Nvidia on Monday announced a new graphics processor with 240 computing cores, giving PCs the horsepower needed to run three-dimensional games and scientific applications.
The new GeForce GTX 280, the largest GPU ever built by Nvidia, includes 1.4 billion transistors and delivers 933 gigaflops of performance. It succeeds the GeForce 8800 GTX, which had 128 cores and delivered 518 gigaflops of performance.
The GPU will bring a new level of realism to gaming, with better character detail and more natural character motion, said Jason Paul, senior product manager at Nvidia.
Users can leave general purpose computing to the x86 CPU and unload advanced processing to the GeForce GTX 280, which will take care of applications like video transcoding and surfing the Web using three-dimensional Web tools, Paul said.
"What you have today are great three-dimensional games [and] high-definition video playback, but now you'll see the GPU becoming the heart of the optimized PC. It's going to provide the second processor of the PC," Paul said.
The chip includes support for PhysX, a hardware and software engine that adds physical reality to existing games, like smoke billowing from an object after an explosion, or the behavior of a rock after it hits a target.
New to the GeForce GTX 280 architecture is support for double-precision, 64-bit floating point computation that can deliver the horsepower necessary to perform high-end scientific applications, Paul said.
Nvidia may paint this as a more powerful alternative to the CPU, but this is a specialized processor designed for high-performance computing applications like weather predictions and, in some cases, CAD programs, said Jon Peddie, president of Jon Peddie Research. "This is not a general-purpose replacement for x86 processors, this is a specialized processor to take apps originally running on an x86 and now they can run it on a GPU," he said.
People have had to run scientific applications on scalar processors like the x86 chip because they couldn't afford high-end vector processors like the ones found in IBM's supercomputers, Peddie said. With the new Nvidia chip, vector processors can now be built into a home PC, Peddie said.
"This is such an incredible technology. We can go out and buy a teraflop computer for under $1,000," Peddie said.
However, a vector processor should be limited to high-performance computing because x86 chips are capable of handling general-purpose computing, Peddie said. The GeForce GTX 280 GPU can be used as a high-end gaming machine chip or on workstations to process graphics-intensive applications.
Today's applications are mostly designed for x86 processors and will need to be recompiled to take advantage of GeForce GTX 280, Peddie said. Nvidia provides CUDA, a tool that allows programmers to design and port programs to run on the new GPU.
The GeForce GTX 280 is available for US$649.