Students gain skills at first Virtual School summer school

released 12.02.08

Summer school attendees

GPU cluster project

NCSA's Innovative Systems Laboratory maintains a 16‑node cluster that combines both GPU (graphics processing units) and FPGA (field-programmable gate array) technology in order to explore the potential of these novel architectures to accelerate scientific computing. The 16 compute nodes feature: two dual-core 2.4 GHz AMD Opterons with 8 GB of memory, four NVIDIA Quadro 5600 GPUs, each with 1.5 GB of memory, and a Nallatech H101-PCIX FPGA accelerator, 16 MB SRAM, 512 MB SDRAM.

NVIDIA donated the GPUs to the research team led by Wen‑mei Hwu, the Sanders-AMD Endowed Chair in Electrical and Computer Engineering at the University of Illinois at Urbana‑Champaign. Hwu is also a principal investigator on the Blue Waters project to build the world's first sustained‑petascale computing system for open scientific research, and the leader of a research theme in the University's Institute for Advanced Computing Applications and Technologies.

Since the cluster was installed in 2007, it has been in heavy use as NCSA staff, campus researchers, and researchers from other institutions have explored the potential these technologies have to accelerate a wide range of science and engineering applications, from molecular dynamics to weather modeling. The center plans to upgrade the cluster to 32 nodes and 128 GPUs in the near future.

By Trish Barker

School fills educational gaps to help students become the new innovators in high-performance scientific computing.

Dozens of graduate and doctoral students from a wide range of disciplines gathered at NCSA in August for the first summer school offered by the Virtual School of Computational Science and Engineering, part of the educational component of the Blue Waters petascale project.

Led by Wen-Mei Hwu, the Sanders-AMD Endowed Chair in Electrical and Computer Engineering at the University of Illinois at Urbana-Champaign, and David B. Kirk, chief scientist for the NVIDIA Corporation, the weeklong summer school focused on harnessing graphics-based and multicore processors to accelerate science and engineering applications.

"High-performance computing is entering a new era of massively parallel processors and petascale systems that will enable researchers to solve a new class of complex problems," says Hwu. "But to fully leverage these resources, researchers will need in-depth knowledge about parallel programming principles, as well as the parallelism models, communication models, and resource limitations of these processors."

Through lectures, hands-on labs, access to NCSA's 16-node cluster of graphics processing units (GPUs), and time for networking and collaboration, summer school participants gained knowledge about accelerators and experience in the programming required to leverage them for science and engineering research.

Those skills can be difficult for students in computational science to acquire, says Sharon Glotzer, a University of Michigan professor and director of the Virtual School.

"A huge challenge in teaching computational science is that computer science departments typically focus on topics that computer scientists need to know, while in 'domain' science and engineering departments, like physics, chemical engineering, or aerospace engineering, computational science courses focus on applications of simulation to those disciplines," Glotzer says. "Many aspects of the nuts and bolts of computational science then fall between the cracks, and as a result, it is not easy for today's students to learn all they need to know to become tomorrow's innovators in high‑performance scientific computing."

The Virtual School aims to fill those gaps, providing the next generation of computational scientists with the skills they need to leverage emerging petascale systems.

"Although I still have a ways to go on learning the specifics of fully utilizing accelerator technology, I feel that I have been given a firm foundation in the way the architecture works," says Rebecca Owston, a mechanical engineering graduate student at Purdue University and both a National Science Foundation Graduate Research Fellow and a National Defense Science and Engineering Graduate (NDSEG) Fellow. "I will be much more mindful of the types of engineering problems which can exploit the capabilities of GPUs now, especially considering their cost effectiveness compared to CPU clusters."

"This summer school has given me a starting point for CUDA programming for the NVIDIA GPUs that I plan to explore in depth," says Eric Jankowski, a University of Michigan chemical engineering graduate student and NDSEG Fellow. "Writing multi-threaded code for my research in nanoparticle self-assembly will help us look at bigger systems and longer time scales that are currently inaccessible. Maybe it will help us get results faster, or maybe it will help us look at things we couldn't even think about looking at before."

Summer school participants were also excited about the opportunity to connect with peers.

"The contacts I made will help me keep in touch with a circle of researchers from different areas that are interested in GPU programming," Jankowski says. "The contacts and friendships that were gained over the past week will hopefully be used to keep on top of cutting-edge GPU research and maybe even for a future collaboration."

Some participants who were brought together by the summer school have already begun collaborative projects. "The project I started during the summer school with Tom Henretty (Ohio State University) will hopefully provide the basis for porting coupled-cluster methods in NWChem and other quantum chemistry codes to GPU-enabled hardware," says Jeff Hammond, who works in chemistry and computer science at the University of Chicago and is a Department of Energy Computational Science Graduate Fellow.

Many summer school participants also were eager to share the knowledge and experience they gained with fellow students at their institutions.

"I plan to use what I learned at the summer school to consult on the application of GPU programming in the research projects of my fellow IGERT students through our regular seminars," says Kevin R. Tubbs, a National Science Foundation Integrative Graduate Education and Research Traineeship (IGERT) Fellow who works in computational fluid dynamics at Louisiana State University. "With the trends of high-performance computing moving toward heterogeneous systems, learning the basics and practical applications of GPU programming is a necessity for a young researcher in computational fluid dynamics and in fact all computational sciences."

In addition to the more than 40 participants who spent the week at NCSA, many more students were able to access the summer school thanks to streaming video, slides, and other content on the Web, and the University of Illinois at Chicago hosted a live high‑definition feed at its Electronic Visualization Laboratory.

For more information: www.greatlakesconsortium.org.

del.icio.us

Slashdot