News and Announcements
TOP 500 – June 2014
The 43rd TOP500 list was released on June 23rd at ISC’14 in Leipzig, Germany. The top 5 machines are identical to the November 2013 list, with China’s Tianhe-2 maintaining the top spot (33.86 petaflop/s), followed by DOE’s Titan (17.59 petaflop/s).
There was an interesting addition to the top 10 machines, however, with a Cray XC30 machine, installed at an undisclosed U.S. government site, hitting 3.14 petaflop/s to earn the #10 spot. Take a look at the entire list here.
| Rank | Site | System | Rmax (TFlop/s) |
|---|---|---|---|
|
1 |
33,862.7 |
||
|
2 |
DOE/SC/Oak Ridge National Laboratory |
Titan – Cray XK7 |
17,590.0 |
|
3 |
DOE/NNSA/LLNL |
17,173.2 |
|
|
4 |
RIKEN Advanced Institute for Computational Science (AICS) |
K computer, SPARC64 VIIIfx |
10,510.0 |
|
5 |
DOE/SC/Argonne National Laboratory |
8,586.6 |
|
| See the full list at TOP500.org. | |||
Jack Dongarra and Erich Strohmaier discuss the TOP500, recent trends in HPC funding, and the value of the LINPACK benchmark.
HPCG: One Year Later
Since its introduction over a year ago, the HPC Preconditioned Conjugate Gradient (HPCG) benchmark has delivered promising results. Intended to be a new HPC metric, HPCG is designed to measure performance that is representative of modern HPC capability by simulating patterns commonly found in real science and engineering applications.
As Jack Dongarra noted at ISC’14, the HPL benchmark—used for the TOP500 rankings—is not the best analog for modern software applications. To keep pace with the changing hardware and software infrastructures, HPCG results will be used to augment the TOP500 rankings to show how real world applications might fare on a given machine. Plans to further develop the HPCG rankings will be revealed at SC’14. For now, you can see how the HPCG benchmark would have ranked its top 14 machines in the table below.
Conference Reports
International Supercomputing Conference
This year’s International Supercomputing Conference (ISC’14) was held on June 22nd through the 26th in Leipzig, Germany. ICL’s Jack Dongarra and Jakub Kurzak both made their way to Europe’s premier HPC conference, which draws attendees from industry, government, and academia from across the globe, with strong attendance this year from the U.S. and Japan.
On Sunday, Jack, Jakub, and ICL alum Hatem Ltaief teamed up for a Linear Algebra Software tutorial. Jack was also on hand for the TOP500 award ceremony on Monday, and then gave a talk on Fault Tolerance in Numerical Library Routines on Tuesday. On Thursday, Jack gave a presentation on the HPCG benchmark and its progress over the last year, which drew a lot of excitement.
Interview
Florent Lopez
Where are you from, originally?
I was born in Bordeaux, located in the southwest of France near the Atlantic coast. The region is mostly known for its good food and prestigious wineries. In addition, the historic center of Bordeaux is listed as an UNESCO World Heritage site because of the outstanding architecture of the urban area and quality of the local cuisine.
Can you summarize your educational background?
I earned my degree in computer science at ENSEEIHT, an engineering school in Toulouse where I developed a strong interest in numerical linear algebra and parallel computing. I spent the last semester of my engineering studies at EPFL in Lausanne where I also had the opportunity to work for a research laboratory. This work in particular was a fascinating experience and motivated me to pursue a PhD. I moved back to Toulouse in 2012 where I enrolled in the PhD program at the IRIT laboratory.
Tell us how you first learned about ICL.
My advisor, Alfredo Buttari, talked a lot about ICL as he spent some time in the laboratory as a post-doc. I also had the opportunity to work with several ICL collaborators at INRIA Bordeaux, such as Emmanuel Agullo and Mathieu Faverge, with whom I had many occasions to discuss the laboratory.
What are your research interests?
My current research interests include sparse matrix algorithms and high performance computing.
What are you working on during your visit with ICL?
My work consists of developing sparse direct methods for heterogeneous architectures, like GPU-based systems, in the context of qr_mumps, which is a multifrontal solver originally designed for multicore architectures. In order to tackle the complexity of sparse matrix algorithms and modern architectures, my approach consists of relying on runtime systems such as ICL’s PaRSEC. For this reason I decided to visit ICL in order to work on a PaRSEC-based version of qr_mumps in collaboration with George Bosilca.
What are your interests/hobbies outside work?
I really enjoy hiking, climbing, and kayaking. During the winter season I also like skiing.
Tell us something about yourself that might surprise people.
I am also very interested in computer security. When I was studying at EPFL, I followed many courses in this field and I worked on a project whose aim was to evaluate threats to the privacy of mobile users using device-to-device communication technologies. I was all the more excited about this work when it allowed me to get my first publication.
Just a reminder, ICL and UTK will be closed on July 4th to observe Independence Day.