IT MIGHT no longer be the undisputed leader in graphics cards, but Nvidia has managed to power what is currently the fastest publicly known supercomputer.
The Tianhe-1A was revealed at the HPC 2010 China conference today with an operational performance 2.507 petaflops, easily surpassing the AMD Opteron shod Cray XT5 nicknamed Jaguar that currently heads the Top 500 list. Nvidia had already powered the Dawning Nebulae to second place on the list, with a theoretical peak performance that surpasses both the Jaguar and today's Tianhe-1A.
However the difference between operational and peak performance is significant, and to produce a 42 per cent increase over the Jaguar's operational performance is very impressive. Nvidia claims not only the performance crown but touts the energy efficiency of the Tianhe-1A machine, saying that it offers three times higher power efficiency compared to traditional CPU based clusters.
The supercomputer was designed at the National University of Defense [sic] Technology in China and is installed at the National Supercomputer Center in Tianjin. Nvidia says the Tianhe-1A is already fully operational and will be used as an open access system for scientific computation.
At present Nvidia has not disclosed the configuration of the nodes that make up Tianhe-1A, simply saying that there are Tesla GPUs inside.
Both AMD and Nvidia maintain that GPUs are the way towards attaining exascale computing, though this announcement will certainly hurt AMD. Not only will it lose the top spot on the prestigious Top 500 list, it will lose it to Nvidia, its principal rival in the GPGPU market.
It will be interesting to see whether Nvidia has managed to up the performance of Dawning Nebulae and possibly take the top two spots on the list. What is for certain, however, is that come November, the latest edition of the Top 500 list will have a new machine leading the pack. µ
Xbox One price of Microsoft-owned game stays the same
And let you stuff more crap into every tweet
Inspector Clouseau and friends pay a visit
Thanks to Parallels Access 3.1