Final Doom3 benches at HardOCP

cool, I can see my mobo, cpu and GPU in there.

Thanks Kyle

 

hey kyle will the 5700 ultra be in there with the other cards?

 

Well, I personally would never spend $400-$500 on a 3d card simply based on pre-game-release closed benchmarks conducted on the premises of a known TWIMTBP developer, any more than I'd spend a like amount buying a 3d card on the basis of pre-game-release closed benchmarks conducted on the premises of a known GITG developer (speaking of ID Software and Valve, respectively, of course.)

While I might be interested in how *both* of these games run on one or the other product, I'd also be very interested on how either prospective product will run all of the other 3d games aside from HL2 and D3 that I plan on buying throughout the lifespan of the product I choose.

Basically, I think anyone who'd actually let this kind of PR hype influence his buying decision to a distinct degree is going to wind up a day late and a dollar short, and may well regret such a decision made in haste. After both D3 and HL2 ship there will be plenty of time to patiently and calmly analyze the competing products in depth to see which provides the most for the money. That's certainly what I plan to do.

Again, if you go back to the '03 [H] D3 Preview closed benchmarks and revisit those scores, it might also have appeared evident that there was "no way" R350 was going to catch nV35. But as we all know, the truth of the matter was precisely the other way around... "Haste makes waste," I believe is the applicable sentiment in this case. So, a person can either respond emotionally to this kind of calculated PR blitz, or respond rationally and simply...wait awhile...to see what "shakes out," right?

 

OK, thanks for answering. I now have two additional queries...

(1) What other games aside from D3, apart from D3-engine games, might we expect to provide us with "D3-style" rendering?

(2) What R4x0 architectural advantages do you see, if any, which might serve to offset such nV40 architectural advantages as you believe exist? I'm actually asking since, obviously, there's a whole lot more going on in rendering a 3d-game aside from the elements you've listed for a "D3-style" rendered game.

Curious as to how you'll answer, as obviously there are as many "styles" of 3d rendering as their are various 3d-game developers, right?

 

Although I'm fascinated that you conclude that with the '04 nV release of the 6800 it's only taken nVidia ~2 years to equal ATi's ~2-year-old 2d IQ (as originally offered in the 9700P in '02), I still must question your methodology. It's not entirely clear how much of the IQ improvement you've observed comes from your new monitor, which certainly it seems to me could also provide better 2d IQ support for your original 9700P... Or, are you just saying that 2-year-old 9700P's offer better 2d IQ on older monitors than even the current 6800s...? It's not really clear...

DV is the one "feature" that I believed I would miss when moving from the Ti4600 to a R9700P in 9/02. nV had me convinced that such a feature was a necessity, as nV drivers required me to use it in 3d-gaming to clearly distinguish between colors like oranges and reds, and to distinguish between low-intensity grays.

In fact, it took me two weeks on the 9700P before I realized that although the drivers lacked DV, I could clearly distinguish between orange and red and among the low intensity grays that were only visible with my former nV product when running some level of DV.

I accordingly reached the conclusion that DV was a deliberate manipulation added to the drivers as a type of marketing bullet, which did not actually have to be present at all, if one's desire was to view his 3d-games in the proper color balance and contrast intended by the developers.

IE, it became obvious to me that nV's DV-off color saturation and contrast settings were deliberately sub-optimal so that with DV on and set to at least the lowest level suddenly things appeared much more clear and less washed out... What I got from the normal operation of the Catalysts was very close, if not superior to, what I got from nV with my DV turned on and set to the lowest level (anything higher I found bizarre.) In my third week of owning a 9700P and playing the same games I'd been playing with my Ti4600, I realized I'd been "had" by DV all along. I've never missed it since for even a day.

The subject of DV has been beaten to death and I have no desire to debate it any further than this statement. Suffice it to say that I can see some utility in it for people who wish to slightly or egregiously over saturate and distort their colors and contrast so that they display much differently than intended to display by game developers--but I'm not one of those, and DV's primary utility for me when I used it was to obtain the degree of color saturation and gray-scale contrast that I get from the Catalysts normally without any need for a DV-like contrivance.

I do find it odd that with your older monitor, though, you found the 9700P's display superior, since I should think the only practical use for DV might be as a color-saturation and contrast band-aid for older monitors whose display capabilities have degraded over time.

 

its not really the style of rendering that created the performance gap. Games using heavy use of shadows will run better on Nvidia's cards but its marginal performance increases, (maybe 5% at most). Shadow culling just improves performance on shadows. So games like doom 3 lets say 100k polys are being rendered with all volumetric shadows that equal to 200k polys on cards without shadow culling (ultrashadow 1 or 2). Culling might take out 5-10% of those polygons that were used to create the shadows. The major performance was becuase of Ogl, not from ultrashadow.

 

I haven't followed the games market all that much lately, but I don't expect many games to follow exactly that route (except D3-engine games, of course).

However, there are already some games out which use stencil shadows, and others will follow (remember, UE3 uses stencil shadows for some objects). Also, fast Z fill equally helps shadow buffer performance, where NVidia has another advantage (since NV20 actually), "free" percentage closer filtering (PCF).

Unfortunately, two of those optimizations mentioned (depth bounds test and depth clamp) aren't even available in D3D. Certainly, NVidia could try to go the FOURCC hack route again, but that's abusing the API. Which is why I believe DirectX is seriously lacking an extension mechanism.

The single most important advantage R420 has over NV40 is clock speeds.
Then there's 3Dc, which brings a bit of a quality improvement over DXT5. It might have an advantage in early Z culling, but I don't know the relevant details of both chips here. It has no temp register issues. And its vertex shaders seem to be faster.

There are certainly lots of other differences, most of them we'll never know.

 

My God you love ATI. No further comment.

Dumbass, I didn't buy the new monitor at the same time. I've had this monitor for a while. And no, the Radeon 9700 Pro was quite blurry with the old monitor above 1024x768.

 

I don't know why you think NVidia needed new cards to "equal" ATI's 2D IQ, since they already had good cards with the same IQ (with the notable exception of overlay quality). Signal quality depends on the card manufacturer. My GF3Ti200 certainly had better signal quality than the R9800 I own now, but since I use DVI now this is a non-issue.

How come you know exactly what the developers intended? Did you do color calibration with your monitor?
If not, I suggest you give it a try before drawing premature conclusions about DV.
NVidia allows you to adjust the saturation to the characteristics of your monitor (although they offer too few levels, IMO). ATI does not.

 

Heh, the AV8 and 3500+ is mine in there as well.

Looking forward to it.

 

I disagree. "because of OGL" sounds like you attribute most of the difference to the drivers. I don't think that's the main factor.
Depth bounds test doesn't cull polygons, it rejects per screen space tile (and per pixel, if you can't trivially reject per tile). And I think the rejection ratio is quite a bit higher on average than the 5-10% you mentioned (even with scissor rectangle). It very much depends on the size of the room, of course.
UltraShadow consists of other parts which are equally important.

On one point however I do agree, I don't think the big performance difference came from depth bounds test...

 

DV is a bit like the loudness button on amplifiers. If it makes people feel better so be it. Kinda like many people like to run their screens at 9300K because the white looks whiter and the overall picture fresher.

 

It is likely that future games will still use a combination of shadow volumes, shadow maps, and projective shadow techniques, plus reliable 'ole "baked in" shadows for static content and also precomputed radiosity. You need only look at Unreal Engine 3. That's a next-gen "in the future" 3D engine right?

Shadow maps still have significant issues that preclude them from being a no-brainer to use, just like shadow volumes have their own pros and cons.


Dave, the reason som are accusing you of bias is simple: Click on your link, "find all posts by Dave Baumann" and a surpising number of them (re: big majority) defend ATI directly or indirectly. I mean, I can understand when you have actual info to provide to correct someone, but, for example, your speculation that the D3 engine is limited to D3 style-content with few polygons or players, is mere speculation, downtalking D3. Frankly, you don't know what the D3 engine is ultimately capable of in terms of settings (outdoor vs indoor) or number of simultaneous characters or players. I doubt even JC knows, as it is up to third parties to stretch the technology to its limits.

I have my doubts as to whether all of this D3 criticism would even exist if the benchmark results had been reversed. Why bother with negative speculation at all? Did you have any negative comments for 3dMark GT2/GT3 (which use degenerate quads to extrude volumes in vertex shaders, which no one in a real game would want to do) when ATI was leading in those benchies? Why not call Futuremark on their dead-end game-test and poor implementation?

In any case, if you want to argue for shadow buffers, Nvidia has advantages there too. First, they have direct support for them and HW PCF. If you want to go further and do filtering in the shader, dynamic branching in SM3.0 allows you to do 16x-64x oversampling where needed, and it is definately a performance and IQ win. So really, it doesn't matter whether you think D3 is a "dead end" technologically or not. It was the first large scale test bed for shadow volume techniques, and the lessons learned will be applied to future engines. If Tim Sweeney really thought they were a blind alley, they wouldn't feature prominently in UE3.

 

Why is it your remark here is as unsurprising as your earlier post?....

Before resorting to a juvenille retort as a substitute for an answer, I'd encourage you to re-read your original post and reflect on why it was you chose not to include this information in it. I really wish I could read minds, but unfortunately I can't quite do that yet, and since your earlier post didn't include this information it seemed appropriate to ask you for clarification. Agreed that it must've been your monitor.

 

Let me stop you right there.... I was characterizing what it appeared Chalnoth was stating--I wasn't stating it myself.

On what technical specifications do you base this assessment? (Not that I really care, mind you...)

I always color calibrate the monitor. Connected to the same monitor, I had to employ DV to get the same saturation and low-intensity gray scale contrast as I got with the Catalysts in their normal operation. Nothing premature about it, btw, as it initially involved a few hours of on-and-off calibration work on both cards with the same monitor.

I have had no need for DV for the past two years (which was a relief, actually, when I discovered it.) As I said, I think that for some people--people using nVidia graphics--it has a certain utility...

I really don't wish to belabor the subject any longer except to simply repeat that in two years without it I haven't had a single day of missing it... Amazing, no, considering what it is you don't think ATi allows you to do?...

 

This is true but think of what hurts proformance fill rate wise. Polygon fill rates usually hurt more in a doom 3 type game because the shadows themselves are not that *deep*. If you have a shadow that a hundred miles long yeah that will seriously hurt fill rates but Doom 3 doesn't have that, also they are using shadow caps not extruding to point (extruding numerous volumetric shadows to a point will hurt fill rates greatly). Doom 3's room sizes are generally small, now I don't kmow how big they are just guessing from alpha leak but shadow depth it too has effect. The extensive use of shadows and the high polygons of the rooms and characters, the creation of volumtretic shadows has a fill rate drop but its more effected at the polygon fill rate level not really pixel fillrates.

nV architecture is suited better for Doom 3 type games that use heavy shadowing but its limited. COD still has a huge performance gap too between the gf 6 line and the x800's too.

 

If you don't really care, why are you participating in this discussion at all?
Stating things like there being no "physical difference" explaining the benchmark results although this is clearly wrong doesn't help anyone.

That's strange, considering DV doesn't affect gray scales at all.

 

It's gratifying to know, DC, that if the results had been reversed you'd have been very positive and said, "Shoot, go ATi! There's *no way* nV40 will ever catch R420, and this benchmark proves it!" Heh... Uh-huh.

But I'll say again as to my own skepticism, that it is based on the following:

(1) It's yet-another-closed benchmark run, with code unavailable to the public, orchestrated and conducted by private parties only interested in giving us convenient numbers pertaining to frame-rates, along the lines of last-year's infamous GITG HL2 benches and [H]'s 2003 TWIMTBP-sponsored "Doom 3 Preview" (which should have been co-titled "nVidia's nV35 Wet Dream Preview.")

I'm sorry you haven't had enough of that but I certainly have.

(2) Not only is there no public demo-benchmark-whatever of D3 that's been released so that we can all dig our teeth into it and give it the level of scrutiny it deserves, but apparently that aspect of things seems completely unimportant to ID Software--even just two weeks ahead of shipping with the code frozen and as Gold as the teats on King Tut's concubine. I mean, why involve your public when it's so much easier to manipulate them instead?

The D3-ClosedMark situation is only marginally better than Gabe Newell swearing on his mother's pantaloons that we'd see a general HL2 benchmark released into the public domain by Sept. 30th, 2003, only to have so totally forgotten his declaration that you wonder if Alzheimer's has taken root somewhere deep inside Valve's hive mind.

(3) I find it remarkable that we've all been treated to several public statements from Todd HolesintheHead about how the "team" is so busy preparing D3 for shipment that there just isn't any time for ID to release a public demo of the game. You know--it's "Gee, we'd LOVE to release a demo of the game to our fawning public, but right now we're just too busy and all."

But, gosh, not so busy that they can't get behind a TWIMTBP contributor with something like this, where JC himself takes the stand and swears up & down on a stack of Korans that ATi is "fudging" while seemingly unable to recall anything specific about nVidia doing any sort of "fudging the texture filtering" in the drivers he was using to conduct his closed, private testing. Carmack is as vacant-minded about that as Newell is about the HL2 benchmark he publicly promised.

And this whole dog & pony show is what some people call "professionalism"...? Myself, I don't see it. I see various companies trying to make chumps out of their customers, and expecting those customers to roll over and become patsies on the Wheel of Life otherwise known as the Almighty Dollar.

Things are not going to get better until customers demand that they get better. It's for sure that "rolling over and being positive" is absolutely the wrong way to do that. Let these companies back their claims with the public release of demos-benchmarks, and when we can see positive things demonstrated publicly, instead of behind closed doors and hidden under layers of secrecy, there'll be plenty of time to wax positive--not to mention plenty of *reason* to do so.

 

nVIDIA's shadow mapping technique works for conventional shadow mapping, without transparencies. But when you need to do transparent shadow mapping, you'll need another channel in the shadow buffer to store the transparent regions. This certainly cannot be accomplished with D24S8 format that nVIDIA's shadow mapping uses.

Plus, why resort to a depth format when you could use a floating point format for the shadow map? With R32F you could render much larger worlds without having to worry about scaling and biasing positions to make them fit into 24-bit space.

Now why doesn't nVIDIA condemn 24-bit here ? Just kidding...

 

Wow--the non-context readers are out in force these days... What I don't care about is DV. If I didn't care about the thread topic--you're right--I wouldn't be posting.

The context of what I said was this: there's nowhere near the architectural difference between nV40 and R4x0 as was extant between R3x0 and nV3x. What does that mean? Well, it means there are no architectural differences which might account for substantial performance differences between nV40 and R4x0 as were responsible for the kinds of large performance differentials between R3x0 and nV3x. That's not the same as saying what you apparently think I said, that nV4x and R4x0 are identical. (Gee, I didn't think I'd have to get so literal about all of this.) Carmack himself stated that the degree of nVidia's OpenGL driver optimizations specifically for D3-engine-specific elements was so excessive that it would negatively impact, in comparison, both nV40's and and nV3x's performance in other games apart from D3. IE, Carmack was telling you *not* to expect this kind of comparative performance from nV40 (and nV3x) outside of D3--if that doesn't set off alarm bells for you then nothing will. Additionally, he stated that he believed future driver revisions would *change* the current numbers (of course.) So even if you don't care to hear it from me, at least hear it from him...

Even more strange is how some people don't comprehend that low-intensity levels of gray are themselves *colors*--what a shocking concept. Yes, it would be strange to think DV had nothing to do with color saturation and contrast, considering I never knew it to do anything else when I used it. Come on, Xmas...