November 19, 2008
Last week's announcement of the upgraded "Jaguar" system at Oak Ridge National Laboratory (ORNL) had a lot of people, including yours truly, thinking that the Cray super would take the TOP500 crown this time around. It was not to be.
Last week, it was announced that Jaguar's extreme makeover would deliver a world record 1.6 peak petaflops. That would translate into a Linpack benchmark close to 1.3 petaflops, which would have easily beaten the slightly enhanced 1.1 petaflop Roadrunner at Los Alamos National Laboratory. As it turned out, the folks at ORNL ended up splitting the Jaguar into two separate Linpack submissions. The older Cray XT4 part yielded 205 teraflops and came in at number 8 on the new list; the new XT5 part took the number 2 spot with 1.06 petaflops -- a mere 3 percent less than the new Roadrunner.
Why the ORNL machine was split up is a bit of a mystery. One can surmise that ORNL and Cray didn't have time to perform a complete benchmark run before the TOP500 submission deadline. Supposedly it took 18 hours for a single Linpack run on the XT5 part of the system, so it might not have been worth the extra time to get a full clean measurement on the entire system.
All this has nothing to do with the Jaguar's real purpose. The compelling story for this machine will be how it is able to handle "big science" applications in the petascale realm, especially under the DOE's INCITE program. We can expect to see the some of most challenging codes in climate research, fusion, astrophysics, biology, atomic physics, chemistry, and nanoscience running on Jaguar over next few years. (Since Roadrunner lives at Los Alamos, we have to assume most of the cycles on that machine will be devoted to classified work for the nuclear weapons program.)
Following the one-two punch of Roadrunner and Jaguar, the next four machines on the list are each just under half a petaflop. That's a lot of computing power in those first six slots -- nearly 4 petaflops. To give you a sense of how top-heavy the TOP500 has become, the remaining 494 computers on the list represent just (!) 12.9 petaflops.
The U.S still dominates the TOP500 with 291 systems. In fact, the top 9 systems are all deployed in the U.S.; but of the next 20 systems on the list, only 7 are. France, in particular seems to be moving up fast, with four supercomputers in the top 30. The top non-U.S machine, at number 10, is the Chinese-built Dawning 5000A at the Shanghai Supercomputer Center, and is the largest system that is running the Windows HPC 2008 operating system.
As has been the case over the past couple of years, the turnover on the TOP500 was rather high this time around. More than half the systems on the June list -- 267 -- didn't survive the November cut. The machine in the 500 slot is now a 12.6 teraflop Dell cluster.
Posted by Michael Feldman - November 18, 2008 @ 9:00 PM, Pacific Standard Time
Michael Feldman is the editor of HPCwire.
No Recent Blog Comments
10/30/2013 | Cray, DDN, Mellanox, NetApp, ScaleMP, Supermicro, Xyratex | Creating data is easy… the challenge is getting it to the right place to make use of it. This paper discusses fresh solutions that can directly increase I/O efficiency, and the applications of these solutions to current, and new technology infrastructures.
10/01/2013 | IBM | A new trend is developing in the HPC space that is also affecting enterprise computing productivity with the arrival of “ultra-dense” hyper-scale servers.
Ken Claffey, SVP and General Manager at Xyratex, presents ClusterStor at the Vendor Showdown at ISC13 in Leipzig, Germany.
Join HPCwire Editor Nicole Hemsoth and Dr. David Bader from Georgia Tech as they take center stage on opening night at Atlanta's first Big Data Kick Off Week, filmed in front of a live audience. Nicole and David look at the evolution of HPC, today's big data challenges, discuss real world solutions, and reveal their predictions. Exactly what does the future holds for HPC?