Slashdot Mirror
Initial Half-Life 2 Benchmarks Released
dfj225 writes "According to an article on ExtremeTech.com, it looks like ATI has the lead in Half-Life 2 graphics card performance. Valve benchmarked their new game using the top cards from both ATI and nVidia. Results show the ATI Radeon 9800 Pro drawing around 60 FPS while the nVidia GeForce FX 5900 Ultra only draws around 30 in Half-Life 2's DX9 full precision tests. Read the article to see results on other tests that Valve ran." Update: 09/11 13:06 GMT by M : Another article about the presentation.
Sure, they are compiling Direct X 9 for Linux as I'm writing these lines.
Mother is the best bet and don't let Satan draw you too fast.
I take it you guys have seen the ingame movies? Looks very nice, and seems to take game physics to a whole new level, but at the same time it looks as if you need a Pentium 5 to get it to run properly!
If Google really cared they would fix Android Chrome to reflow text, instead of discriminating
How is objectivity of this study any different from, say, a study by Microsoft promoting Windows?
What's this Submit thingy do?
Forget ExtremeTech's article, and go check out the one at The Tech Report. According to Gabe Newell of Valve, one of the graphics card companies was trying to detect when a screen shot was being made, so that it could output a higher resolution frame, hiding the quality trade-offs made by the driver. From the article: "He also mentioned that he's seen drivers detect screen capture attempts and output higher quality data than what's actually shown in-game."
Is it just me, or is ATI pulling a real turnaround? They used to be the underdog for so long -- their drivers weren't the greatest, their marketshare was second-fiddle, and they initially missed out on the Xbox contract. I never thought I'd see the day where nVidia, which is practically the industry standard for gaming, might be challenged on such a thing as actual performance.
;-)
Oh well, at least communication between hardware and game developers has improved to the point that I won't need to specify to the game whether I have a Hercules, Tandy, or Trident chipset...
The coolest voice ever.
And just how long will it be before someone finds out that one or both of those video card manufacturers has been "tweaking" their benchmarks to improve the acheived frame rate?
Anyhow, just who runs Half-Life or anything with all the eye candy maxed up? No serious gamers that I know of, that's for sure. At the settings that hardcore FPS addicts play at, the frame rate delivered by any card currently being shipped either ATi or nVidia will be sufficient (assuming that the rest of the system isn't subpar).
Once again, for those of us without money to burn the smart buy is that $100-$200 card that cost $600 a few months ago, not the one that costs $600 now (and which will be down to $100-$200 just as fast).
"Accept that some days you are the pigeon, and some days you are the statue." - David Brent, Wernham Hogg
What a terrible article. It didn't even say what resolution all that was happening at.
Caffeine Good
nVidia has been circulating its Det50 driver to analysts in hopes that we would use it for our Half-Life 2 benchmarking. The driver contains application-specific optimizations
The article fails to mention whether they actually detect the application and run the driver through a different code path, or if they've made general driver-wide optimizations that happen to also help Half-Life. Knowing the behavior of these video card companies in the past, I would suspect they have huge chunks of code in there devoted soley to Half-Life.
So, now instead of having to hack around and catch companies cheating on drivers, we just have to read as they admit it openly? This is standard operating procedure now???
When I download the latest Detonator drivers for my nVidia card, I want to download a generic D3D/OpenGL driver, not a Half-Life driver. The amount of time they spend "optimizing" for the popular games is time they could have been spending making sure the performance and quality is adequate for ALL games and modeling apps.
I wouldn't value these benchmarks too much, given they're from a game that hasn't yet gone gold. Features could be dropped from the graphics engine that will affect the way each card deals with the graphics.
... you insensitive clod!
Seriously though, are they allowing for people with older cards? (UT 2003 ran fine on my Voodoo3 and still looked pretty darn good, even w/o transparency, anti-aliasing, or any of the other modern GFX buzzwords)
If a user wants to take a screenshot, shouldn't it be at the highest available resolution? If they can do it with a low overhead, they should. It's the lying on the benchmarks that's the problem here.
Karma: Excellent^(-t/Tau), Tau=Wittiness/Trollishness
It's all very nice seeing how the latest and greatest cards perform but how about some test results for older cards.
I prefer to save my pennies and upgrade my graphics card to the one just behind the current generation.
I've got a fever and the only prescription is more COBOL.
NIVIDA fanboy: blah blah blah nvidia has better support... blah blah blah!!!
I'm not sure what is going to end first, the Israel-Palestinian situtation or the ATI vs NIVIDA arguement.
The fact is both regularly cheat on performance and quality benchmarks, and if you think you can actually say one is better then the other you are a biased fanboy.
Just buy the one on sale, please.
Actually, it all boils down to:
DX9: Bad
OpenGL: Good
All Valve is doing is making it harder for other OS's to get their games. So I think I speak for all the *nix users when I say they can go fornicate themselves with an iron rod.
Half Life 2 is DX 9 only.
Mother is the best bet and don't let Satan draw you too fast.
Well, if you consider this is an article about Half Life 2 and not about Nvidia's and ATI's open source strategies you might see the interest.
Mother is the best bet and don't let Satan draw you too fast.
ATi is bundling HL2 with there cards soon so anything in this article at best gives you an idea as to HL2's performance.
Let's also remember that once ATi was much bigger than nvidia in graphics, and charged exorbitant prices for crappy chips, with shocking driver support.
Let's also remember nvidia have much better performance so far in the more important (and independant) doom3 benchmarks (where 16bit floating point precision is used for nvidia cards, instead of 24 for ati and 32 for nvidia, as directx9 was originally going to specify before nvidia and microsoft fell out).
Also remember that nvidia's cards offer better performance in most 3d rendering apps (where both cards use 32bit fp and almost all of ati's advantages evaporate), so driver tweaking on nv's part in games does not necessarily mean they have a lesser part for that.
Finally linux support is a no brainer, nvidia have been doing it well for years (with support as far back as tnt), ATi have made a recent attempt that is not user friendly, or even support all radeon chipsets, let alone rage 128.
ATi are onto a good thing right now with the current directx9 spec giving them an advantage in games that stick to the spec instead of the optimum end user experience. That is about all they have going for them though. This battle has far from swung the other way, it's merely gotten closer than it used to be.
Toms Hardware
FiringSquad
Tech Report
Gamers Depot
Beyond 3d
I often hear people say "after 30 fps you can't tell the difference", or something to that effect. That might be true if you were playing back the frames evenly spaced. However, your monitor runs at a fixed 60 Hz framerate (or 70 or 80, but let's just say 60), so a "fps" of 50 will have you showing 5 frames, showing the last frame again, and then showing 5 more, which can produce a noticable stutter even though the "fps" is 50. So that is one reason why you might want a "fps" of at least 60 (or 70, or 80). Also, the really meaningful value is "minimum fps", because that is what you're going to get when you're fighting the boss and all these guys are coming at you and all these things are happening at once. Usually, a higher average fps (say, 120) indicates that the minimum fps will also be higher. So, a high fps score can still be good even if your monitor can't display 120 frames every second.
main(c,r){for(r=32;r;) printf(++c>31?c=!r--,"\n":c<r?" ":~c&r?" `":" #");}
the thing about it is that lets say the game has a fps of 30, thats an adverage score, which meens there was a high and a low, in some parts the fps may drop to as low as 0 or as higher then 60, if your trying to frag someone and your fps goes down to 3 your in trouble, the higher the fps rating is the higher the min fps is
Actually, if you bother to read the god damn article, you'll find that your 4600 (and my NV28 4800) beat the NV30 cards when the DX9 gubbins is turned off. Given that Valve are saying that it'll run on a DX6 or later card, it looks like this'll be a viable option for us poor bastards with 6 month old hardware.
If you were blocking sigs, you wouldn't have to read this.
These results mirror 3DMark03-results perfectly. It seems that NV's DX9-support is horribly broken. Why else would their cards need separate codepath (In HL2 and in D3(Although D3 is OpenGL-game, it uses many of the same features)) whereas Ati-cards do not? Carmack has said that if D3 does not use the NV-specific codepath, NV-cards will have poor performance.
Lesbian Nazi Hookers Abducted by UFOs and Forced Into Weight Loss Programs - -all next week on Town Talk.
If you take a screenshot while running at 1280x1024, and it outputs a 1600x1200 picture, then it's providing "the highest available resolution". If you take a screenshot while running at 1280x1024, and it gives you the same size image but with all the ugly "trade visual quality for speed without the user's request" hacks turned off, then it's just lying.
This should really be in the slashdot FAQ. It was settled way back in the day with 3DFX's demo comparing 30 and 60fps side by side.
1) The fps number is an average. If you average 25fps, then when things get busy on screen the rate can drop to 15 or something, which is very visible and ugly. I you run at 60, that doesn't happen.
2) 25fps looks bad for rapid movement and panning (ie, most games). Next time you watch a film, look at how blurry everything looks when the camera pans rapidly.
I'm not 100% certain about the specific cards tested, but for several of the highest end NVIDIA and ATI cards a head-to-head comparison for performance doesn't tell the whole story.
This is because ATI cards have implemented a 24-bit floating point pipeline while NVIDIA cards implement a 32-bit pipeline. It is reasonable to expect the ATI card to outperform the NVIDIA card at the expense of some round-off errors. 32 vs. 24 bits on a color pixel is probably no big deal (although some color banding might arise), but when those results apply to vertex positions you could begin to see cracks in objects and shadows.
Note that the ATI card is still faster for Half-Life 2 in 16-bit mode, so it is probably a faster card overall for that game. There are so many ways to achieve similar looking effects on modern graphics cards that even as a graphics expert, I can't tell which card is actually faster.
I've been working with both the GeForceFX and Radeon9800 for some time and both are amazing cards. They have different capabilities under the hood, and can perform different operations at different speeds. Furthermore, under DirectX both cards are restricted to a common API but on OpenGL they have totally different capabilities. I don't think a consumer would go home unhappy with either card, except for the price.
-m
(mind you, there's just been a new driver release from ATI, and I haven't installed that one yet)
I've got a 9700 Pro, and the ATI drivers have given me *a lot* of grief as a developer. There are many times when they are so blatantly non-compliant with the OpenGL standards, it's not funny.
For example, the driver claims to support OpenGL 1.3. With 1.3, ARB_multitexture has been promoted into the core, so they driver _should_ export glActiveTexture & friends without the ARB suffix. Well guess what? It doesn't. You have to use the *ARB versions of the functions.
I guess that a lot of this can be attributed to the fact that ATI is not as long in the Linux driver business as NVidia, and overall, things have in fact gotten better over time. But you should expect a bumpy ride.
About the only thing this is illustrating is that the performance problems with D3D are pretty severe now. DX couldn't correctly render fog or water in the original Half-Done(tm) engine, and going to OpenGL drivers would not only boost the frame rate by as much as 66%, but would also correctly render those effects.
Also, RTFA, Nvidia is a little shy about "optimized" drivers for benchmarking certain applications. They specifically requested that the optimized drivers not be used. No indication that ATI did the same.
I doubt there will be a Linux version of HL2 either, because this new 3D engine appears to only support DirectX.
That's a shame, because the world didn't end with the America's Army developers ported AA:O to Linux. As a matter of fact, it runs quite well, and it didn't take them 5 years to produce nothing but vaporware.
Fred
"A fool and his freedom are soon parted"
-RMS