Ajay Tirumala

Ajay was the architecture lead for several generations of NVIDIA products, most notably the NVIDIA Volta independent thread progress, the NVIDIA Turing data uniformity, reduction features in the GPU, and most recently DPX in NVIDIA Hopper. He has been managing the Streaming Multiprocessor (SM) Core Architecture team for the last few years, responsible for the architecture of SM math (tensor cores, FP and integer datapaths), and control units (branching and instruction pipelines). The unified SM Architecture forms a critical part of both data center (for example, A100 through H100) and consumer (for example, NVIDIA Ampere RTX 3090 through NVIDIA Ada RTX 4090) product lines, accelerating a wide range of workloads from deep learning to cutting-edge graphics. He was also previously part of NVIDIA Research. Ajay obtained his Ph.D. in computer science in UIUC in 2006 and has been at NVIDIA ever since.
Avatar photo

Posts by Ajay Tirumala

Technical Walkthrough 5

Boosting Dynamic Programming Performance Using NVIDIA Hopper GPU DPX Instructions

Dynamic programming (DP) is a well-known algorithmic technique and a mathematical optimization that has been used for several decades to solve groundbreaking... 9 MIN READ