[digitalwanderer]

Originally Posted by Scali

Indeed. Both methods give similar quality, at least in the opinion of Futuremark. You can have a preference for either, but ATi is slyly trying to bias the public to prefer their method. I personally prefer NV's method, because as I said before, the edges seem less 'noisy' or aliased.

Originally Posted by A friend of mine trying to explain it to me

The first issue is that this shader and DST work in different ways - The shader takes four point samples (one per texture), DST takes a single bilinear sample from four textures. This causes image quality differences between the two methods - There isn't a 'better' or 'worse' in this scenario however, as one method can look better than the other depending on the scene.

The (probably more important) offshoot of this difference is the performance implications - DST can do its work in a single cycle, whereas the shader-based version takes more than a single cycle (up to four, one for each sample). This adds up to about a 15% performance gain for boards with DST in hardware, in other words nVidia cards automatically gain 15% performance in 3DMark05 by default.

What further makes this matter controversial is the fact that FutureMark have previously stated that they will only ever use features from within the DirectX specification, and will stay away from any vendor specific features. DST isn't in the DirectX 9 spec, and is very much nVidia specific - In fact, nVidia own the patent on this function, as part of a legal settlement with SGI.

[Scali]

Originally Posted by Bouncing Zabaglione Bros.

So what makes Nvidia's DST so special then? If it's okay for everyone's vendor specific solutions to be ignored, then why are Nvidia's included? Oh that's right, because Nvidia said DST was "important".

Close, but not quite.
It's because many game developers said DST was "important".
DST has been around since the GF3, and it's been ignored all this time. If it was just going to be supported because NVIDIA wanted to, then either 3DMark2001 or 3DMark03 would have already supported it.

It was ignored because it offered no advantages at the time. Shadowmapping required too much fillrate and videomemory to be practical. Times have changed, and next-gen engines will support shadowmapping, and this includes DST when available.
3DMark tries to predict the performance of a system in future games. 1+1=?

[digitalwanderer]

Originally Posted by Scali

Close, but not quite.
It's because many game developers said DST was "important".
DST has been around since the GF3, and it's been ignored all this time. If it was just going to be supported because NVIDIA wanted to, then either 3DMark2001 or 3DMark03 would have already supported it.

It was ignored because it offered no advantages at the time. Shadowmapping required too much fillrate and videomemory to be practical. Times have changed, and next-gen engines will support shadowmapping, and this includes DST when available.
3DMark tries to predict the performance of a system in future games. 1+1=?

Problem: "What further makes this matter controversial is the fact that FutureMark have previously stated that they will only ever use features from within the DirectX specification, and will stay away from any vendor specific features. DST isn't in the DirectX 9 spec, and is very much nVidia specific"

Biiiig problem. :?

[Dave Baumann]

Originally Posted by Scali

Anyway, I think the time for feedback ended at the time 3DMark05 went gold, if not sooner. I don't expect Futuremark to make changes to a released product. So either you were too late, or you are in the minority with your complaints, and they will be ignored.

How many times do I need to type the same thing? We were never made aware of these changes prior to the release. We have also said that we don't expect changes changes, but that doesn't mean that we have to agree with what is finally implemented - none of your ranting will change that.

[PatrickL]

scali stop please to hide besides that

It's because many game developers said DST was "important".

and tell us who are theses games developpers, what part of the market they own ?

[digitalwanderer]

Originally Posted by PatrickL

and tell us who are theses games developpers, what part of the market they own ?

Well there is John Carmack, and every game from now on will be based on his Doom3 engine.

[Scali]

Originally Posted by digitalwanderer

Your friend is not entirely correct. There is only one texture being sampled, namely the shadowmap. Both take 4 samples, the difference is where the samples are taken, and how these samples are processed.
The bilinear filter looks smoother because it uses a continuous function to approximate the amount of surface in shadow. So you can have any value between 0..1. The alternative can only determine inside/outside shadow for each sample, giving only a few possibilities: 0 samples inside, 1 sample inside... 4 samples inside. So basically only 5 possible values. Which is what causes the noise or aliasing.

[digitalwanderer]

In fairness to me friend, they were dumbing it down and giving me the Cliff notes version so I could understand it.

[Scali]

Originally Posted by PatrickL

and tell us who are theses games developpers, what part of the market they own ?

I don't know exactly, you'll have to talk to Futuremark.
But I do know that Carmack will, and Sweeney aswell, I believe.
Those are already two of the biggest engine developers.
Just contact other developers and ask them, I'm sure there are more who are using, or are planning to use DST.

[Scali]

Originally Posted by DaveBaumann

How many times do I need to type the same thing? We were never made aware of these changes prior to the release. We have also said that we don't expect changes changes, but that doesn't mean that we have to agree with what is finally implemented - none of your ranting will change that.

As a beta member you actually have the product, right?. How could you not be aware of the fact that it supported DST? Even I knew that, because a preview article in a magazine that I read, hinted at it.
So I find this rather hard to believe.

[digitalwanderer]

Originally Posted by Scali

I do know that Carmack will, and Sweeney aswell, I believe.
Those are already two of the biggest engine developers.

Yes, and if Sweeney isn't it is just Carmack...isn't it.

[digitalwanderer]

Originally Posted by Scali

How could you not be aware of the fact that it supported DST? Even I knew that, because a preview article in a magazine that I read, hinted at it.
So I find this rather hard to believe.

As Rev mentioned in his initial Pulpit post, they only became aware of how DST worked a few days ago.

[PatrickL]

you start to be boring with carmack and engines dues in 2007 when you think you can use them as smoke screen.
Every time we point you out that theses engines will not come out before YEARS, you just change your argument and hide the fact that you simply have no idea about games scheduled with that proprietary technique. We are in 2004 i expect you to talk about games scheduled for the next 18 months, expected time life for that benchmark ( 20 months between 3dmark03 and 3dmark05 ?)

[Dave Baumann]

Originally Posted by Scali

As a beta member you actually have the product, right?. How could you not be aware of the fact that it supported DST? Even I knew that, because a preview article in a magazine that I read, hinted at it.
So I find this rather hard to believe.

I'm not sure I'm at liberty to discuss this or not, but we only had a build a few weeks before the final RC. We neither had any documentation on how it worked or what the specifics were.

Like I said, we had no idea of the rendering properties of the benchmark beforehand, nor that there was would be any differences as to how they were achieved. And I don't really care for your inferance thanks.

[Scali]

Originally Posted by digitalwanderer

Yes, and if Sweeney isn't it is just Carmack...isn't it.

No, but those are the only ones that *I* am aware of. Note that *I* am not Futuremark. Ask Futuremark for the full list.

Besides, Carmack alone is influential enough, if you ask me.

[Evildeus]

Even if i don't really like FM, wasn't always 3DMark being future proof? Being 2006/7 isn't an issue, au contraire, i would say.

[Evildeus]

Originally Posted by DaveBaumann

I'm not sure I'm at liberty to discuss this or not, but we only had a build a few weeks before the final RC. We neither had any documentation on how it worked or what the specifics were.

Like I said, we had no idea of the rendering properties of the benchmark beforehand, nor that there was would be any differences as to how they were achieved. And I don't really care for your inferance thanks.

Well, what's the point of being part of the beta program if you don't have any information/influence on the next product? A genuine question.

[digitalwanderer]

Originally Posted by Scali

Besides, Carmack alone is influential enough, if you ask me.

No, really? I'd have never guessed ya thought that...

[Bouncing Zabaglione Bros.]

Originally Posted by Scali

Close, but not quite.
It's because many game developers said DST was "important".
DST has been around since the GF3, and it's been ignored all this time. If it was just going to be supported because NVIDIA wanted to, then either 3DMark2001 or 3DMark03 would have already supported it.

But it's not in DirectX, so it's no more valid than anything else. Developers are going to use 3DC, but that's not included either.

[Scali]

Originally Posted by digitalwanderer

No, really? I'd have never guessed ya thought that...

Just because I don't always agree with the way he influences things doesn't mean I deny his influence

[digitalwanderer]

Influential, yes; but on a DX benchmark? Mr. OpenGL his ownself?

Uhm, no.

[Scali]

Originally Posted by digitalwanderer

Influential, yes; but on a DX benchmark? Mr. OpenGL his ownself?

Uhm, no.

Why not? The exact same feature is available in OGL aswell. It's the same hardware, doesn't matter which API you use, in this case.
And you can't deny the influence that Carmack had on 3DMark03. Battle of Proxycon was clearly a clone of Doom3's set of rendering tricks

[Dave Baumann]

Originally Posted by Evildeus

Well, what's the point of being part of the beta program if you don't have any information/influence on the next product? A genuine question.

Well, I think we did have an influence, there are a couple of things in there that I think we assisted in steering (at least one of them was a "Thanks, thats a good idea" type of reply, IIRC), however we hadn't seen any builds until recently and at least one thing, I've only reacently noticed, that we were asking for was dropped in the final release.

[jvd]

Originally Posted by Bouncing Zabaglione Bros.

But it's not in DirectX, so it's no more valid than anything else. Developers are going to use 3DC, but that's not included either.

Right and i have a feeling that in 2005 lots of devs will be using 3Dc , i don't see how ati would leave it out of the xbox 2 / xenon and the new nintendo system. So i suspect in the future lots and lots of games will be using it .

[CMAN]

The more true 3DMark becomes to being a benchmark, the less people/gamers will use it. Gamers like tech demos and such and like to show off the features their cards have and gloat over huge numbers and laugh at people that have smaller numbers. Most review sites cater towards gamers too, so they will want 3DMark. This is a question of the intended audience, and I believe Futuremark did the right thing for themselves. If that is how you made your living what would you do?

I think the true decision needs to be where does Beyond3D sit here. Are we more worried about gaming situations or more worried about a trully even playing field needed for a synthetic benchmark that compares particular features within graphics cards?