurrency issues
- 25:18Performance and Rust semantics
- 27:26Ecosystem benefits of Rust GPUs
- 31:18Cross-pollinating GPU and Rust
- 32:44GPUs as future necessity
- 34:09Q&A session
related talks
28mInterview
39mLevel Up Your Concurrency Skills With Rust
25m"Rust for AI & Accelerated Computing"
59mParallel Programming in Rust: Techniques for Blazing Speed
40mRust’s approach of getting things right
40mWhy Rust?
19mRust Without Tears: A Guide to Avoiding Common Traps
01h 14mRust: A Friendly Introduction