Before you get all hot and bothered because GPUs are starting to wipe the floor with CPUs, ask yourself: What's the difference? Isn't a GPU just a CPU on steroids?

NVIDIA wants you to believe there's more to it than that. At their recent "CUDA Day" on February 4th in San Francisco, the company invited the Press in for tea at the Clift Hotel and staged a series of testimonials from the likes of Adobe, MotionDSP, Kaspersky Labs, CyberLink, Agilent, muvee, and Siemens Medical -- customers from a broad spectrum of technologies, using GPU processing power to crunch an even broader spectrum of applications.

There's no doubt that the customer demos at the Clift were impressive, no doubt that NVIDIA's own clip of fast-paced gaming imagery was breath-taking. But again, the question:

Could equivalent levels of performance in video processing, medical imaging, data smoothing, and test and measurement be accomplished using multi-core CPUs? Is a GPU the only processor architecture with sufficient mojo to make this stuff zing?

Per their recent event in San Francisco, NVIDIA clearly believes that to be the case, as they openly challenged Intel's dominance during the presentation. More likely, however, the jury's still out on the question, even as the lines between GPUs and CPUs blur with definitions becoming less distinct.

NVIDIA's CUDA parallel computing hardware architecture "supports standard computing languages ... and other APIs for general purpose computing on the GPU."

NVIDIA is clearly announcing to the world that they are the future. Only time and market share, however, will tell if their bravado is worse than their byte.

March 4, 2010

************************************

Print Version