News and Announcements
Jack Dongarra in Manchester
As a Turing Fellow at the University of Manchester for the past several years, Jack has had the opportunity to spend part of each summer interacting with the people in the University’s schools of Mathematics and Computer Science. Below are two pictures from this summer’s visit. The first shows Sven Hammarling, his wife Pam (center), and Magaret Wright from NYU, all enjoying a salubrious libation in celebration of Sven’s seventieth birthday.
The second photo shows Jack and Iain Duff finishing a “10 + 2” year old bottle of Scotch. Both Iain and Sven are two of Jack’s long-time friends, and have often collaborated with Jack and ICL.
SAAHPC ’11
Mitch Horton and Stanimire Tomov recently attended the Symposium on Application Accelerators in High Performance Computing (SAAHPC), on July 19-20, in Knoxville. As the name indicates, the majority of presentations at the conference focused on the use of accelerators, namely Nvidia GPUs, to increase performance in scientific computing. Greg Peterson, from UT’s Electrical Engineering and Computer Science department, was the symposium chair, and Pradeep Dubey, from Intel, gave the keynote address.
Mitch’s presentation and paper, “A Class of Hybrid LAPACK Algorithms for Multicore and GPU Architectures,” describes what is essentially an extension to the current release of MAGMA for QR, LU, and Cholesky factorizations, in which an attempt to fully utilize all available CPU cores is made while using a single GPU on a single node.
The Exaflops Target
In the August issue of Communications of the ACM, Jack Dongarra weighs in on the goal of exascale supercomputing. The article, which also discusses Asia’s rise to prominence in high performance computing, gets down to brass tacks when talking about the potential energy consumption of an exascale system. An analogy is also established between the jump from teraflops to petaflops, which took roughly a decade, and the challenging jump to exascale. Can a 1,000-fold leap in performance be repeated, despite limitations in software parallelism and energy efficiency? The race is on to find out.
Piotr Makes the News
Our very own Piotr Luszczek made the front page of the July 26th issue of The Daily Beacon for his work on the LINPACK Benchmark app for iOS. Specifically, Piotr is trying to break one gigaflops on the iPad 2 by optimizing the LINPACK Benchmark to run more efficiently on the Apple device.
This just in: On August 1st, “just before lunch,” Piotr successfully broke the one gigaflops barrier on the iPad 2, using the LINPACK Benchmark app!
The End of TeraGrid and the Dawn of XSEDE
Shirley and Piotr recently attended TeraGrid’11 which was held in downtown Salt Lake City the week of July 18. Piotr helped with a FutureGrid tutorial on the 18th and also graciously carried Shirley’s PAPI poster to SLC, given that Shirley was staying 4+ miles away due to not being able to get into the conference hotel block. Shirley co-authored a paper with ICL alumnus Haihang You, who then presented the paper in the Tuesday afternoon Science and Technology session. Shirley also presented a poster on new features of PAPI in the Tuesday evening poster session.
Much of the talk during the conference was about TeraGrid ending and being replaced by Extreme Science and Engineering Digital Environment (XSEDE). In fact, you’re not supposed to go to the TeraGrid website anymore. Instead, the latest and greatest info is to be found on XSEDE’s website. A big emphasis with XSEDE is including a wider range of resources in the cyberinfrastructure, not just supercomputers, but also scientific instruments, data collections, etc., and making it all easy to access and use together, as well as broadening the user base. Consequently, projects such as Eclipse Parallel Tools Platform and Science Gateways are getting a lot of NSF funding and attention. But that’s not to say that good old-fashioned performance measurement and optimization aren’t still important. There were plenty of visitors at the PAPI poster, including several NSF program managers. Most were already familiar with PAPI, and were just interested in hearing about recent developments, but a few newbies stopped by who didn’t know anything about it yet.
All in all, it was an information- and fun-filled week, including an excursion to the Clark Planetarium with educational IMAX movies on black holes and such things. Since it was summertime, many attendees brought spouses and/or kids as well. I guess we’ll look forward next year to XSEDE’12.
ICL Retreat 2011
This year’s ICL retreat will be held on August 11-12 at the Buckberry Lodge in Gatlinburg. Please talk to Leighanne about transportation if you would like to carpool.
The Lodge at Buckberry Creek
961 Campbell Lead Road
Gatlinburg, TN 37738
Click here for directions or use the map below.
Recent Releases
PLASMA 2.4.1 Released
The Parallel Linear Algebra for Scalable Multi-core Architectures (PLASMA) 2.4.1 release is now available for download.
This package contains two major bug fixes:
- Fixed bug with Fujitsu compiler reported on the forum.
- Unbind threads in PLASMA_Finalize to avoid problem of binding in OpenMP section following PLASMA calls (still possible on Mac and AIX without hwloc). A better fix is to create the OpenMP thread in the user code before any call to PLASMA, thanks to a fake parallel section.
See the software page to download the tarball.
QUARK Users’ Guide Released
The QUeueing And Runtime for Kernels (QUARK) Users’ Guide has been released. QUARK is used for scheduling in both the PLASMA and MAGMA projects. The guide is available as a PDF here.




















