[digitalwanderer]

Have you even confirmed if Sweeney is planning to use DST yet Scali?

[Blastman]

Originally Posted by Scali

This theory is not entirely plausible in my opinion.
Namely, 3DMark05 slaughters the GeForce FX even worse than 3DMark03 did.

But only because NV has some illegal “optimizations” on 3DM3. Which they agreed not to do on 3DM5.

I think Ichneumon points may be right on the mark.

The question then arises -- why would ATI cards like the X800pro have a sizable lead over a 6800GT (-10%) without DST? The 3 games are “suppose” to represent a heavy DX9 “standard” workload. Why would the benchmark favor ATI without DST? Is there anything in these 3 games that Futuremark thought favored ATI? If so, then why didn’t they go back and adjust the rendering of the 3 games to make it more neutral. From ATI’s point of view, they could say, well, we work harder on DX9 and deserve that lead. From NV’s POV, they’d say the benchmark isn’t realistic and we’d be forced to go back to what happened with 3DM3.

If 3Dc was put in 3DM5 and DST left out, it’s likely ATI would have a huge lead and NV would be leaving Futuremark again. ATI must have surely knew about this, so maybe they reluctantly agreed to DST to achieve the close parity of the cards. They have a sizable clock advantage with the X800 XT, so maintain a sizable lead at the top of the heap. If this is the case, I hope they don’t regret this down the road if NV closes the GPU speed gap with an update of the NV40 -- or even takes the lead.

[Fodder]

Originally Posted by Scali

This theory is not entirely plausible in my opinion.
Namely, 3DMark05 slaughters the GeForce FX even worse than 3DMark03 did.

A couple of points, someone posted a while back that NV3x suffers even more than NV4x when DST is disabled, and it'd be hard to cater specifically for NV3x (DX8 path? PP?) without people kicking up a stink, given they'd essentially be coding last year's driver cheats right into the benchmark.

[Scali]

Just because games are going to use 3Dc doesn't mean it's an interesting feature for 3DMark05.
3Dc won't greatly improve or decrease performance compared to DXT5.
It could give slightly better image quality in some cases, but that's not that interesting for 3DMark05.

[Scali]

Originally Posted by digitalwanderer

Have you even confirmed if Sweeney is planning to use DST yet Scali?

I never said I was going to. I just vaguely recall Sweeney or someone else saying that UE3.0 will use DST for its shadowmaps.
With Carmack I'm 100% sure he will. With Sweeney I'm not.
Then again, I don't care. I'm smart enough to figure out for myself that DST will be used a lot in the future.
If anyone really needs confirmation from other developers, I suggest they get that themselves, I'm not going to do that for you. I'm sure there are people around here with much better connections to various developers.

[Scali]

Originally Posted by Blastman

But only because NV has some illegal “optimizations� on 3DM3. Which they agreed not to do on 3DM5.

Even without "optimizations" I don't think the 9600XT would score a few hundred points above the 5950XT. So it seems to me that the gap has grown even larger, because more and heavier ps2.0 shaders are used now.

For the rest, it's just a theory, you can't prove it. It could also just be a coincidence that the 6800 and X800 are so close now.
I wouldn't be surprised if it was, because it's very hard to tweak a benchmark's content so that it will give comparable results on completely different hardware. Especially the content of 3DMark05, which contains lots of complex scenes, textures and shaders. Getting everything to look good AND perform the way you want, I think that's nearly impossible.
Especially since the actual hardware hasn't been available to Futuremark all that long. NV40 and R420 weren't available yet (certainly not in the final retail form/speed) at the time 3DMark05's scenes were designed and modeled.

[digitalwanderer]

Originally Posted by Scali

I never said I was going to. I just vaguely recall Sweeney or someone else saying that UE3.0 will use DST for its shadowmaps.
With Carmack I'm 100% sure he will. With Sweeney I'm not.
Then again, I don't care. I'm smart enough to figure out for myself that DST will be used a lot in the future.

No offense intended Scali, but I think you're taking a huuuuuge leap of faith by assuming that just because Carmack does it that way in openGL that everyone plans to do it that way.

Please quit saying "all future games will use DST" until you can come up with at least one example of a game that is not Doom3 or Doom3 engine powered which will use DST, otherwise it just looks like foolish fanboyism.

[Scali]

Originally Posted by digitalwanderer

No offense intended Scali, but I think you're taking a huuuuuge leap of faith by assuming that just because Carmack does it that way in openGL that everyone plans to do it that way.

Your assumption is wrong. I don't assume that everyone plans to use DST because Carmack does. I assume that the reasons why Carmack uses DST make sense for me and others aswell, so we will be using DST aswell.
In fact, even if Carmack wouldn't be using DST, the reasons would still be valid (in case you forgot: the main reasons were that shadowmaps are generally faster and more flexible than shadowvolumes, because they are image-space techniques, not based on geometry, and DST can make shadowmap sampling faster and/or produce higher quality shadows).
Carmack just happens to be the only one who publicly announced this, so it's 100% sure in his case.

Please quit saying "all future games will use DST" until you can come up with at least one example of a game that is not Doom3 or Doom3 engine powered which will use DST, otherwise it just looks like foolish fanboyism.

Excuse me, but Doom3 does not use DST, it doesn't even use a shadowing method that COULD use DST. We are talking about the engine Carmack is developing NOW.
Also, I never said "all future games will use DST", and I am not the one who should be defending the statement.
Futuremark is the one who should be defending it, since they based their decision on this. I am merely explaining how Futuremark came to its decision.
As I said before, if you want to verify this, contact Futuremark or other developers, don't bother me with it.

[martrox]

Originally Posted by Scali

Then again, I don't care. I'm smart enough to figure out for myself that DST will be used a lot in the future.
.

Jeez....don't all of us wish they had your crystal ball.......

If nVidia has a patent on DST - which they do - then I guess, with your ability to see the future you know ATI will have to go out of the video card business........

For someone who sees the future and complains about what a piece of crap AMD is, where were those smarts when you bought a KT133 chipset MB?

For someone who sees the future and knows that nVidia will rule....um, where were those smarts when you bought a ATI 9600?

[Scali]

Originally Posted by martrox

For someone who sees the future and complains about what a piece of crap AMD is, where were those smarts when you bought a KT133 chipset MB?

The KT133 was the newest they had at the store back then, when I asked for the KT800. I already knew that a few years later the AMD Athlon64s would be the fastest CPUs ofcourse.

For someone who sees the future and knows that nVidia will rule....um, where were those smarts when you bought a ATI 9600?

I don't think my choice of videocard in my PC at home has anything to do with any of the statements I've made.

[suicuique]

Originally Posted by Ylandro

Intersting point.
Xbitlabs says that Farcry is using PSM (I have no idea personally), and this certainly looks completely similar to what we're seeing in 3Mark05...

Here is a image I just took on a 6800U (66.51 Drivers) and FC Patch 1.1



Doesnt show the hard Shadow Edges as in the XBitlabs pics.

In Fact when i played FC (some months ago) and used several driver versions i never saw hard shadow edges IIRC (!).


EDIT: I screwed up in a major way. To cut a long story short, i was playing FC on my 6800U under the "ATI Path" (fix by Tommti-systems).
But as this is going major OT, see http://www.beyond3d.com/forum/viewtopic.php?t=16675 for details.
Sorry for the fuckup.

regards, alex

[kyleb]

Wasn't the nvidia shadow stuff added in the 1.2 patch?

[thop]

Hmm i like how it looks, however i think the animation is a bit rusty. Those soldiers could have some better movement and when that big seasnake like thing comes out of the water it is kinda stiff (same when it dives in).

Textures and lighnting ain't everything

[Reverend]

Originally Posted by digitalwanderer

That would be very bad for FM.

Dave has said privately that he is indeed considering pulling out. The underlying principles behind the whole DST-on-by-default has something to do with this but there are other considerations as well, stuff that we/B3D has always been concerned about (like the usefulness of the game tests to B3D, like our suggestions of more specific purely-synthetic tests in 3DMark05 not being implemented, like complaints by FM that we don't use enough of their game tests -- although they later admitted this isn't part of the BDP agreement and withdrew their complaint -- which is something we don't want to be continuously reminded of for every 3DMarkXX version that comes out and other misc. stuff).

Yesterday, I actually asked/told Dave to withdraw.

Originally Posted by Reverend, to Dave,

Regardless of what you decide, I have personally decided not to waste my time and will not provide any contributions if we are still in their BDP (if you decided to continue with our BDP status) and FM asks us for suggestions for future 3DMarks.

Until FM convinces me otherwise, that is my current reaction/thoughts.

[digitalwanderer]

Originally Posted by Reverend

Dave has said privately that he is indeed considering pulling out. The underlying principles behind the whole DST-on-by-default has something to do with this but there are other considerations as well, stuff that we/B3D has always been concerned about (like the usefulness of the game tests to B3D, like our suggestions of more specific purely-synthetic tests in 3DMark05 not being implemented, like complaints by FM that we don't use enough of their game tests -- although they later admitted this isn't part of the BDP agreement and withdrew their complaint -- which is something we don't want to be continuously reminded of for every 3DMarkXX version that comes out and other misc. stuff).

Yesterday, I actually asked/told Dave to withdraw.

Until FM convinces me otherwise, that is my current reaction/thoughts.

This is as big a deal as I thought it was, thanks for the info Rev.

[tEd]

...so who will replace b3d? anand maybe , [H] perhaps 8)

[jimpo]

Originally Posted by martrox

Jeez....don't all of us wish they had your crystal ball.......

Well. That's not only Scali's crystal ball that is showing DST. It's also Carmack's, and Sweeney's, and Futuremarks (as has been repeated).

Futuremark's job is to look into the crystal ball, and make educated guesses on what 2005-06 holds. Now they have guessed that it holds DST. (Have they guessed that it does not hold 3Dc? According to their own claim, no, but they did not have time to implement it). Of course this guess could prove to be wrong, but can't see why we should crucify them before we know whether it does or not. It's not easy to predict the future but I see 3DMark's main purpose as just that, not as much being absolutely "fair" towards all players. If the future won't be fair, then they shouldn't either.

[Ylandro]

Don't forget that this is supposed to be a benchmark.

You can argue that Futuremark thinks that DST is the future. That nobody cares it's patented and not inside DirectX, that other techniques like 3dc are irrelevant, etc etc. As it's all about predicting future behaviour of games, there are no facts, only opinions, and it's difficult to either defend or attack those decisions. (*)

However, in a benchmark you need to compare cards.

Considering that you can always exchange image quality of performance, any video card benchmark should either:
- keep the same image quality for all cards, and compare the performance.
- keep the same performance, and compare image quality.

That is a simple fact. Most take the first approach, as does 3dMark.

However, using DST on NVidia cards, you get different (lower) image quality with different cards. That means you simply can't compare performance anymore. You've justed screwed your benchmark.... It's not comparing apples to apples anymore.

If DST would produce the same image quality (or better), the argument would be different. However, that's not the case. Even though some argue that it's theoretically better, in practice as shown in the 3dMark05, it's clearly producing a lower quality shadow.

If this lower quality would happen in any other situation, everyone would argue that this would be either a driver bug, or a very obvious cheat. The Farcry review of Xbitlabs that has come up in this thread is a perfect example.
Just consider this: Suppose that 3dMark05 was release with DST set off.
Suppose that NVidia made a driver optimization turning it on, with the current difference in image quality.
Wouldn't that be totally unacceptable, both for us, as well as for Futuremark? Wouldn't that be just as worse as any 'optimization' in the last years? Everyone would again acuse NVidia of cheating. And now this "cheat" is put into the benchmark by default!!!


Again: 3dMark is a benchmark. Ofcourse, game developers might choose to use different renderpaths that produce different image quality for different cards. Like Valve does for HL2 for the GeforceFX. You know any FX owner will use the DX8.1 renderpath, while the Radeon owners will use the DX9.0 path.
But that doesn't mean we use those paths, when using HL2 to benchmark the cards, does it? In that case, you want the same image output, to compare apples to apples.

3dMark05 is simply not comparing apples to apples. That is always a bad thing in a benchmark.


(*) You certainly can questions the consistency of Futuremarks decisions what to include, and what not. Not going into that now....

[martrox]

Originally Posted by Reverend

Yesterday, I actually asked/told Dave to withdraw.

That's what I really like about B3d...... This is one of the very few sites that try and deal objectively with this market. Guys, I'm behind you 1000%.

[Scali]

Originally Posted by Ylandro

However, in a benchmark you need to compare cards.

In essence that is what they're doing. One card has DST, the other doesn't. They're comparing that.

However, using DST on NVidia cards, you get different (lower) image quality with different cards.

Different? yes
Lower? subjective

Just consider this: Suppose that 3dMark05 was release with DST set off.
Suppose that NVidia made a driver optimization turning it on, with the current difference in image quality.
Wouldn't that be totally unacceptable, both for us, as well as for Futuremark? Wouldn't that be just as worse as any 'optimization' in the last years? Everyone would again acuse NVidia of cheating. And now this "cheat" is put into the benchmark by default!!!

That's the difference, isn't it? Because Futuremark recognizes that DST will be used in games anyway (without NV having to use driver cheats), they chose to include it in the benchmark.
So this time NV is playing by the rules. And Futuremark is more a gamers benchmark than ever.

But that doesn't mean we use those paths, when using HL2 to benchmark the cards, does it? In that case, you want the same image output, to compare apples to apples.

In case you don't know, you can turn DST off. So like with HL2, you can still choose to render the same path on all cards, if you so please.
You'll never get the same image output though, so it's always debatable whether you really compare apples to apples.

[PowerK]

Originally Posted by martrox

That's what I really like about B3d...... This is one of the very few sites that try and deal objectively with this market. Guys, I'm behind you 1000%.

I agree.

[martrox]

Originally Posted by jimpo

Well. That's not only Scali's crystal ball that is showing DST. It's also Carmack's, and Sweeney's, and Futuremarks (as has been repeated).

Futuremark's job is to look into the crystal ball, and make educated guesses on what 2005-06 holds. Now they have guessed that it holds DST. (Have they guessed that it does not hold 3Dc? According to their own claim, no, but they did not have time to implement it). Of course this guess could prove to be wrong, but can't see why we should crucify them before we know whether it does or not. It's not easy to predict the future but I see 3DMark's main purpose as just that, not as much being absolutely "fair" towards all players. If the future won't be fair, then they shouldn't either.

3DM is a DX9 benchmark, period. FM has stated they would not use vender specific addons, which DST is and, not only that, DST is NOT DX9 so it has absolutely no place in the benchmark, much less be on by default. Also, DST is patented by nVidia, so no one else can use it, has not been, at this point, even been mentioned as being a candidate for future DX9 releases.

IT HAS NO PLACE IN THIS BENCHMARK

If you like it, fine, no problem. But that doesn't mean it should be a default in 3DM05.......

[jimpo]

Originally Posted by Ylandro

However, using DST on NVidia cards, you get different (lower) image quality with different cards. That means you simply can't compare performance anymore. You've justed screwed your benchmark.... It's not comparing apples to apples anymore.

If DST would produce the same image quality (or better), the argument would be different. However, that's not the case. Even though some argue that it's theoretically better, in practice as shown in the 3dMark05, it's clearly producing a lower quality shadow.

True. I personally do not have a problem with DST being used per se. I do have a problem with the fact that it is used so that it produces lower quality. I know that some people (Scali?) state that they like NV's shadows better than ATI's - but to me they look absolutely butt-ugly!

Let's say DST would produce the same image quality - would you accept it's use then? It seems that many people here don't care about the quality of DST, they just object to it's use in principle. Which I do not understand.

What kind of performance impact would NV cards get if DST would be used so that it produced similar IQ to ATI's cards?

[Bjorn]

Originally Posted by jimpo

True. I personally do not have a problem with DST being used per se. I do have a problem with the fact that it is used so that it produces lower quality. I know that some people (Scali?) state that they like NV's shadows better than ATI's - but to me they look absolutely butt-ugly!

That's my problem with it too. Using it is ok, but i have yet to see a screenshot where it looks as good or better. Only one where it looks a lot worse.

[Scali]

Originally Posted by jimpo

Let's say DST would produce the same image quality - would you accept it's use then? It seems that many people here don't care about the quality of DST, they just object to it's use in principle. Which I do not understand.

I think most people who are whining about it, probably wouldn't even have noticed if ATi didn't point it out and insinuated that it's lower quality.
It's the exact same thing as with 3DMark03 all over again, except this time ATi is spreading the FUD, not NVIDIA.

What kind of performance impact would NV cards get if DST would be used so that it produced similar IQ to ATI's cards?

The same impact as not using the DST path on NV hardware. About 10% lower. Problem is, then ATi's quality would be considered lower by most people. Might aswell flip a coin.