DirectX 9 Performance Impact

Now that you've seen what improvements Half Life 2's DX9 path can give you, let's take a look at the price to pay for some of those impressive visual effects. In order to measure the impact of the DX9 path we did the following: ran benchmarks using both the DX8 and DX9 paths, then took the percentage decrease in performance seen by going to DX9. We then averaged the percentage decrease across all five of our custom Half Life 2 benchmarks, per card, per resolution. We will look at actual performance numbers shortly, but this is just to give you an idea of what's to come:

At 800 x 600 the game is mostly CPU bound on cards like the Radeon 9600XT, thus the performance drom from DX8 to DX9 is quite small. Even on cards like the X300 and the Radeon 9550 the performance hit isn't too bad at less than 20%. But here's the kicker, the GeForce 5900XT sees almost a 60% drop in performance by going to DX9 mode. This type of a performance drop should be relatively consistent across the entire NV3x line (e.g. FX 5900 Ultra, FX 5600, etc...).

Half Life 2 DirectX 9 vs DirectX 8 Performance Penalty

At 1024 x 768 now all of the GPUs are in double digit performance losses, but even the GeForce 6200 with its 25% performance hit is nothing compared to the 5900XT which incurrs a 65% performance hit when going to DX9.

Half Life 2 DirectX 9 vs DirectX 8 Performance Penalty

At 1280 x 1024 things get just a little worse, but you should get the picture by now - the GeForce FX line does not perform well as a DX9 part under Half Life 2.

Half Life 2 DirectX 9 vs DirectX 8 Performance Penalty

You will undoubtedly see these statistics reflected in the actual performance of the 5900XT in the coming pages, but basically if you are a NV3x GPU owner you will want to run Half Life 2 in DX8 mode and not DX9 mode.

Now let's take a look at how the rest of the GPUs perform in both DX8 and DX9 modes. For these tests we used the exact same drivers and platforms as our first article, just with different video cards so the numbers are comparable.

DirectX 9 vs. DirectX 8: Image Quality Battle in the Canal
POST A COMMENT

62 Comments

View All Comments

  • stealthc - Monday, January 03, 2005 - link

    I have geforce4 mx 440, with 1.7ghz intel cpu and the frame rate is unacceptable, jittery all to hell, stupid HL2 autodetects it as dx7.0 card, I forcefully correct but still. How the hell are you pulling off frame rates like that with that particular card? I barely get a stable rate at 800x600 res at 20fps. I see lots of red and usually yellow when I use cl_showfps 1 in console
    There's something completely messed with this game it shouldn't run like this.

    Counter strike source is unplayable, I can't get over 10fps in it. Usually it hangs at around 3 to 6 fps.

    I'm pissed off I spent $60 on a freaking game that does SH*T for me. I already beat it but it's so not smooth and jittery I'm surprised I could even hit the side of a barn.
    Reply
  • Barneyk - Monday, December 06, 2004 - link

    "The next step is to find out how powerful of a CPU you will need, and that will be the subject of our third installment in our Half Life 2 performance guides. Stay tuned..."

    Yeah, im waiting like hell, i wanna se how HL2 performs on old CPUs, i have a TB1200 and a GF4 Ti4800SE, Graphics perfomance was OK, but the game playing was really sluggish, but still very playable.
    And i wanna se som comparison graphs on how different CPUs perform, and i've been waiting, when do we get to see the HL2 CPU roundup? :)
    Reply
  • charlytrw - Thursday, December 02, 2004 - link

    take a look on this link, here are some interesting info about how to improve performance on geforce fx users in about a 50% on dx9 mode. Try it yourself. The link to the article is:
    http://www.hardforum.com/showthread.php?t=838630&a...
    I hope this can help to some geforce fx users... :)
    Reply
  • clstrfbc - Wednesday, December 01, 2004 - link

    Did anyone else notice the piers and pylons don't have reflections in the screen shots on page 2, Image Quality, the second picture, the one of the water and damn/lock.

    When you roll over, you see the non-reflexion version in dx8, rollback and something looks funny. The refelctions look good, but there are lots of things missing, most obviously the piers, aslo some bullocks? (things you tie ropes to)
    Hmm, maybe the pier is a witch, or is it a vampire, so it has no reflection....

    Other than that, the game is awesome. I'm well into city 17, and only took a break because the wife was becoming more dangerous than the Combine.

    Running on a Athlon 1.7 512Mb and Saphyre Radeon 9000 128Mb?, it it plays fine excepte the hickups at saves and naptimes at loads.
    Reply
  • MrCoyote - Monday, November 29, 2004 - link

    This whole ATI/Nvidia DirectX/OpenGL optimization is driving me insane. Developers need to stop trying to optimize and code for "universal" video cards.

    One reason why DirectX/OpenGL was created, was to make it easier on developers when accessing different video cards. It is a "middle man", and the developer shouldn't have to know what card someone has in their system.

    But now, all these developers are using stupid tricks to check if you have an ATI or Nvidia card, and optimize the paths. This is just plain stupid and takes more time for the developer.

    Why don't they just code for a "universal" video card, since that's what DirectX/OpenGL was made for?
    Reply
  • Alphafox78 - Wednesday, November 24, 2004 - link

    'Glide' all the way!!!
    I used to run Diablo on a voodoo2 and you could switch from glide to Dx and glide looked so much better! even after upgrading to a geforce card glide looked better... not that its relevant here. heh
    Reply
  • nserra - Wednesday, November 24, 2004 - link

    #55 OK I agree. No I was not trying to saying that 6xxx is bad, or that it will be bad. Nor I think it superior do Ati X line and vice versa. But we already have the "bad" example of the nvidia FX line (for games), how will both scale?! I didn’t like 3Dmark bench, but was 3Dmark03 that bad? It was already showing something at the time….

    One day I said here it was important to bench an Ati 8500/9000 vs Geforce3/4 using current drivers (and platforms maybe too) with new and older games to test.
    This is very important since you will know if support to older hardware is up to date, and also how is performance on these cards today.
    I am not saying that while 8500 loosed to Geforce4 in the past, that it will win today, I just want to know how things are now! Is PS1.4 of the 8500 giving it any thing now since then it doesn’t? I don’t know!

    But everyone said the test was pointless and that there where no need to bench "older" hardware, since no one was planning to buy it....
    Reply
  • Lord Banshee - Tuesday, November 23, 2004 - link

    well it kinda matter what you do with your computer. Personaly i like my 5900XT over the 9800Pro i have because it is faster in 3D Modeling software.. But for gaming the 9800Pro is fare better.

    But as you say that the NV3x wasn't bad then but now, well i think it was bad then just had nothing to prove it. Now that their is a lot of DX9 games out the poor mistake on nvidia engineers show. And they know they missed up thats what the 6800 can use a 16x1 pipeline config. I belive the 59xx's were 4x2 and the 9800 was 8x1. the 6600GT also shows nvidia's mistake being that it is a 8x1 and beating the paints off the 4x2 5900 series.

    But i was unsure about if you were tring to say in your last post if you think nvidia fudged again and that it will show in due time with the 6800 series, but if this is the case and you are tring to say that. I don't think so and i believe Nvidia truelly has a very compeditive card that meets the demands of all users.
    Reply
  • nserra - Tuesday, November 23, 2004 - link

    #52 You are right. I was talking about the NV3x core. But dont forget that this CHIP was not that bad at the time, its really bad TODAY!

    Example: Now who is better served, who had bought a 5600/5700 or a 9600 card?
    Reply
  • pio!pio! - Monday, November 22, 2004 - link

    #48 agreed..it may be that dx8 water looks that crappy while dx8.1 water looks like what you and I are seeing Reply

Log in

Don't have an account? Sign up now