CHIP DESIGNER Nvidia has been knocked off the top spot in the Top 500 list of supercomputers, with a Sparc-based machine taking its place.
The who's who of the high performance computing (HPC) world turned up in Hamburg to find out whether China's Tianhe-1A has retained the top spot on the prestigious Top 500 list. Instead Japan took the lead for the first time since NEC's Earth Simulator with a 548,352 core Sparc system assembled by Fujitsu.
Known as the K Computer, the Sparc beast doesn't feature a single GPGPU, instead it runs 68,544 Sparc64 VIIIfx eight-core CPUs at 2.0GHz. On the Top 500 list the numbers are always big, but the K Computer is significantly larger than the Tianhe-1A, with twice the number of cores of any system in the Top 500. It's faster than the rest of the top five systems combined and it isn't even fully up to speed yet.
The reason why Fujitsu went for the seemingly mundane name of K Computer is due to the Japanese word 'Kei' which is 10^16 in Japanese. That number signifies 10 petaflops, the ultimate goal of the K Computer, though as of today its recorded speed is a mere 8 petaflops.
According to the organisers of the Top 500 ranking, the June 2011 list is the first where every computer in the top 10 exceeds one petaflop. Intel might well flood the list with its Xeon-based clusters but worryingly for the firm only one machine in the top 10 is running purely x86 Xeon chips, the Pleiades SGI Altix cluster in eighth place.
On the other hand, while Nvidia might have lost the top spot on the list, three out of the top five clusters use Nvidia GPGPUs. All three machines use Intel Xeon chips but the majority of their horsepower comes from Nvidia's Tesla boards.
Going back to the machine of the moment, we asked around the show floor whether a 10 petaflop K Computer will be fast enough to run Crysis, but we couldn't get an answer. So we will just have to recommend that the chaps at Riken Advanced Institute for Computational Science in Kobe turn down the detail a bit and hope for the best. µ