Bill Dally is NVIDIA’s chief scientist and senior vice president of research. A computer architect and former Stanford computer-science chair, he has worked on interconnection networks, experimental parallel computers and efficient processing architectures, while leading NVIDIA Research since 2009.[1][2]
Start with Bill Dally Caltech alumni interview →
Education
Links
NVIDIA Research StanfordChief Scientist · NVIDIA
Senior Vice President of Research · NVIDIA
Adjunct Professor, Emeritus · Stanford University
Career
Dally earned an electrical-engineering BS at Virginia Tech and MS at Stanford, followed by a computer-science PhD at Caltech. He taught at MIT from 1986 to 1997, then Stanford, where he chaired computer science during 2005–2009. He joined NVIDIA as chief scientist in January 2009. Stanford now lists him as an adjunct professor, emeritus.[1][3][6]
Contributions
Reduce the cost of communicating between processors
Dally’s early work examined how processors in a parallel computer exchange messages, including network topology, routing and flow control. His Caltech work produced a torus routing chip. At MIT, his team built the J-Machine and M-Machine, exploring low-overhead communication and synchronization. These systems made moving and coordinating work between processors a central architecture problem.[1][4]
Build and evaluate an inference machine from chiplets
Dally coauthored Simba with Yakun Sophia Shao and a larger NVIDIA, Stanford and MIT team. The work fabricated and tested a 36-chiplet prototype for deep-learning inference. Its architecture distributes computation and storage across smaller dies, while mapping and tiling strategies reduce costly communication between them. The paper measures the resulting performance and energy tradeoffs rather than treating packaging alone as a speedup.[5]
Ideas
NVIDIA’s account of its research organization describes Dally’s emphasis on risky technical questions that could materially benefit the company. Researchers collaborate with product teams to discover which ideas survive implementation. The account traces contributions from many teams across graphics, architecture and AI; Dally’s role is research leadership, alongside his own technical work.[2]
Tools, writing & talks
Writing, research & talks
Start here
Bill Dally Caltech alumni interviewengenuity.caltech.eduAn accessible first-person account of interconnection networks, stream processing and the J-Machine.
Start with his explanation of how computers communicate and the J-Machine example. The interview supplies historical context without requiring circuit-design experience; continue to Simba for a technical treatment of the same communication costs in modern inference hardware.
Interview participant[4]A fabricated chiplet-based inference system and evaluation of mapping, communication and energy costs.
Coauthor[5]The company’s account of research-to-product collaboration and contributions from its research groups.
Research leader and quoted participant[2]