Bio
Abdul Dakkak is Chief Scientist at Modular, where he works on AI compute, GenAI performance, Mojo, kernels, framework and serving layers for Modular's platform.
Session (1)
Additional Speaker Details
Using publicly available information we constructed an analysis to help you get a feel for this speaker before deciding to attend their session.
Abdul Dakkak is a Distinguished Engineer at Modular leading the GenAI team, where he works on deep GPU kernel optimization in Mojo — co-authoring Modular's engineering deep-dives on high-performance Blackwell matmul kernels (TMA, tensor cores, shared-memory swizzling) and rapid hardware bring-up on AMD's MI355 — as well as Mojo's GPU compilation infrastructure. His session offers systems-level insight into how next-generation AI compute is built, from an engineer doing the work.
Speaking style
Recent talks (3)
GitHub
@abduld- abduld/libwb★ 87C++ utility library for CUDA programming, used in Coursera's Heterogeneous Parallel Programming course (archived)
- abduld/Parboil★ 34Collection of computation-intensive GPU/CPU benchmark algorithms from the University of Illinois, using CUDA and OpenCL
- abduld/WebGPU★ 21GPU computing resources related to WebGPU (Mathematica)
Recent writing (3)
- Matrix Multiplication on Blackwell: Part 2 - Using Hardware Features to Optimize Matmul · blog · Sep 2025
- Matrix Multiplication on Blackwell: Part 3 - The Optimizations Behind 85% of SOTA Performance · blog · Sep 2025
- Achieving State-of-the-Art Performance on AMD MI355 — in Just 14 Days · blog · Oct 2025