Nvidia sails to Cuda in bid to stem ATI's Stream

Revolutionary?

By Tony Smith • Get more from this author

8th November 2006 20:50 GMT

First AMD/ATI introduced Stream Computing, a scheme to accelerate scientific, engineering and other non-gaming apps using GPU parallel processing technology. And now here's Nvidia's version, CUDA.

Not that it's clear what CUDA stands for - Nvidia didn't provide a solution to the acronym in its press release. Computationally Ubiquitous Digital Architecture, perhaps? Other, more inspired sugestions to the usual email address, please.

Whatever the moniker's meaning, it boils down to Stream. As it were. Nvidia can at least leverage all 128 of its GeForce 8800 GTX's unified shader processors for high-end apps. For now, AMD/ATI has to split tasks into code more suited to either vertex or pixel shaders, but as with DirectX 10 support, AMD/ATI's probably not that far off unified shaders of its own. RS700, anyone?

Not that gaming's out of the question, and CUDA is likely to underpin the 8800 GTX's Quantum Effects Technology, of which much has been mentioned but very little detailed by Nvidia itself. In September, Asus let slip Nvidia was working on a physics add-in card - a mistaken reference to the three-GPU solution announced today as the nForce 680i SLI chipset, or a hint at the physics processing capabilities of the G80's Quantum Effects and CUDA?

Nvidia also launched a specialised C compiler today to allow developers to code technical computations to run on a G80-class GPU. The SDK can be downloaded here. ®