The future of high-performance computing might very well depend on graphics cards. It's not just CPUs that are crunching numbers faster and faster. Today's GPUs are just as efficient at performing floating-point calculations--perhaps even more so, especially when special floating-point hardware is integrated directly into the GPUs:
"At the International Supercomputing Conference, AMD (NYSE:AMD) today introduced its next-generation stream processor, the AMD FireStream 9250, specifically designed to accelerate critical algorithms in high-performance computing (HPC), mainstream and consumer applications. Leveraging the GPU design expertise of AMD’s Graphics Product Group, AMD FireStream 9250 breaks the one teraflop barrier for single precision performance. It occupies a single PCI slot, for unmatched density and with power consumption of less than 150 watts, the AMD FireStream 9250 delivers an unprecedented rate of performance per watt efficiency with up to eight gigaflops per watt."
Stream Computing is a form of parallel processing that utilizes the additional computational power of a graphics engines. GPUs, such as AMD's FireStorm cards, which are designed with this specific functionality, are referred to as General-Purpose computing on Graphics Processing Units (GP-GPU). GP-GPUs act essentially as powerful math co-processors that perform heavy-duty number crunching beyond just the graphics calculations for which the chips were originally designed.
Stream Computing use has been building momentum in scientific and financial sectors, where heavy-duty number crunching is mission critical. It hasn't quite made its way into consumer-level applications yet, however, unless you count Folding@Home.
It's a good bet that with AMD's recent announcement of partnering with Havok, AMD is likely counting on using Stream Computing techniques in its consumer-level Radeon GPUs to perform real-time physics calculations for 3D games. AMD has already demonstrated Havok-based physics calculations performed by a dedicated GPU; but by implementing Stream Computing components into the GPUs, the need for an additional, dedicated GPU for physics, should no longer be necessary.
The FireStom 9250 and its SDK will be available sometime in the third quarter of 2008 for $999. This is a sizable drop in price from the previous generation, FireStrom 9170, which currently sells for $1,999.