AMD says the R600 breaks teraFLOPS barrier
Last year, Intel unveiled a prototype 80-core chip it said was capable of computing a trillion floating point operations per second (teraFLOPS). Intel recently revealed that it had no plans to commercialize that particular chip, but AMD isn't sitting idly by. At a press event in San Francisco late yesterday, AMD demonstrated a "Teraflop in a Box" system built around its upcoming R600 graphics processor. The system, which was based on AMD's existing dual-core Opteron processors and contained two R600 Stream Processors, was reportedly capable of performing more than a trillion calculations per second using a general multiply-add (MADD) calculation.
AMD touts the achievement as an important milestone towards its "Accelerated Computing" vision of having specialized co-processors speed up conventional systems. That said, the R600's computing power may not be unique. Nvidia says its G80 graphics processor can handle around 520 gigaFLOPS, so a pair of GeForce 8800 GTX graphics cards would presumably be able to break the teraFLOPS barrier just like AMD's dual R600 Stream Processors.
In related news, ZDNet blogger David Berlind was at the event in San Francisco yesterday and snapped a few pictures of the R600. AMD allowed him to release the pictures on condition that he refer to the product as an R600 prototype. The photos are similar to other leaked shots of the card we saw last month, but they show just a bit more detail. Interestingly, beneath the card's massive (and reportedly foot-long) cooler lives a PCB that actually looks somewhat shorter than that of Nvidia's GeForce 8800 GTX. Let's hope the final retail cooler isn't quite as gargantuan.