Here is a collection of highlights from this week’s news stream as reported by HPCwire.
DOE Announces Annual Supercomputing Awards Bonanza, 1.6 Billion Hours!
It’s that time of the year again, when the folks at the US Department of Energy (DOE) get the fun job of awarding computer time to the most deserving high-end projects. This year, the numbers are impressive: 1.6 billion hours have been awarded to 69 projects. As always, this is all thanks to INCITE — the Innovative and Novel Computational Impact on Theory and Experiment program, now in its seventh year.
According to the press release, the INCITE program, “provides powerful resources to enable scientists and engineers to conduct cutting-edge research in just weeks or months rather than the years or decades needed previously, … facilitat[ing] scientific breakthroughs in areas such as climate change, alternative energy, life sciences, and materials science.”
The cutting-edge projects were selected based on a peer review process and on their computational readiness to advance scientific discovery. They will get time on DOE’s Leadership Computing Facilities at Argonne National Laboratory in Illinois and Oak Ridge National Laboratory in Tennessee.
24 million of the 1.6 billion hours will go toward helping researchers design materials for developing lithium air batteries, capable of powering a car for 500 miles on a single charge. To compare, lithium-ion batteries, which are used in currently-available plug-in hybrid electric vehicles, have a range of only 40 to 100 miles.
The National Center for Supercomputing Applications (NCSA) and the Institute for Advanced Computing Applications and Technologies (IACAT) have also announced some details of their supercomputing awards, which includes 28 million hours for a new project titled: “Performance Evaluation and Analysis Consortium End Station.” This important endeavor aims to maximize the utility of DOE leadership class systems, such as the Cray XT4, Cray XT5, and IBM Blue Gene/P.
You can read about this project and all of the INCITE awardees, here.
Let’s take a quick look at how this year’s awards stack up to previous years. First off, breaking the billion hour mark and tearing past it — the 1.6 billion figure is an impressive 80 percent higher than last year’s. In December 2008, the DOE Office of Science announced 66 projects and nearly 900 processor-hours. That 900 hour figure is over three times the number of hours announced in January 2008, when the DOE awarded 265 hours (shared among 55 projects).
One trend that sticks out is that the number of projects are not increasing anywhere near as sharply as the number of hours awarded. To recap, for the last three years, the hours awarded go from 265 to 900 to 1.6 billion, and the number of projects goes from 55 to 66 to 69. Applications, and science in general, have become so complex and data-intensive, that it’s a given projects will require more and more computing time to process all of the information. To illustrate: there’s a push in climate modeling to reflect real-world systems as much as possible instead of simplifying the problem by approximating physical processes. For example, representing the earth as an ellipsoid requires more supercomputing power than using the less-complex (but less accurate) spherical shape. And we haven’t even touched on how the current drive toward real-time systems will increase our computational demands!
NVIDIA Partners with University of Illinois on GPU Programming Textbook
A “first of its kind” textbook, Programming Massively Parallel Processors: A Hands-on Approach, was announced today from GPU innovator NVIDIA and its partner-in-processing, the University of Illinois at Urbana-Champaign. The 256-page text is authored by Dr. David B. Kirk, NVIDIA Fellow and former chief scientist, and Dr. Wen-mei Hwu, Chair of Electrical and Computer Engineering at the University of Illinois at Urbana-Champaign. The material is aimed at teaching advanced computer science students and professionals the fundamentals of parallel programming and GPU architectures.
The announcement explains the impetus for the book:
With conventional CPU-based computing no longer scaling in performance and the world’s computational challenges increasing in complexity, the need for massively parallel processing has never been greater. GPUs have hundreds of cores capable of delivering transformative performance increases across a wide range of computational challenges. The rise of these multicore architectures has raised the need to teach advanced programmers a new and essential skill: how to program massively parallel processors.
The press release claims that this is the “first and only text that teaches how to program within a massively parallel environment.” The material in the book has been tested at over 300 universities worldwide, and drafts of the book have been used in classes at the University of Illinois. The book utilizes OpenCL and CUDA C, the parallel computing language developed by NVIDIA specifically for massively parallel systems.
The publisher, Morgan-Kauffman, has set up a site for the book here.