updated 08:05 pm EDT, Fri March 26, 2010
GeForce GTX 480 and 470 finally official
NVIDIA at PAX East tonight finally released the first video chipsets based on its Fermi architecture. The top-end GeForce GTX 480 leads the group and is billed as the "fastest GPU in the world:" it has 480 visual processing cores, 16 geometry units and four raster units that combined should beat the ATI Radeon HD 5800 series in real-world tests. It also has major optimizations to multi-card SLI that produce a 90 percent speed boost with a second card, making the case for multiple GeForce 400 series cards in high-end systems.
Both it and the more modest GTX 470 are NVIDIA's first cards to support DirectX 11 (and eventually OpenGL 4.0) and the accompanying benefits, including hardware geometry tessellation. Fermi's much more generalized design should also help significantly for general-purpose computing in CUDA, DirectCompute and OpenCL. Either new GeForce card is powerful enough to handle raytraced lighting and realistic hair in real-time under certain conditions.
3D Vision has been given a boost as two or three cards can drive as many as three displays with 3D active.
The GTX 480 at $499 brings a main 700MHz core clock speed, 1.5GB of GDDR5 memory clocked at 1.85GHz and a high 250W power draw. NVIDIA scales the GTX 470 back considerably to a 607MHz core and 1.2GB of 1.67GHz GDDR5 memory, but it costs a considerably lower $349 and carries a slightly less demanding 215W energy requirement. Boards based on the technology should be available from third parties as of today in limited quantities. Third-party PC builders such as CyberPower have already signed onboard to carry the faster video hardware.
GeForce GTX 470
Hair and raytracing demos