Basser Seminar Series

High-Performance GPU Computing with CUDA and the NVIDIA "Fermi" Architecture

Speaker: Dr Mark Harris, NVIDIA

Time: Thursday 12 November 2009, 2:00-3:00pm **Note different day/time.

Location: The University of Sydney, School of IT Building, Lecture Theatre (Room 123), Level 1

Add seminar to my diary

Abstract

Modern GPUs provide a level of massively parallel computation that was once the preserve of specialized supercomputers. NVIDIA's latest GPUs are fully programmable, massively multithreaded processors with hundreds of scalar processor cores capable of delivering hundreds of billions of operations per second. The NVIDIA CUDA architecture provides a parallel programming model that enables developers to program GPUs in C, C++, and Fortran, as well as specialized GPU Computing languages such as OpenCL and Microsoft DirectCompute. Researchers across many scientific and engineering disciplines are using this platform to accelerate important computations by up to 2 orders of magnitude.

This talk will provide an overview of GPU Computing using the NVIDIA CUDA programming model and architecture and the applications and research that it is enabling. The talk will also provide a taste of the upcoming NVIDIA GPU architecture, codename "Fermi". NVIDIA's "Fermi" flagship GPU has 512 processing cores and 8x the peak double-precision floating point throughput of its predecessor used in the NVIDIA Tesla C1060 and S1070 processors, as well as Error Correcting (ECC) memory support. Fermi has a new cache hierarchy and up to 3x as much on-chip shared memory as its predecessor, and supports enhanced software features including capability to execute C++ code. These new features represent a huge step forward in GPU computing that will enable high performance computing on the desktop as well as in the next generation of data centers and clusters.

Speaker's biography

Mark Harris is a Senior Developer Technology Engineer at NVIDIA, where he works with developers around the world on software for computer graphics and high-performance computing. His research interests include parallel computing, general-purpose computation on GPUs, physically based simulation, and real-time rendering. He earned his PhD in computer science from the University of North Carolina at Chapel Hill in 2003 and his BS from the University of Notre Dame in 1998. He founded and maintains GPGPU.org, a web site dedicated to general-purpose computation on GPUs. Mark has recently moved to Brisbane after living in the United Kingdom for five years.