: News
: Reviews
: Editorials

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




The GeForce 7600 GT


eVGA Superclock Style


by Josh Walrath


            The $179 to $249 range of the graphics market is the meat and potatoes of the industry.  Margins at this level are typically quite good as compared to the budget level, and the sell through with these products can help to make or break a company for that particular quarter.  Needless to say, this is a very important market in terms of monetary value as well as mindshare.  Late last year NVIDIA released the 6800 GS card to provide an alternative to the X800 and X1600 series of parts from ATI.  While it was fairly successful as such, it was basically a rehash of older technology given a speed boost.  It was a solid performing product, but it did not share any of the new and improved features of the GeForce 7 series of cards.

The cool, blue eVGA box doesn't even hint at the monster inside.

            Going back to August of 2004, NVIDIA introduced a part that they hoped would grab the midrange crown and keep it.  The GeForce 6600 series took that crown, and it has kept a tight hold of it through the years.  As time went by, the prices on these cards went down, but no matter what price level it was set at, the 6600 series consistently provided the best price to performance ratio of any product in its range.  The 6600 GT has today finally made it down to sub-$120 levels, and it has proven to be a long legged and stalwart part.  Coming up on nearly two years old makes it positively geriatric in this industry, and NVIDIA feels it is finally time to release the spiritual replacement of the 6600 series.

Transistors, Die Sizes, and Margins

            Keeping track of costs to produce a video card is of supreme importance, especially in the above described market.  If a company has to keep up with the competition by producing a large GPU die, as well as expensive memory and PCB costs, then margins will predictably be low for that product.  Companies want to move away from that particular paradigm, and they do so by utilizing advanced fabrication techniques, as well as good design decisions to keep production costs low.

            Let us use the 6800 GS as a good example.  This card sports a NV42 chip based on TSMCís 110 nm process.  This chip is comprised of approximately 198 million transistors running at a maximum speed of 425 MHz to 450 MHz, depending on the SKU.  The chip features 8 ROPS, 12 Pixel Shader Pipelines, 12 texture units, and 5 vertex shaders.  It also has a 256 bit memory controller (4 x 64 bit crossbar) attached to 256 MB of 1 GHz GDDR-3.  When the 6800 GS was introduced, it was priced around $249.  Throughout the last few months that price has dropped to $199.  While the NV42 is not exactly a new chip, and it is made on TSMCís value 110 nm process, it is still a large chip for its price range.  Add to that the extra PCB and chip packaging costs to support a 256 bit memory controller, the 6800 GS was starting to turn into a low margin part.  This is where the 7600 GT comes in.

eVGA leaves a little window open on the back so the buyer can see if the card inside the box matches what is printed on the outside.

            NVIDIA designed the 7600 GT to slot into the $179 to $249 range.  To be able to hit production cost targets, several important design decisions had to be made.  The first was to adopt a 128 bit memory controller, as that lowered the costs of both the PCB and the chip packaging.  It also allows the board to use four pieces of memory vs. the 8 that are required for 256 bit.  The next decision was what process this new chip should be made on.  The initial plan to use TSMCís 90 nm was actually a conservative one, even though at the time design work started on the chip TSMC had not yet produced a single 90 nm part.  About a year ago TSMC started to really churn out 90 nm parts, and on the graphics side ATI tried their hand with this process with disastrous results.  The problem ATI had was not with TSMCís 90 nm process, but rather a third party software error in the design stage.  TSMC had a clean process by early 2005, and NVIDIA knew that it would be quite mature by the time they required TSMC to fabricate the 7600 GT.

            The next question faced by the engineers at NVIDIA was how many functional units should the 7600 GT have?  With the success of the NV42 in a wide range of applications, it was thought that the same setup would be quite effective in terms of both performance as well as die size.  The 7600 GT has 8 ROPS, 12 Pixel Shader Pipelines, 12 texture units, and 5 Vertex Shaders.  These engineers then decided to spice things up a bit.

            The 7600 GT would of course embrace all of the GeForce 7 features including transparency AA and the dual issue MADD ALUís in the pixel shader pipelines.  The interesting part in all of this is that the engineers went through the design with a fine toothed comb and removed transistors while still keeping all the features and performance.  The school of thought on transistor design is that to speed up a GPU, more pipeline stages are needed to lessen the propagation delay between stages, thereby making the design faster.  But the more stages there are, the greater the transistor usage.  NVIDIA apparently is using a much faster transistor, as well as some custom logic, to make the 7600 GT very fast, yet at the same time cutting down the pipeline stages and thereby reducing the transistor count.


Next: More 7600 GT


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


Copyright 1999-2006 PenStar Systems, LLC.