Speaker: Melih Elibol & Stephen Jones
Location: Soda Hall, 510
Date: Friday,Feb 13th, 2026
Time: 12 – 1 pm PT
Title: From CUDA to Rust: Scaling GPU Performance with Tile-Based Programming
Abstract:
As GPU architectures boost performance with specialized hardware like Tensor Cores, tile-based programming has emerged as a vital strategy to simplify development by shifting the focus from scalar processing with individual threads to collective operations on high-level data “tiles.” This seminar is presented in two parts: first, we will introduce the tile programming model, detailing how Tile IR and cuTile make GPU programming easier by targeting block-level parallelism. Second, we will do a deep dive into our work on GPU programming in Rust. We’ll present our new tile-based Domain-Specific Language (DSL) built atop the Tile IR compiler stack, which we’ve embedded directly in Rust. We also go over our approach to high performance programming on multiple GPUs, enabling Rust-based tools, such as inference engines, to efficiently interface with our GPUs.
Speaker Bio:
Melih Elibol is a Senior Research Scientist in Programming Systems and Applications research at NVIDIA. His research aims to make GPU applications easier to write and scale using modern programming languages, numerical optimization, and machine learning. He completed his Ph.D. at the University of California, Berkeley and A.L.B. at Harvard University.
Stephen Jones is a Distinguished Software Architect of CUDA, working on defining the language, the platform, and the hardware that it runs on, to span the needs of parallel programming from high performance computing to artificial intelligence. Prior to his present position, he led the Simulation & Analytics group at SpaceX, working on large-scale simulation of rocket engines. He has worked in diverse other industries, including networking, CAD/CAM, and scientific computing. He has been a part of CUDA since 2008.