: News
: Reviews
: Editorials

: Archives
: About
: Contact
: Advertising
: Privacy
: Links
: Forums

 

 

 

NVIDIA’s GeForce 7800 GTX

 

Walk Softly…

 

by Josh Walrath

            April 2004 was the introduction of NVIDIA’s latest architecture, the NV40.  This little number brought NVIDIA out of the dark ages that started with the now infamous NV30, aka GeForce 5800.  While NVIDIA improved the overall performance of the NV3X series with the NV35 and its derivatives, they still lost a lot of marketshare to ATI and their excellent R3x0 based products.  Now we are finally at the point where NVIDIA takes its next step forward.

            It is hard to believe that it has been almost 15 months since the release of the GeForce 6800 series.  NVIDIA in that time has released the 6600 and 6200 series of products based on the architecture, but the top end products have never received a refresh after all of this time.  Typically 6 to 8 months after the initial architectural release we see a refresh of the high end and midrange parts.  This was not the case with the NV40.  From sources around the industry, NVIDIA came very close to producing a refresh part, but in the end found that it was not truly economically viable to release a new high end part that would not dramatically improve the performance over the current 6800 Ultra (though it was rumored to be around 10% faster overall, and featured the fully functional video portion that the NV41/42/43/44 chips boasted).

            ATI released their X800 series of cards a month after NVIDIA released the 6800s, and ATI had a bit of a rough time with those chips.  While the X800 Pro’s were in strong supply, the X800 XT and PE editions were not so lucky.  Due to what appears to be a lack of overall availability of TSMC’s 130 nm Low-K process and lower than expected yields, ATI was unable to provide as many chips as the market demanded.  After a metal layer revision, ATI released the X850 series of chips in December 2004 which boasted improved clock speeds and greater availability.  Still, ATI’s refresh was not as significant as previous refreshes in the past.

            As GPUs become more complex, and the underlying process technologies not advancing as fast as they once did, we can expect to see longer and longer times between major architectural transitions and refreshes.  NVIDIA used to be on a 6 to 8 month refresh with a major transition every 12 to 16 months.  Now we are seeing these times starting to lengthen out.  This brings up a good question though… is the 7800 GTX a major transition or merely a refresh?  The answer to this is not so easy, but I think it is a mixture of both. 

The G70 Chip

            The G70 chip is the beating heart of the 7800 GTX video card.  This 302 million transistor part is fabbed on TSMC’s 110 nm process.  The 110 nm process is a size optimized process, and has the performance characteristics of TSMC’s 130 nm FSG process.  This basically means that more die can be placed on a wafer as compared to the 130 nm process, but the transistor performance should be nearly identical to the 130 nm process.  I have the distinct impression that the engineers at TSMC have added some performance features to their 110 nm process, as it does appear to have superior speed and power characteristics over the 130 nm process.

I actually thought this was one of the more fascinating pictures of the G70.  Here we can see the bottom of the flip chip, with all of the hundreds of solder balls for making contact with the substrate (which then mounts to the PCB of the video card itself).

            One would expect a GPU which has approximately 77 million transistors more than its predecessor and runs 30 MHz faster would have a greater power draw and heat dissipation.  This is not true.  The minimum power requirements for the 7800 GTX is actually 50 watts lower than the 6800 Ultra (350 watt PS vs. 400 watt PS), and the power requirements for 7800’s in SLI are identical to that of the 6800 Ultras.  By all current thinking, the G70 should be a hotter running chip that requires more power.  In between TSMC’s process and NVIDIA’s engineering talent, they have kept the thermal and power envelope pretty stable from last generation to this one, all the while achieving this without a major process node change.

 

Next: More G70

 

If you have found this article interesting or a great help, please donate to this site.

 

Copyright 1999-2005 PenStar Systems, LLC.