Even at the launch of GeForce4 last February, with NVIDIA’s insistence of being the first to market with DirectX9 hardware, NV30 has been one of the most hotly anticipated graphics chips for a long time, with the hardware community lapping up every rumour or detail available.

In July/August, roughly when we’d begin to expect hearing some details on NV30, ATI threw everyone for a curve with the introduction of the Radeon 9700 PRO, with ATI touting DirectX9 support (long before the API’s availability) and benchmarks showing areas with a greater than twice the performance difference over previous generations of hardware. While a greater than two times performance has been achieved before from one generation to the next, with 3D accelerators as powerful as they are it was quite a surprise to see it again. The talk then became of how much faster NV30 would be over the Radeon 9700 PRO.

During the period of the Radeon 9700 PRO’s release and availability, NVIDIA’s investor conference calls, whilst painting a relatively healthy picture for the company in the face of a tech sector slump, made for difficult listening for those keen to glean details on NV30. A confused picture was painted as to whether the chip had or hadn’t taped out (leading to the numerous discussions over the definition of ‘tape-out’!); however, the company line was firm: this will be NVIDIA’s fall product. At least, that was until the last quarter earning in which the CEO admitted disappointment in missing the holiday period.

So, while it seems unlikely that we'll see many NV30 based products before Christmas, today we do get to learn more about the products based off the CineFX architecture, henceforth to be formally introduced as GeForce FX.

GeForce FX Overview

The NV30 based GeForce FX will be the first series of products based on the CineFX architecture. As we’d expect, GeForce FX has a considerable amount more raw power than previous generations. Here’s a rundown of the raw performance elements of the architecture:

  • 8 Pixels Per Clock – providing, on a clock for clock basis, twice the pixel performance of GeForce 4 Ti, current games can be rendered faster (CPU willing) and allows more flexibility in future applications for increased per pixel operations whilst still maintaining playable frame rates.
  • 1GHz DDR-II Memory Interfaces – With increased pixel performance so too is increased bandwidth important. NVIDIA is maximising the bandwidth by utilising the latest available high speed memory chips provided by the DDR-II memory specification.
  • NVIDIA Intellisample Technology – As with some of the previous generations of GeForce, GeForce FX implements Z compression algorithms to reduce the Z buffer bandwidth and attain greater performance. GeForce FX also implements colour compression, thus further reducing the overall frame buffer bandwidth utilisation. The biggest performance benefit will be with FSAA enabled.
  • AGP 8X Host to Graphics interface – GeForce FX utilises the latest AGP 3.0 specification to get the highest possible transfer rates between the host system and graphics controller, allowing for greater geometry and texture transfer rates.

Here's a quick comparison table of GeForce 4 Ti to GeForce FX:

DDR Memory @ ~ 600MHZ DDR-II Memory @ 1GHz Nearly doubles the raw peak bandwidth to local memory
4 Pixels per clock 8 pixels per clock Twice the pixel fill rate available
Accuview Antialiasing Intellisample Doubles efficiency of memory Read/Writes
AGP4X AGP8X Doubles transfer to host

The compared memory rates between GeForce4 and GeForce FX are consistent with GeForce FX utilising a 128bit memory bus as opposed to a 256 bit bus utilised on recent boards such as the ATI Radeon 9700 (PRO), Matrox Parhelia, and 3Dlabs Wildcat VP.