November 17, 2009
With all the talk this year about Intel's Nehalem CPU, it was AMD's Istanbul chip that helped vault Cray supercomputers to the number 1 and number 3 spots on the TOP500. After a thorough upgrade of ORNL's "Jaguar" with AMD's six-pack Opteron, the renovated Cray XT5 system managed a 1.75 petaflop Linpack score, which was more than enough to beat out last year's number one supercomputer, the 1.04 petaflops IBM "Roadrunner" system at Los Alamos. The other Cray super in Tennessee, nicknamed "Kraken," also got the Istanbul treatment over the last few months and managed the number 3 spot with 832 teraflops, barely edging out the 826 teraflops recorded by the IBM Blue/Gene P "JUGENE" at Juelich.
As it turns out, this is the first time in the TOP500's 16-year history that Cray grabbed the number one spot on the list. And even though these Linpack rankings tend to be much ado about nothing, it was gratifying to see the iconic supercomputer company land atop the iconic supercomputer list.
We also saw some new blood in the top 10. The recently announced Tianhe-1 supercomputer installed at the National University of Defense Technology in China comes in at number 5 with 563.10 teraflops. That system is interesting in a number of ways. For one thing, it's the FLOPiest supercomputer ever installed in China, and garnered the highest TOP500 ranking that the country has ever attained. It also happens to be built from a mix of Intel CPUs and AMD GPUs -- an unusual team-up outside of the consumer space. Each of the 2,560 compute nodes sports two Xeon E5540s hooked to two ATI Radeon HD 4870s, representing over a petaflop of peak performance and about half of that in Linpack.
The other new entry at the top of the list is one from Oracle -- oops, I mean Sun. The "Red Sky" supercomputer at Sandia National Laboratories is a Nehalem-InfiniBand vintage Sun Blade super that grabbed the number 10 spot. Its Linpack number was 423.90 teraflops.
Skipping to the end of the list, at number 500, we have an IBM BladeCenter PowerPC cluster that comes in at just a skosh above 20 teraflops. For those of you keeping score at home, that's 3 teraflops more than the number 500 system 6 months ago.
If you'd like to see your machinery on the T0P500, but are on a budget, you might be able to manage 20 teraflops with dozen or so of the new Fermi GPU-equipped Tesla 20-series servers hooked up to some cheap x86 boxes. Of course, by the time the new Teslas hit the streets next May, it will almost be time for the June TOP500 list, which undoubtedly will raise the cost of admission once again.
Posted by Michael Feldman - November 16, 2009 @ 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?