NVIDIA is stressing with the new GTX 200 family that the GPUs go beyond
gaming and are one of the most powerful processors in a PC and can be used for
rendering video and other functions. NVIDIA says that its goals with the architecture
of the GTX 200 series were to design a processor twice as powerful as the
GeForce 8800 GTX, rebalance the architecture for future games with more complex
shaders and more memory, improve efficiency per watt and per square millimeter,
provide enhanced CUDA performance, and add a significant reduction in idle
NVIDIA says that the GTX 200 line provides nearly a teraflop of
computational power. The GTX 200 family also offers support for PhysX powered
physics processing right on the GPU. Both the new GTX 280 and GTX 260 support
SLI and 3-way SLI. The previous NVIDIA 9800 GX2 could not support 3-way SLI.
Key features in the new GTX 200 GPUs include support for three times the
number of threads per flight at any given time. A new scheduler design allows
for 20% more texturing efficiency. The memory interface for the GPUs is 512-bit
(GTX 280) and full-speed, raster-operation (ROP) frame blending is supported.
The GTX 200 series also features twice the number of registers for longer and
more complex shaders and IEEE754R double precision floating-point. The GTXC 200
line also supports 10-bit color scan out via the DisplayPort only.
One of the main goals with the GTX 200 line was improved power management.
Both the GTX 200 series GPUs have idle power requirements of about 25W; during
Blu-ray playback power requirements are around 35W; full 3D performance
requirements vary with the most power needed being 236W (GTX 280). The GTX 200
line is compatible with HybridPower, which makes the power needs of the GPU
The GTX 280 is built on a 65nm process and has 1.4 billion transistors. The
stock video cards have a graphics clock of 602 MHz, processor clock of 1,296
MHz, and a memory clock of 2,214 MHz. The GTX 280 has 1GB of GDDR3 and 240
processing cores and 32 ROPs.
The GTX 260 is also built on the 65 nm process and has 1.4 billion
transistors. The graphics clock for the GTX 260 is 576 MHz, the processor clock
is 1,242 MHz, and the memory clock is 1,998 MHz. The memory interface on the
GTX 260 is 448-bit and it has 896MB of GDDR3 memory. The GTX 260 has 192
processing cores and 28 ROPs. The maximum board power is 182W.
Both video cards will support PhysX processing on the GPU. NVIDIA purchased
Ageia in early 2008.
GTX 280 video cards will be available tomorrow for
$649 and the GTX 260 cards will be available on June 26 for $399.
quote: However, the performance gains that were observed in the retail version are inaccurate since the implementation was wrong and a part of the rendering pipeline was broken.
quote: Unfortunately, our original implementation on DX10.1 cards was buggy and we had to remove it.