AA Comparison

And now the fun part: playing around with images. Certainly everyone has their own taste when it comes to AA, but we've cropped and blown up this 800x600 screenshot from Oblivion in order to better show what's really going on. As resolution increases and pixel size decreases, the impact of higher AA modes also decreases. This is useful to keep in mind here.

A few key points to check out: compare the interior of textures between either no AA image and any of AMD's tent filters. Notice how the detail on interior textures is significantly decreased. It can be quite frustrating to enable a high anisotropic filtering level to increase the detail of textures only to find them blurred by your AA mode. Also, note how NVIDIA's 8x CSAA and 16x CSAA modes only subtly change some of the pixels. This is because CSAA actually attempts to better understand the actual geometry that a pixel covers rather than going around looking for data outside the pixel to bring in.

These screenshots are with gamma correction enabled on NVIDIA hardware in order to give the best comparison with RV770 which does not allow us to disable gamma correction. We do prefer disabling gamma correction for the average case and especially for anti-aliasing thin lines.

Click the links in the table below to change the AA images displayed


AMD RV770 No AA

AMD RV770
NVIDIA GT200

Click here to download all the full resolution, uncompressed images used in this comparison

Fixing AMD's Poor AA Performance The Test
Comments Locked

215 Comments

View All Comments

  • NullSubroutine - Wednesday, June 25, 2008 - link

    It scaled more than 100% in a few games?
  • DerekWilson - Wednesday, June 25, 2008 - link

    greater than 100% scaling is due to margin of error combination for both single card and dual card tests in the vast majority of cases.

    we also tested single card performance on an nvidia system and crossfire performance on an intel system, so the different computers will also add margin of error.

    two card solutions generally don't scale at greater than 100% except in extraordinarily odd situations (where rebalancing loads might help with scaling on both individual cards -- but that's odd and rare).
  • Sind - Wednesday, June 25, 2008 - link

    Why no 260 and 280 SLI?
  • ImmortalZ - Wednesday, June 25, 2008 - link

    Because, with that kind of money, one can an entire system with one 48xx :P

    Also, page 10 appears to be broken.
  • Lifted - Wednesday, June 25, 2008 - link

    No 260 or 280 SLI in the benchmarks, but they included them in the power charts. Odd.
  • Anand Lal Shimpi - Wednesday, June 25, 2008 - link

    The power data was simply taken from the GTX 280 review, we just added to the list.

    As for the GTX 280 SLI numbers, we didn't include them as it it's mostly out of the price range of the Radeon HD 4870 ($1300 vs. $600 for two 4870s). We can always go back and redo the graphs to include them if you guys would like, but in the interim I would suggest looking at the GTX review to get comparison numbers.

    Take care,
    Anand
  • DerekWilson - Wednesday, June 25, 2008 - link

    we actually only have one GTX 260, so we can't test that
  • Clauzii - Wednesday, June 25, 2008 - link

    Yes, and the click to enlarge doesn't work.
    And believe it or not, posting right now from a AT page that looks like 1994...!
  • ImmortalZ - Wednesday, June 25, 2008 - link

    Insert a buy in there. Need edit!
  • TonyB - Wednesday, June 25, 2008 - link

    but can it play crysis?

Log in

Don't have an account? Sign up now