By Patricia Resende / CIO Today. Updated October 01, 2009.
Nvidia used the GPU Technology Conference in San Jose, Calif., to show the world it has reached a new milestone in graphics processing. Nvidia demonstrated its next-generation GPU architecture, code-named Fermi. The new architecture will not replace the CPU, but will secure a significant place in PC system architecture.
Fermi's graphics capabilities will mean substantial improvements to game play, multimedia encoding and enhancement, and other PC applications, according to the company.
"It is completely clear that GPUs are now general-purpose parallel computing processors with amazing graphics, and not just graphics chips anymore," said Jen-Hsun Huang, cofounder and CEO of Nvidia. "The Fermi architecture, the integrated tools, libraries and engines are the direct results of the insights we have gained from working with thousands of CUDA (compute unified device architecture) developers around the world."
Huang provided several examples of Fermi's potential. In one demonstration, he created a realistic physical reaction for a game by throwing rag dolls at destructible walls. In another, he showed its value in 3-D stereoscopic video. A third demonstration showed how GPUs can be used to enhance processing of ultrasound recordings to detect breast cancer.
Nvidia said Fermi has increased the performance startup by eight times over Nvidia's last-generation GPU. The increase is critical for high-performance computing applications such as quantum chemistry, linear algebra, and numerical simulation, according to Nvidia. It added that Fermi also provides supercomputing features and performance at one-tenth the cost and one-twentieth the power of traditional CPU-only servers.
The GPU architecture is designed for C++, makes parallel programming easier and increases performance on a greater variety of applications than in the past, the company said. Performance increases are seen in ray tracing, physics, high-precision scientific computing, sparse linear algebra, and sorting and search algorithms, according to Nvidia.
Fermi is also the first GPU to provide error-correcting code (ECC) protection for DRAM, Nvidia said.
Analysts have touted Nvidia's new architecture as the first complete GPU architecture. Because Fermi is derived from Nvidia's graphics products, it ensures the company will sell millions of software-compatible chips to PC gamers, according to Peter Glaskowsky, a senior technology analyst for Envisioneering Group.
Nvidia also garnered support for Fermi from various companies, including Dell, Hewlett-Packard and Microsoft.
But it was Oak Ridge National Laboratory that stepped up to announce it will use Nvidia's GPU architecture for a new supercomputer. The supercomputer, expected to be 10 times more powerful than the current fastest supercomputer, will be used to research energy and climate change.
"This would be the first coprocessing architecture that Oak Ridge has deployed for open science, and we are extremely excited about the opportunities it creates to solve huge scientific challenges," said Jeff Nichols, Oak Ridge associate lab director.