by Andy Walsh

Last Thursday and Friday, Georgia Tech hosted a tutorial and workshop for computational scientists who are eager to run their simulations on the recently activated NSF Keeneland cluster. Keeneland is a NSF project to bring innovative computer architectures to NSF’s computational science community. To deploy Keeneland, Georgia Tech has partnered with the University of Tennessee-Knoxville, the National Institute for Computational Sciences and Oak Ridge National Laboratory. The workshop was intended to teach researchers how to get the most out of this prominent NSF resource for open science. The workshop was announced only three weeks before the event, but registration still sold out quickly.

The initial deployment of Keeneland is a seven cabinet HP solution powered by 360 Tesla 20-series GPUs.

Scientists from Brown, the Center for Disease Control, Florida, Florida State, Notre Dame, Oak Ridge, Oregon, Purdue, Rice, Stanford, Tennessee, USC, and Yale all joined in the event. All in all, about 70 researchers from 25 institutions came to learn more about GPU supercomputing. And another dozen researchers had to be turned away.

Although it’s clear there is a lot of enthusiasm in the open science community around Keeneland, it was surprising to see just how popular this event was.   “We only announced this workshop three weeks ago in order to respond to user requests,” said Keeneland Project Director and workshop organizer Jeff Vetter of Georgia Tech and ORNL. “It was like a flash mob. We are already starting to plan more of these events.”

There are already about 60 active projects on Keeneland including studies on advanced physics, bioscience, fluid dynamics, molecular dynamics, neural science, nuclear physics, protein folding, and even a GPU MapReduce project.

During the workshop I had lunch with a researcher who was using GPUs to simulate the expected orbital trajectories of spacecraft, satellites and space debris. He turned a traditionally serial recursive math function into a parallel problem – instead of one Earth, he modeled it as 40,000 masses each with their own gravity.  His research promises to provide better predictive capabilities and minimize the uncertainty of situations like what occurred two weeks ago, when astronauts almost evacuated the  International Space Station, before tracking data confirmed oncoming space debris would not hit.

Thanks to Georgia Tech for hosting this important event, I learned a lot.

If you are a GPU Computing developer wanting to learn more about programming the GPU, then be sure to sign up for our registered developer program. Registered CUDA developers get access to in-depth Webinars and interactive Q&A sessions with experts from NVIDIA and Partners. These webinars are your opportunity to get live answers to your questions and to get hints and tips directly from the folks on the leading edge of GPU Computing.