Final Words

When Valve and ATI came together to show us the first inklings of Half Life 2 performance last year, it did not look pretty for NVIDIA.  NVIDIA’s highest end card at the time, the GeForce FX 5900 Ultra, could not even outperform a Radeon 9600 Pro in most tests – much less anything from ATI at its price point.  Even though we haven’t shown it here (that’s coming in Part II), the situation has not changed for NVIDIA’s NV3x line of GPUs – they still must be treated as DirectX 8 hardware, otherwise they suffer extreme performance penalties when running Half Life 2 using the DirectX 9 codepath.  To give you a little preview of what is to come, in DirectX 9 mode, the GeForce 5900 Ultra offers about 1/3 of the performance of the slowest card in this test.  If you’re unfortunate enough to have purchased a NV3x based graphics card, you’re out of luck with running Half Life 2 using the DX9 codepath (at any reasonable frame rates). 

What we were missing from looking at Half Life 2 performance a year ago was the release of NVIDIA’s NV4x line of GPUs, which have effectively “saved” NVIDIA from delivering embarrassing performance under Half Life 2.  In fact, NVIDIA’s GeForce 6 line of GPUs actually runs Half Life 2 extremely well, even when pitted up against equivalently priced competition from ATI. 

Our final Head to Head comparisons revealed a few interesting things:

The GeForce 6800 Ultra performs very similarly to the X800 XT as long as antialiasing and anisotropic filtering are disabled.  With those two features enabled, the X800 XT begins to show a performance advantage that is truly seen at 1280 x 1024 and 1600 x 1200 with 4X AA enabled.  If you are running with AA disabled, the two GPUs perform very similar to each other.  It is only at 1600 x 1200 that the performance becomes somewhat noticeable between the two, as the X800 XT averaged 8% faster than the 6800 Ultra.  However, turning on antialiasing and anisotropic filtering gave the X800 XT between a 4 – 20% advantage depending on resolution, which definitely isn’t shabby. 

At the $400 price point, the X800 Pro and the GeForce 6800GT are basically equal performers in all of the resolutions we tested (regardless of whether or not AA/aniso was enabled).  So the recommendation here goes either way, look at the performance of the cards in some of the other games you play to determine which one is right for you. 

If you’re spending $200 - $300 you’ve got three choices for PCI Express graphics cards, and one for AGP.  The NVIDIA GeForce 6800 is 12-pipe underclocked version of the 6800GT/Ultra and currently sells for close to $300, however in Half Life 2 the performance of the regular 6800 is not any better than the cheaper 6600GT, thus making our NVIDIA recommendation clear.  But how does the 6600GT stack up to the X700 XT?  The two GPUs are basically equal performers under Half Life 2, although the X700 XT is faster with AA enabled. If you need an AGP card however, then the 6600GT AGP is your only option (and far from a bad one at that).

We’ve left a number of questions unanswered here today involving older/slower hardware, so be sure to check back for part II of our Half Life GPU comparison to find out how well older hardware performs under Valve’s amazing game.  Thanks for taking a break from playing Half Life 2 to read this, now get back to it…

Head to Head: NVIDIA GeForce 6800 vs. NVIDIA GeForce 6600GT
Comments Locked

79 Comments

View All Comments

  • araczynski - Thursday, November 18, 2004 - link

    yawn, i'm too busy enjoy the game (6800gt) to read the article and/or care which card is better :) i'm playing at 1600x1200 0AA/4AF (2.4@3.3/1GB) and have absolutely no complaints, other then knowing that the game will eventually end :(
  • Jeff7181 - Thursday, November 18, 2004 - link

    #57... poorly :)
  • blckgrffn - Thursday, November 18, 2004 - link

    I would also like to to see how the 9200/9000 series Radeons perform too, and if you have extra time, the 8500/9100.

    Again, Thanks!
  • Jeff7181 - Thursday, November 18, 2004 - link

    #16... that's correct, although the only REAL observation that needs to be made is that Half Life 2 makes heavy use of pixel shaders which is very GPU dependant, and GPU's are just now growing the required testicles to process those shaders :)
  • blckgrffn - Thursday, November 18, 2004 - link

    Anand -

    I would like to see how the 6600 performs. As an 8 pipe card, it should perform better than the 9600xt and a little under a 9700 Pro, but it would be interesting to see if that is true. It is a great budget PCIe card along with the x700.

    Thanks!
    Nat
  • eva2000 - Thursday, November 18, 2004 - link

    nice review downloaded your demos to run on AMD64 3700+ @ 12x 222 = 2664mhz with 1GB BH-5 @ 222mhz 2-2-2-6 1T and Powercolor X800XT PE @ 520/560 and all demo results were within 3-4fps of the reviews :)
  • Live - Thursday, November 18, 2004 - link

    Good reading as always. Would like to see minimum FPS tough. I find it very important to see how low the cards drop when stressed. You can't see that with only average FPS.
  • housecat - Wednesday, November 17, 2004 - link

    So... wheres the Nvidia SLI versus ATI results??

    Muwahahaha.
  • Avalon - Wednesday, November 17, 2004 - link

    Hey Anand, I have an interesting request. Could you try Rivatuner on your 6800, unlock its pipes, and then bench it again? :P
    Just kidding. Actually, I'm glad you've confirmed what I've been thinking...that AF hasn't been doing much for me. Since I'm running on a lowly 9700, I think I'll just turn it off now, and enjoy a nice speed boost.
  • PrinceGaz - Wednesday, November 17, 2004 - link

    How about throwing a GF 5600 and maybe even a GF 5200 in as well for part 2, as an awful lot of people have them. Ultra versions of either if you prefer.

    I don't have one of them myself as I'm still using a Ti4200, but it would be interesting to see how they stacked up in the DX8 codepath against the Ti4600 you are planning to test. And it should be worth a giggle to see just how "fast" the 5600 or 5200 can manage the DX9 codepath :)

    Thanks to the resolution scaling-graphs this review included and how the fastest cards were generally CPU limited with that A64 4000+ when the resolution was dropped to 1024x768, I'm not sure how much a CPU scaling article for part 3 will show that can't already be quite accurately guesstimated from how different CPUs generally tend to perform in games. But a comparison of the Athlon 64 4000+, against an Athlon XP, a Prescott, a Northwood, and if time permits a fast P3, Duron and Celeron also, would be great.

Log in

Don't have an account? Sign up now