A publication for the faculty, staff, administrators, and friends of California State University, Chico
February 11, 2011 Volume 41 / Number 4

Chris Mauzey, a recent graduate of the CSU, Chico Computer Science Department, installs a Graphical Processing Unit (GPU) in a new computer in the Department of Physics as part of a National Science Foundation research project.
Photo: Shawn Mayor

Parallel Processing in the Department of Physics

Despite transistors becoming smaller and more numerous on chips, there is general agreement among computer scientists today that microprocessor clockspeed won’t become much faster than at present. This explains the recent trend for multi-core processors on common personal computers.  Today’s typical PC now has at least two, and sometimes as many as six cores. The fastest clockspeeds are around 3.5 GHz.

Recognizing this trend in computer architecture, Dr. Shane Mayor, a Research Professor in the Department of Physics recently invested in a “personal supercomputer” in order to speed the calculations necessary for his research project that is funded by the National Science Foundation.  The machine is only slightly larger than the average desktop computer. However, two NVIDIA Tesla c2050 Graphical Processing Units (GPUs) inside make it extremely powerful. Each GPU provides 448 processors operating at 1.15 GHz. NVIDIA donated a third Tesla c2070 card as part of an academic partnership program bringing the total number of processors inside up to 1344.

GPUs have evolved in response to the demand for high performance from video games.  Typical video cards today have dozens to hundreds of processors. Within the last few years, NVIDIA began marketing GPU cards specifically designed for scientific computing. The cards consume far less electrical power than computer clusters and are much less expensive. However, in order to take advantage of the architecture and make programs run faster, programmers must know how to distribute computations among multiple processors — a science known as parallel programming.

Enter Chris Mauzey, a 2010 graduate of the Computer Science Department.  During the last few years, Mauzey (from Willows, CA) learned how to write programs to tap the multi-core power of GPUs. Mauzey currently works part-time in Mayor’s lab as a research assistant and is optimizing the programs that Mayor has written. Mauzey reduced the execution time of the programs by factors of 10 and 100 even before the new workstation arrived by implementing instructions to take advantage of the architecture of common multi-core processors. He plans to begin graduate studies this year.

Mayor and Mauzey’s goal is to accelerate the calculations so that they can be completed in real-time. That means computing hundreds of thousands of motion vectors from image sequences in a handful of seconds. The result will enable the Raman-shifted Eye-safe Aerosol Lidar (REAL) to provide immediate remote wind information. The technology is likely to be useful in applications such as wind energy, wind shear detection, atmospheric research, and improving short-term predictions of the dispersion of particulate matter.