Pimp My CPU

Unlike grid computing, which improves processing by throwing more commodity processors onto the fabric, massive parallel computing looks to offload as much parallel-friendly processing as possible from the individual CPU.

Right now, the industry has access to a host of field-programmable gate arrays (FPGAs), but the really interesting technology is coming from the major graphics processing units (GPUs) manufacturers. It shouldn't be surprising, as parallel computing is their bread and butter. In this issue, AMD, which acquired graphics card provider ATI Technologies, plans to launch its GPU-CPU hybrid in 2009 (see story, page 4). Let's hope that Nvidia hasn't already stolen the market with the Tesla line of GPU computing appliances that it is currently shipping.

Nvidia's offerings range from cards to rack-mountable and free-standing chassis so that firms don't have to muddle about with their servers' motherboards. More importantly, its has CUDA, its toolbox for the C development environment that lets programmers exploit the parallel processing environment.

However, the clock is ticking. Once Fusion goes into production, AMD will be able to deliver it with its CPU without having to deal with the marketing stigma usually associated with technology sporting the appliance label.

The race is heating up and it could be anyone's game.

Comments? Send them to me at rob.daly@incisivemedia.com.