When we first found out about NVIDIA's NV10 (as it was called then) way back at Comdex '98, we were extremely excited by the prospect of this new architecture. By WinHEC this year it became apparent that NVIDIA's NV10 engineers would debut an on-board graphics co-processing unit for hardware transform and lighting (T & L) within their architecture to offload some of the calculations that would normally bog down a CPU. These new features have certainly caught the imaginations of both gamers and game developers.
By having what NVIDIA likes to call a 'GPU' take over most of the complex number crunching involved within the graphics pipeline, game developers are allowed the opportunity to employ the reclaimed CPU horsepower for a host of new features and enhancements while still maintaining high frame rates in complex environments. We're talking more elaborate AI, better physics models, visual enhancements like cube environment mapping, vertex blending, as well as more detailed environments and characters.
A Labor Day weekend US launch was originally planned for the first retail boards, but unfortunately the date slipped back until the middle of October but it's here now and we've put together this piece to help decipher what's what in terms of this new animal, especially the DDR-SGRAM version.
The NV10 code name has since been replaced by 'GeForce 256' and is fairly simple in its meaning. The "Ge" is short for 'geometry' (not General Electric as we joked about last time) and the 256 is a reflection of NVIDIA's new 256-bit architecture (the TNT2 was 128-bit). The 256-bit architecture (not memory bus width) is certainly a step in the right direction and in layman's terms refers to a QuadPipe Rendering engine. There are four 64-bit rendering pipelines (4 x 64 = 256) that together are able to process four 32-bit pixels per clock cycle, which is double the TNT2's capability.
The 64-bit pipelines can render in 32-bits each (RGB and 8 bits of Alpha), with a 24-bit Z Buffer and 8 bits of stencil buffering (great for those shadows in Quake 3 or extra ground detail in flight sims). In a nutshell, the GeForce 256 has more 'muscle' than its predecessor does, OK? The NVIDIA QuadPipe architecture may have a slight technical disadvantage when compared to S3's Savage 2000 Quad Texture engine. When multitexturing, a GeForce 256 can render a single texture on four pixels per clock or the same two textures on each of two pixels per clock, whereas the Savage 2000 will be capable of these functions in addition to applying four textures per pixel per clock cycle. Exactly how useful this feature will be is still up in the air so we'll cross that bridge when we come to it… right now multitexturing support in OpenGL games such as Quake2/3 is for two textures per pixel.