GeForce 7900 Battle: GT vs. GS

With equivalent stock clock speeds and potential 14% and 20% advantages in vertex and pixel processing respectively, we have the opportunity to better understand how G71 scales with parallelism. The following graphs are percent increases in performance from the 7900 GS to the 7900 GT for teach game at each resolution tested.

800 x 600 1024 x 768 1280x1024 1600x1200 1920x1440

The 7900 GT shows at most a 15% performance improvement over the 7900 GS in our testing. The largest performance improvement is under Oblivion at a relatively low resolution. Most of the biggest improvements seem to appear at 1024x768. As we increase resolutions, we become more bandwidth limited, and with both cards offering the same amount and speed of memory the scores begin to tighten up. Interestingly, Half-Life 2 shows zero performance difference between the GS and GT at higher resolutions. These cards are by no means CPU bound under such conditions, so we seem to have run into a bandwidth bottleneck.

The Test XFX 480M Extreme vs. Stock Performance
Comments Locked


View All Comments

  • phusg - Tuesday, September 12, 2006 - link

    Hi Derek,

    I'm a little late to the ball but still

    > cheaper price tag

    really grates me! I know it's pretty endemic but it's still logically incorrect. A price tag can be lower of higher, but not cheaper, unless it's the price tag being sold. It's the product itself that can be cheaper.

    Cheers Derek and don't let me catch you making this one again or there'll be hell to pay ;-)

  • imaheadcase - Thursday, September 7, 2006 - link

    Could you post a link to the bf2 demo you use, so we can compare are systems video cards to new ones?
  • Stele - Wednesday, September 6, 2006 - link

    At first glance, it seems that ATI has markedly improved their OpenGL implementation, at least for the Doom 3 engine:

    ...the latest ATI OpenGL enhancements that have drastically improved Doom 3 engine based game performance.


    ...clench Quake 4 as a benchmark that greatly favors ATI hardware when running at the highest possible quality settings. This is the exact opposite of what we have been saying about Quake 4 performance ever since the game launched....

    However, after a moment's thought considering the vast difference in performance from before, and also the following qualifiers:

    Of course, not all OpenGL games faired well with the latest round of drivers from ATI, with City of Heros/Villains performing very poorly in spite of its use of OpenGL.


    ...but it seems ATI has finally solved their OpenGL performance issues -- at least with this particular engine.

    one can't help but wonder - just wonder - if there's anything here that smells like the last quake.exe driver optimisation trick ... which, curiously enough, was also pulled by ATi (iirc it was during the Radeon 8500's time?). I wonder!
  • Ryan Smith - Wednesday, September 6, 2006 - link

    There's no quackery as far as we know of. The problems with City of Heroes is a shader corruption bug, and a bug related to rendering on a secondary buffer, according to Cryptic(the developers of CoH). Whatever ATI did to speed up OpenGL performance here, they apparently didn't take in to account CoH.
  • Stele - Thursday, September 7, 2006 - link

    Excellent! Am deciding between the X1900GT and 7900GS (when the latter shows up in the channels), and this improvement would help strengthen the case for the X1900 a bit. :)
  • S3anister - Wednesday, September 6, 2006 - link

    found an XFX version on this card on newegg for 189MIR.">
  • emilyek - Wednesday, September 6, 2006 - link

    A worthless sku. x1900gt and x1800xt/gto2 are better and almost $50 cheaper.
  • sharkdude - Wednesday, September 6, 2006 - link

    The Oblivion percentages are the same in this graph as in the graph on page 4 for all resolutions when in fact only the 800x600 numbers should be the same. On page 5 the numbers should be 4.1%, 10.1%, 6.4%, and 7.3% for 800x600, 1024x768, 1280x1024, and 1600x1200. Note the text below the chart should also change 15% to 10%.
  • DerekWilson - Wednesday, September 6, 2006 - link

    corrected -- but your number for 16x12 appears to be wrong as well. :-)
  • Lifted - Wednesday, September 6, 2006 - link

    Thanks for including the 6600 and 6800 cards in the benchmarks.

Log in

Don't have an account? Sign up now