Specifcations

Building the GeForce 6800 GS

With manufacturing of the GeForce 7800 GT and GTX going along so well, the high-end card of yesteryear, NVIDIA’s GeForce 6800 Ultra, is no longer in production. In fact, NVIDIA’s board partners have told us that production of the GeForce 6800 and 6800 GT just recently came to an end as well (although production may be over, these cards can still be found in the sales channel, but once supplies dry up these boards will be gone forever). Instead, NVIDIA has turned to an entirely new process for the GeForce 6800 GS – TSMC’s tried and true 110-nanometer manufacturing process.



This is the same manufacturing process NVIDIA has used for a string of products. The GeForce 6200 and 6600 family are both built on TSMC’s 110-nm process, as well as the high-end GeForce 7800 GT and 7800 GTX. NVIDIA’s previous manufacturing partner for the GeForce 6800 family was IBM, whose 130-nm process was used on NV40 (GeForce 6800 Ultra/GT), NV45 (PCI Express variant of the GeForce 6800 Ultra/GT), as well as NV41 (PCI-E native version of GeForce 6800). With the introduction of the GeForce 6800 GS, NVIDIA’s essentially moved all their production back over to TSMC.

Like NV41, NVIDIA’s NV42 GeForce 6800 GS GPU is built entirely from the ground up with 12 pixel pipelines and 5 vertex units. (The GeForce 6800 Ultra and GT both feature 16 pixel pipes and six vertex units.) This means that like NV41, you won’t be able to unlock any additional pixel or vertex pipelines on GeForce 6800 GS – they simply aren’t there.

Thanks to the smaller manufacturing process, NVIDIA can cram the GeForce 6800 GS’ transistors into a smaller die area. This also helps to reduce manufacturing costs, as NVIDIA/TSMC yield more chips per silicon wafer. With fewer functional units inside, transistor count is reduced from 222 million in 6800 Ultra/GT to 202 million in GeForce 6800 GS (in comparison, the X1600 XT contains 157 million transistors while NV41 contained 190 million. According to NVIDIA the difference between transistor counts in NV41 versus the newer NV42 “is related to different libraries and manufacturing processes between IBM and TSMC”). Although the 6800 GS’ NV42 die is a little larger than X1600 XT (and thus costs a little bit more for NVIDIA to produce than the X1600 XT for ATI assuming equal yields), thanks to its smaller process, it’s still the least expensive to manufacture GeForce 6800 variant NVIDIA has produced to date. As a result, the GeForce 6800 GS carries an MSRP of $249, that’s $51 lower than the GeForce 6800 when it was introduced.

Besides lower production costs, another added benefit of TSMC’s 110-nm process is that it has proven to scale ridiculously well to high clock speeds. Even massive 300+ million transistor chips like the GeForce 7800 GTX have hit clocks of over 500MHz on retail graphics cards, while the less complex GeForce 6600 family has scaled well also.

The GeForce 6800 GS is no exception, as NVIDIA clocks the graphics core at 425MHz, that’s 100MHz higher than the GeForce 6800. Meanwhile the memory subsystem runs at 500MHz (1.0GHz effective), giving the GeForce 6800 GS the same 32GB/sec of peak memory bandwidth found on NVIDIA’s GeForce 6800 GT. This is an improvement of 200MHz over the 256MB GeForce 6800, which shipped with 300MHz memory (the memory on 128MB GeForce 6800 cards ran at 350MHz). Also like the GeForce 6800 GT, the GeForce 6800 GS ships with 256MB of GDDR3 memory.

This chart summarizes the key features found in NVIDIA’s GeForce 6800 GPUs:

GeForce 6800 Series Comparison
	GeForce 6800 Ultra	GeForce 6800 GT	GeForce 6800 GS	GeForce 6800
Core Clock (MHz)	400	350	425	325
Memory Speed (MHz)	550	500	500	300 (256MB) 350 (128MB)
Memory Interface Width	256-bit	256-bit	256-bit	256-bit
Memory Size & Type	256MB GDDR3	256MB GDDR3	256MB GDDR3	256MB DDR1, 128MB DDR1
Memory Bandwidth (GB/sec)	35.2	32	32	19.2 (256MB) 22.4 (128MB)
Texel Fill-Rate (Gtexels/sec)	6.4	5.6	5.1	3.9
Vertices/sec (million)	600	525	531.25	406
Pixel Pipelines	16	16	12	12
Vertex Units	6	6	5	5