Final Doom3 benches at HardOCP

Two Z's per clock is the max. This includes AA.

 

I will back Walt up on this one. It seems both companies have applied some optimisations. ATI have used their 'soon to be patented' optimisation (as mentioned by Carmack) & maybe others too, whilst im sure nVidia have also used every doom 3 trick they have tested to date on alpha & beta versions. I dont think ATI should be pointed out with fudging anything here as they have been clean in publicly disclosing their universal optimisation.

There is no screen captures, only numbers here so this does not show the buyer what they will recieve at 'x' fps. I would like to see how cards compare visually across the different rendering paths as well as performance.

Lets also see say-
2xAA - 6x(8x)AA, and combos with AF 2x, up to 16x- starting as resolutions 800x600-1600x1200 (where applicable) - with a variety of cards - so people can see what settings they are likely to use (as most dont own the high-end beasts tested). There seems to be a few abnormalities with comparitive frame loss or this might be just architectual or driver related?

Where is the beasty deltachrome with their optimised z-rendering??

 

Hi Mintmaster,

What do you think is the point of the "opengl rewrite" that people have been talking about? Just memory controller optimization?

Nite_Hawk

 

I am sorry, but I dont remember ATI advertising the optimizations they use before public found out for themselves.

 

Cool jb! It would be nice, however, if everybody knew who has the mod rights. I don't wanna tell someone to shut up that has told me to shut up only to find out that I really should have shut up in the first place (ifyouknowwhatImean)

 

So is ATI rewriting the OpenGL drivers or not? And will they be finished within the next 2 weeks?

 

2 weeks? No 2 years

 

They are but I highly doubt it's gonna be finished by then...

 

Some other games have lower framerates on ATI hardware when CPU-bound. For example, look at Serious Sam. This is nitpicking, though. I don't care about going from 170fps to 190 fps.

The other reason is that ATI seems to have implemented some optimizations in D3D only, for example the low performance hit of AF. I haven't seen this in OpenGL, and NVidia have mentioned this as well ("ATI is cheating in D3D!"). My guess, though, is that Doom3 is a special case and probably already has these optimizations through app detection (it would be stupid for ATI to do otherwise).

Personally I don't see why there's such a big fuss about the re-write. Something else seems to be holding ATI back, and if they find it, it'll just be a few chip configuration adjustments that are necessary, not a complete rewrite.

 

Perhaps on the OpenGL side of things their code is horrible. I wouldn't be terribly suprised if half the reason for the rewrite is to clean it up to be in the same state as their DX code. It'd certainly be nicer for them to have a clean and reasonably generic way to enhance the opengl drivers without having to do application specific optimizations that could be general case ones. Especially if they are gearing up for opengl 2.0 compatability and expect anyone to use it.

Nite_Hawk

 

The DX driver is a clean slate fore R300 onwards. OpenGL wasn't - not only do they have the historical code in there, but also a bunch that was added after they bought FireGL.

Naturally, though, there is only so much that a driver can do and over half the passes taken in each frame of D3 will favour nvidia architecture more than ATI's.

 

Strange thing is that with Half Life based mods, like CounterStrike 1.5, I used to have 100 fps no matter what happened. However, when Valve launched Steam, performance in every Half Life based mod dropped considerably. To the point where newer gamer produce better graphics and perform equally as good.

Strange thing is that it only seems to affect ATi users. Maybe it is some compatibility problem with their OpenGL drivers. Hopefully the complete rewrite you guys mention might solve those problems.

 

I dunno if anyone else saw this, so what the hell... I will post it just incase.

G4 TechTV: Doom 3 promo

Not sure if what I heard was accurate, but if you increase the speed of the playback, I could have sworn you hear J.C. say, "ATi suXor."














j/k... hehe

 

Yes, but it would still be nice to know that it's not their own sloppy / messy code holding them back, but factors outside of their influence.

 

Yeah, but I'd still like to see theoretical results. I thought NV3x was the same until I saw this:
CPU : Intel(R) Pentium(R) 4 CPU 3.00GHz
GFX : NVIDIA GeForce FX 5950 Ultra
OS : Microsoft Windows XP
Settings : 1024x768 32 bits D16 4x FSAA

Benchmark Result
-----------------------------------------------------------
FrameBuffer Clear : 3251.2 FPS
Color Fill : 1736.442 M-Pixel/s
Z Fill : 1869.821 M-Pixel/s

I've seen programs show the X800XT can do 4.1 Gpix/s Z fill with 4xAA, and would assume the 6800GT can do 2.8, yet the latter still wins in D3.

The only thing I can think of is that Carmack's reverse algorithm is incompatible with hierarchical Z, since it reverses the depth test. R420 doesn't have a simultaneous min-max HiZ system, AFAIK. It can only use HiZ during either the lighting passes or the stencil passes, and I would assume they use it for the former since that's how all other games work. It would be interesting to see ATI use HiZ for the stencil volumes, but I don't know if that's possible or even advantageous.

If this is indeed the bottleneck (not being able to rapidly cull both stencil and lighting), then I think the R420 architecture is in trouble for Doom3. Hopefully they can use HiZ for the stencil pass, and Early-Z for the lighting pass, and that'll get them up to speed. Again, I would think that they've already tried that, so it doesn't look good...

 

I just laughed when I read Walt's post a page back. Yes, of course! It all makes sense now! If someone tells you that something is not the end-all-be-all, it must not be taken into consideration at all!

Gotta love fanboys.

And, it's still too early to say NVIDIA has won this generation, but they definitely have the advantage. Then again, maybe the only ATI drivers that [H] could find that work with Doom 3 only expose 128 megs of RAM...

 

Re: PC gaming

Not really supposed to be except for lighting. There is some other effects like heat waves and the sort but besides the lighting algorithms which aren't too complex

Hopefully ATI has some leeway in reworking Z performance and will concentrate on that for their next card. ATI is only going to lag behind as more and more games use stencil shadows if they don't prolly. Curious how does ATI compare in Thief 3 to Nvidia? I would be curious to see FRAPs scores in areas with high levels of shadows in that game.

But really stencil shadows is definately going to run into a wall due to its brute force nature. Heh, maybe we will see cards start to add in clipping for creation of light volumes for handling shadows instead.

 

Yeah, regarding drivers being rewritten - I found it: http://www.rage3d.com/board/showthread.php?p=1333009108#post1333009108

 

Rev, I quoted the appropriate passages directly from Carmack as quoted by [H]. I do not understand how you might think I'm "reading" anything except the Carmack quotes I reference as the basis for my remarks... (?)

Plainly, although he states that ATi's optimizations *do not visibly impact the quality of D3,* he complains that they exist, and states that when they are turned off (automatically) ATi performance drops: "the ATI cards did show a performance drop when colored mip levels were enabled" (which is exactly and precisely what ATi has always said would happen, is it not?)

OTOH, his complaint is that nVidia's drivers specifically "have been tuned for Doom's primary light/surface interaction fragment program," which means other non-D3 OpenGL games (if not D3d-based games, too) will "fall off the fast path" and therefore experience "significant performance impacts, especially on NV30 class cards." Quote, unquote.

His criticism of ATi's optimization is presumably D3-specific (since it would be improper for him to speak for other developers about other software), nevermind that the ATi optimization functions the same in all 3d games regardless of API, and that ATi has explained it all many times prior to Carmack's statement, and that Carmack himself states that render-quality for D3 isn't affected by it. His criticism of nVidia's D3-specifc driver optimizations is that performance will fall off in non-D3 OpenGL games, especially in the case of nV30-based products. Although he's plainly talking about apples and oranges, since ATi's optimizations are global and nVidia's are D3-specific, he doesn't care to make that distinction. It's a big distinction, to me.

So, what is it exactly that you think I'm "reading" into his comments that he's not quoted by [H] as having said? If all I wanted to say was "John likes NVIDIA," that's all I would have said. But, would that statement have had any meaning?

Instead, I chose to address his specific remarks as quoted by [H]. I'm not sure how that would present a problem for B3d's bandwidth... Really, I don't know what you mean, as I can't see how it would have been more helpful for me to simply state my opinions without referencing the Carmack quotes which caused me to reach those conclusions. Without proper attribution it becomes very difficult to justify and explain such conclusions, I think. One must never assume that he is writing to an audience of mind readers and telepaths, such that he doesn't actually have to illustrate his points to have them understood, imo...

If you have any alternative conclusions based on your interpretation of these remarks from Carmack as quoted by [H], I'd be happy to hear them... I suppose that's why I bothered posting my own interpretations in the first place.

 

Boy, did I pick the wrong day not to be on the boards!!!

Question; do we know if ATi brought their rewritten openGL bit with 'em to test it with or was this run using old code? (Sorry, I'm just trying to get up to speed with the discussion)