New teraflop/s GPU cluster facility at BU

BU’s College of Engineering, the Center for Computational Science (CCS), Information Systems & Technology (IS&T) and the Bioinformatics Program are introducing a significant expansion to the facilities for research
computing with the acquisition of a 160 GPU-accelerated cluster. At 80 teraflop/s in peak double-precisionperformance, this system will be the fastest ever deployed at BU.

The new computing resource was made possible by the investment of IS&T funds, a contribution from the CCS and a donation by Hewlett Packard of the 160 GPU cards, valued at $320,000.

It was also made possible by the catalyzing effect, and strengthening of a community, started by a few researches at Boston University. Over the past four years, experiments with GPU hardware were seeded by an NSF ARRA stimulus grant to build a prototype “Experimental GPU Cluster for Fundamental Physics” awarded to Professor Richard Brower as principal
investigator, Lorena Barba and Claudio Rebbi, as co-principal investigators.

The new system represents a build-out from the EAGER-supported BUNGEE (Boston University Networked GPU Experimental Environment) cluster. The new donation from HP was motivated by the expertise that was developed over the past years thanks to the EAGER seed grant and the BUNGEE cluster.

“We are extremely happy that our experimental GPU project has brought us to the stage where we are able, in partnership with IS&T, to bring this technology to the whole university” said Prof. Brower.

The system, to be named BUDGE (Boston University Distributed Gpu Environment) is being installed now, and consists of 20 HP 12-core Intel Xeon servers with 48GB of RAM each (soon to be doubled to 96GB), with eight NVIDIA Tesla M2070 GPU accelerators, each with 6GB RAM per accelerator, for a total of 160 GPUs in the deployment. Network is provided by a Cisco 3500 gigabit ethernet switch on the front end and a Mellanox QDR Infiniband 40 gigabit switch on the back end.

To receive updates about its status as well as other GPU initiatives at Boston University, please see our web site at http://blogs.bu.edu/gpu, where you can also subscribe to the gpu@bu.edu mailing list.

Rich Brower, Lorena Barba, Claudio Rebbi and Glenn Bresnahan

GPU@BU workshop announcement

hicThe second GPU@BU workshop will be held on November 8–9, 2011, with sponsorship from the BU Center for Computational Science and the Hariri Institute of Computing and Computational Science & Engineering, and additional funding from NSF.

The two-day event will assemble several researchers from Boston-area universities, and guests, to discuss the latest research in using GPU hardware for scientific computing applications.

This workshop is organized by the co-investigators of NSF award OCI 0946441: Richard Brower (PI), Lorena Barba and Claudio Rebbi. Current confirmed speakers include:

  • Richard Brower, Boston University
  • Cris Cecka, Harvard University
  • Ben Chandler, Boston University
  • Michael Clark, currently at Harvard but joining Nvidia soon
  • Jonathan Cohen, Nvidia
  • Andrew Corrigan, Naval Research Laboratory
  • Chris Hill, MIT
  • David Kaeli, Northeastern University
  • Simon Layton, Boston University
  • Miriam Leeser, Northeastern University
  • Rio Yokota, formerly at Boston University and recently at KAUST

See more details on the GPU@BU workshop page, or go to the Registration page!

Welcome to GPU@BU

Several research groups at Boston University are leveraging the power of graphics processors, known as GPUs, for scientific computing.  Through this blog, we hope to disseminate the research output of these groups, as well as to share our experiences and successes.

Experimental GPU cluster at BU

Thanks to NSF and NIH funding,  a new GPU-enabled computational cluster has been installed at Boston University.  The cluster offers peak performance in excess of 32 teraflop/s (or 32 million million operations per second), at an unprecedentedly low cost for such number-crunching power.

This supercomputer will be used by the principal investigators for research in particle physics and fluid dynamics (Richard Brower, Lorena Barba, Claudio Rebbi), as well as bioinformatics and computational biology (Martin Herbordt).  An additional goal is to foster a “GPU community” at BU; to this end, the cluster will also be made available to other researchers for exploratory projects. This initiative has already seen involvement from several other research groups.

In November 2009, a collection of faculty, postdoctoral researchers, and graduate students came together to learn about GPU computing in a workshop organized by the co-PIs of the NSF-funded project “Experimental GPU cluster for experimental physics.” This workshop, dubbed GPU@BU, was attended by researchers from all around the Boston area — including Harvard, MIT, Northeastern, and Brown — as well as distinguished guest speakers from NVIDIA Research and Nagasaki University.

Links