IntroductionNVIDIA’s fourth generation graphics architecture has proven to be a very successful one for them. The first available graphics architecture to support DirectX Shader Model 3.0, the generation of parts that started with NV40**, and is finishing with G70** and friends as NVIDIA forge onwards towards generation five, has spawned SKU after successful SKU, both desktop and mobile.
Support for Shader Model 3.0 has been paired, on PCI Express at least, by NVIDIA’s SLI** technology, allowing for another add-in board sharing the same GPU to take part in the rendering process and accelerate performance and image quality.
Recent dalliances with 110 and 90nm process technology have spawned mobile parts with excellent performance-per-watt stats, along with all the technology features that the desktop parts, well covered by us on HEXUS.core, are known for and sport.
A largely redesigned fragment shader core, reputedly courtesy of ex-3dfx engineer Emmett Kilgariff, fine per-quad texturing ability and (plentiful) full-featured vertex shader units have seen them expand what started with NV40 into G70 in recent times. So G70 takes the best bits of what debuted with NV40, adds more of them across the board, and further enhances what the fragment hardware can do, to create the current performance champion when clocked at 550MHz (or more**) and paired with half a GiB of very high-speed GDDR3 memory.
However, while GeForce 7800 GT, GTX and GTX 512 get power from the G70 GPU, there’s still a large market for NV4x-based products as NVIDIA hold back on a top-to-bottom 7-series transition that they’ll likely make in the New Year. Money talks, as millions of consumers stay away from those £200 or more products, preferring instead to shop where it’s more fiscally attractive. NVIDIA have traditionally offered well in that segment, GeForce 6600 GT and the standard 6800 dominating for months on end.
However, with NV45 shelved in recent times, its 130nm roots too expensive for NVIDIA to bother ordering from its fab’ partners in the small-ish quantities the market is asking for, NVIDIA has turned to one of its 110nm NV4x parts to fill a gap vacated by the NV45-powered GeForce 6800 GT and Ultra SKUs.
Clocked attractively and sporting three fragment quads (12 ‘pixel shader’ units), 5 vertex shader units and 8 render output units for writing pixels on the screen, NV42 is the basis for GeForce 6800 GS.
The 202 million transistor chip is essentially just the 110nm optical shrink of the 130nm NV41, with both sporting the same basic 5/12/8 configuration. So, the GS is designed to fill GeForce 6800 GT’s boots, sitting next to it in terms of performance, but sitting proudly as what’s on store shelves, in the absence of the trailblazer it replaces.
And with whispers abound of an AGP version to satiate users of that older graphics interconnect, users that didn’t manage to snag a 6800 GT or Ultra while they were around on AGP in the beginning, GeForce 6800 GS looks like quite the stocking filler for some this coming holiday season. Join us as we take – an admittedly belated – look at NVIDIA’s last NV4x gasp on the desktop before it rolls out a fuller family of GeForce 7-series SKUs in 2006.