Nvidia says its new Fermi GPU will run supercomputers
- — 01 October, 2009 10:51
Nvidia showed off a new GPU architecture on Wednesday that it hopes will allow it to move beyond gaming to play a greater role in the supercomputing market.
The new Fermi architecture should provide more realistic graphics for gamers, but it also includes technologies that make it well-suited to highly parallel computing environments, Nvidia CEO Jen-Hsun Huang said in a speech at the company's GPU Technology conference, which was webcast from San Jose, California.
He showed a graphics card on stage with a prototype of the new chip inside it, but he didn't announce any specific product plans Wednesday or say when Fermi would hit the market.
Oak Ridge National Laboratory will use Fermi chips in a supercomputer for scientific tasks such as climate modeling, said Jeffrey Nichols, associate lab director, who joined Huang on stage.
The lab currently has the Jaguar supercomputer developed by Cray, which is based on Advanced Micro Devices Opteron processors.
The Fermi chip has 3 billion transistors and 512 processor cores, and includes a new technology called GigaThread 3.0 that can manage thousands of threads in parallel, Huang said.
Fermi will succeed Nvidia's G80 GPU architecture, which was introduced last year and is used in the company's GeForce GT200 graphics cards, which include up to 240 processor cores.
Fermi also contains an ECC memory correction technology to shield bits of data from electromagnetic waves or particles from the environment that can affect thread execution.
The GPU will double the memory bandwidth of current architectures with support for GDDR5 DRAM memory, and support up to 1TB of GPU memory, Huang said.
The architecture also supports double-precision floating point, which could provide up to an eight-fold performance boost for certain scientific and mathematical tasks, Nvidia said.
And it will support C++ programming, in addition to the C programming of current designs.
The Fermi chips will also be compatible with Nvidia's CUDA development environment, which helps developers to write parallel code.
Also, Nvidia and Microsoft jointly announced Nexus, a development environment for the Fermi architecture that has been integrated into Microsoft's Visual Studio developer environment.