Tech Talk: Nvidia: GPU Architectures from Shield to Supercomputers

Information Session: External Relations Group | January 23 | 5:30-6:30 p.m. | Soda Hall, Wozniak Lounge

 Electrical Engineering and Computer Sciences (EECS), Nvidia

Given By: Dr. Craig M. Wittenbrink, Director, 3D Graphics Architecture
Craig M. Wittenbrink is an Architecture Director, 3D Computer Graphics at NVIDIA. He has worked in computer architecture, scientific visualization, and Graphics Processing Unit (GPU) architecture, and has over 70 publications and patents. At NVIDIA, Craig has developed new performance modeling strategies, new display verification processes, and memory system architectures. He managed the research and development of the first Kepler core for Tegra—GK20A--, and also managed the SOC (system on a chip) architectural development for the Tegra K1 (Logan) SOC. He received his B.S. EE/CS from Univ. of Colorado and M.S/Ph.D. EE from the Univ. of Washington. He’s a member of ACM SIGGRAPH, a senior member of IEEE, and a member of IEEE Computer Society

NVIDIA recently announced the Tegra K1 SOC (system on a chip), and now has a unified GPU architecture from mobile to high performance computing. I will provide an introduction to the advancements in power efficiency that scale a single architecture from low power (< 2 Watts) to server processors (250 watts). We use different process technologies and power efficient design techniques. I will cover applications that require significant computation for handheld game players (NVIDIA Shield) to HPC servers (NVIDIA Tesla). And, I will present power efficient design examples in our SM (Streaming Multiprocessor), the engine of the GPU.