NVIDIA's upcoming DirectX 10 part has some rather flabbergasting specifications, if the rumours are correct.

The launch flagship GeForce 8800 GTX, whose core clock is 575 MHz, will apparently have 128 unified Shaders clocked at 1350 MHz. According to my calculations that's an awesome peak pixel fillrate of 172.8 Gigapixel/s (other sources seem to disagree) vs. the current fillrate champion Radeon X1950 XTX's 31.2 Gigapixel/s. That's still 180% faster than a Radeon X1950 XTX Crossfire setup.

Much more significantly, this should translate into a absolutely stunning peak transformation rate of 57.6 GVerticies per second, vs. the mere 2.6 GVerticies achieved by the Radeon X1950 XTX.

But the jaw-droppers don't stop there. The G80 will also come equipped with 768 MB of GDDR3 attached to a 6-channel memory controller clocked at 900 MHz. That's 384-bit/1800 MHz in marketing speak, or 86.4 GB/s, compared again to spec-king X1950 XTX's 64 GB/s as provided by a 2 GHz 2-channel (256-bit) GDDR4

The junior model, GeForce 8800 GTS brings back the old nonclemature from the GeForce 2 days. Originally, GTS meant "Gigatexel Shader" - the GeForce 2 GTS was the first graphics card capable of fillrates in excess of 1 Gigatexel. Now, it's means a 500 MHz Core clock, 96 Unified Shaders clocked at 1200 MHz and 640 MB of GDDR3 on a 5-channel 900 MHz bus (320-bit, 1800 MHz). This gives it a raw fillrate of over 115 times that of the original GTS.

Daily Tech reports.


Unified Shaders are part of the Shader Model 4.0 specified in DirectX 10. In the Shader Model 3.0 and earlier model, texture operations are handled by Pixel Shaders, while vertex operations are handled by Vertex Shaders. In most DirectX 8 compliant graphics controllers and later, there are approximately a third the number of Vertex Shaders as there are Pixel Shaders. This makes fairly good sense in terms of silicon real estate, as Pixel Shaders perform much more complicated algorithms requiring more resources.

A single Unified Shader can perform both Pixel and Vertex operations, in addition to the newly defined Geometry operations, which facilitate the manipulation of entire geometric objects (as opposed to just individual vertices as per Vertex Shader). This radically improves a graphics accelerator’s ability to load balance, but also dramatically improves the accelerator’s peak geometry processing capabilities. As a direct result, objects with poly counts many fold higher will be feasible.