Graphics chips challenge x86, Cell

Are graphics chips destined to penetrate supercomputer applications? Nvidia and Advanced Micro Devices seem to be headed down that road as they pack hundreds of specialized calculating engines on new graphics chips in pursuit of teraflop performance, reports today’s Wall Street Journal. Although the companies are targeting gaming applications with their chips, they are “also are marketing the technology to accelerate the performance of desktop and server systems on a wide range of technical chores,” the Journal reports, as they look to use the graphics chips to displace x86 processors from server farms and Cell-type processors from supercomputers.

But skeptics caution that the graphics chips, despite their impressive performance, can be hard to program for non-graphics-intensive applications. The companies are addressing the programming challenge, concludes the Journal: “Nvidia has an internally developed programming scheme called CUDA; AMD plans to use a programming technology called OpenCL that Apple Inc. and other companies are backing.” In other differences, Nvidia is focusing on "extreme performance" on a single chip, while AMD is focusing on “smaller chips that cost less to make and consume less power.”

You can read the complete article here; subscription required.