Late, noisy, HUGE!!! - makes 17k Mark...

[Nick[FM]]

Yes it uses Pixel Shader 1.4, BUT also looks the same on Pixel Shader 1.1 so again I ask whats advanced about it that is allows the older version to look the same.

It maybe looks the same, but with PS1.4 it renders the water surface in 1 pass.

According to ATI on their Ocean screen saver PS 1.1 couldn't render it with any speed, there is two Pixel Shaders being used, the water and the sky..something similar should have been done here.

So what you mean is that we should have done something special to favour some company? But that's again against what you are striving at.. That's controversial. You say that we mustn't favour anyone, yet hint that we should favour some company. So, how is it now?

I am using my head ,I don't consider a DX 8.1 requirement being put with EMBM 'supporting it'...sorry.

Errrr.. Please explain. I don't get this.

If DX 8.1 PS 1.4 is a feature then so is DX8.0 PS 1.1 so remove nature from scoring.

PS & VS were introduced in DX8.0. They were kind of the key elements for DX8.0. In DX8.1 the PS was only updated to 1.4. There's a big difference. Again, I ask you to use your head.

 

[DemoCoder]

There is no pixel shader that can be rendered in PS1.4 that can't be rendered with PS1.1. You seem to think that PS1.4 is a "feature" like EMBM, that if you don't have it, you don't have it, and anything that uses that feature (say EMBM), must be switched off. (and no, a software fallback for PS1.0 is nonsense. There is a difference between 2-passing something, and REFRASTing something, which effectively switches off your 3D card)

The only real difference is performance. Imagine if the Advanced Pixel Shader test was written in HLSL without regard to PS version and that a runtime compiler generated PS1.1 and PS1.4 versions. Are you going to claim that it is unfair that a fallback PS1.1 version gets generated?


What's "advanced" about a pixel shader should be the level of complication of its shading algorithm, not some contrived program that is supposed to be "1.4" only, since it's unlikely you'll be able to come up with one that can't be trivially converted to 2-pass.

Futuremark should write all of its future shaders in DX9 HLSL to avoid being accused of bias. Oh, but then maybe they'll be accusations that MS's compiler is somehow biased towards NVidia like the REFRAST accusations.

Bah.

I think Sharkfood has lost whatever little credibility he had in this thread.

 

[Doomtrooper]

] It maybe looks the same, but with PS1.4 it renders the water surface in 1 pass.

No point in calling it advanced then is it ...PS 1.4 could do the same with future..but that would be too easy.

So what you mean is that we should have done something special to favour some company? But that's again against what you are striving at.. That's controversial. You say that we mustn't favour anyone, yet hint that we should favour some company. So, how is it now?

Woha back up the truck PS 8.1 is a DX 8.1 feature, just because ATI was the only card supporting you are not showing any favortism in any way, but showing full DX 8.1 support...BTW reverse that and your same mentality can be used for PS 1.1 since their was only one manufacturer that supported on 2001's release

Errrr.. Please explain. I don't get this.

Ummm ok, you have the DX 8.1 Feature Test stuffed in with EMBM, DOT 3 bumpmapping...are you calling DX 8.1 in a whole a feature ??

PS & VS were introduced in DX8.0. They were kind of the key elements for DX8.0. In DX8.1 the PS was only updated to 1.4. There's a big difference. Again, I ask you to use your head.

Why is there is a difference in thinking at futuremark whenever you guys see fit, there was a significant change in DX 8 going from 1.1 to 1.4 (PS 2.0 is built off PS 1.4) yet Futuremark felt it doesn't deserve to be counted into scoring, or in other words you support whatever you feel is necesarry, tell that to Microsoft.

 

[Doomtrooper]

There is no pixel shader that can be rendered in PS1.4 that can't be rendered with PS1.1. You seem to think that PS1.4 is a "feature" like EMBM, that if you don't have it, you don't have it, and anything that uses that feature (say EMBM), must be switched off. (and no, a software fallback for PS1.0 is nonsense. There is a difference between 2-passing something, and REFRASTing something, which effectively switches off your 3D card)

Nobody said that, whats the point of rendering a scene when you are getting 10 fps (PS1.1 multipass) vs 25 fps or 30 fps (singlepass)...what does 3Dmark put in the bottom right hand corner (framerate is life).

Like ATI's reply to David Kirks comment about PS 1.4 not being a big step forward (kind of like Worms comments here).

http://www.rage3d.com/board/showthr...=20&highlight=david kirk&pagenumber=2

That's just blatant misinformation. I'm shocked it was even published. For example, the ocean water screen saver has two shaders: water, clouds. Neither can possibly be done on a GF3 (or pre 1.4 pixel shader part) regardless of how many passes you break it down to. The math simply cannont be expressed without 1.4 pixel shaders. We wrote them, we know.

The statement of 1.4 being a small step is just plain wrong. The 8500 can do 12 texture samples and 16 generalized math ops in a single pass. The GF3 isn't even close. Not a single developer I've spoken to doesn't think ps1.4 is leaps beyond the GF3's pixel pipe.

Alex V, ATI

 

[Rancidm]

There is no pixel shader that can be rendered in PS1.4 that can't be rendered with PS1.1.

I'm no graphics programmer, but one person who worked on the Ati demos doesn't seem to agree with you.

http://www.rage3d.com/board/showthr...ighlight=alex AND we know we 1.4&pagenumber=2


quote:
--------------------------------------------------------------------------------
Originally posted by Zarich
Everything that is done in ps1.4 can be done without ps1.4. ps1.4 is just suppose to make things work a bit simpler. So yes the geforce 3 can run all those demos. Whether they run them better or not .. well thats another story.
--------------------------------------------------------------------------------


That's just blatant misinformation. I'm shocked it was even published. For example, the ocean water screen saver has two shaders: water, clouds. Neither can possibly be done on a GF3 (or pre 1.4 pixel shader part) regardless of how many passes you break it down to. The math simply cannont be expressed without 1.4 pixel shaders. We wrote them, we know.
..................

And later...

quote:
--------------------------------------------------------------------------------
Originally posted by Zarich
thats interesting alex.. cause there are people at nvidia claiming they have already duplicated your demos on the geoforce 3. I assume you would know though... since you coded it.
--------------------------------------------------------------------------------


That's almost funny. It's physically impossible to do most of the shaders we've written on a GF3. Period. Whoever said that is blowing smoke. Looking at just the screen savers, all (except for the bubbles) just can't be done without 1.4 pixel shaders.

Alex V, ATI

Surely you've seen these posts before? I'm sure it's been refered to a number of times over the years.

Are you making this statement from personal experience?....or has Alex's statement been disproven in the past?

edit: I guess doomtrooper beat me to the punch

 

[DemoCoder]

Doomtrooper,
PS1.4 is not a "feature" that makes it possible to render things impossible with PS1.0. 1.4 is not a significant change over 1.0 like 2.0 is. I don't get your PS2.0 "based off PS1.4" remarks. PS2.0 doesn't have phase markers. PS2.0 separates texture registers into samplers and coordinate registers. In what way is it "based" off 1.4 anymore that it was based off 1.0 that preceeded it?

 

[Sharkfood]

worm-

There's no use talking to you. You have no idea what you are talking about, and all your "proof" is only something you have made up in your head.

followed by Nvidiacoder-

I think Sharkfood has lost whatever little credibility he had in this thread.

Im sure glad you two "peas in a pod" subscribe in the "make shallow, empty statements" rather than actually trying to retort or prove otherwise with facts, examples or reasoning. lol.

I made the point pretty clear. I also thank Rancidm for posting the reality as explained pretty clearly from ATI.

There are also at least a few logical thinking people here that will stop and scratch their head at a PS1.4 benchmark that not only runs perfectly on non-DX8.1 compliant hardware, but in most cases equals or exceeds it in performance.. on lesser performing hardware.

So, now what's next? Worm and NVidiacoder roll eyes, self declare that all this is imagined/made up then spin fingers in the air? Well, at least the "air" part is right.

 

[Sharkfood]

On the FUD and nonsensical NVIDIA propaganda posted by Democoder-

What's "advanced" about a pixel shader should be the level of complication of its shading algorithm, not some contrived program that is supposed to be "1.4" only, since it's unlikely you'll be able to come up with one that can't be trivially converted to 2-pass

Same goes with water effects in Nature without using pixel shaders. What stopped them then? Hmmm.. Maybe because the only card available with PS was a Geforce3? Hmm?

This is the double standard very clearly illustrated by your example. Obviously, any effect can be downplayed by simply emulating in such a way to perform best on a lower common denominator baseline.

That is the point. The Advanced Shader Test is nothing of the sort. It wasnt written with the complexity level you have described to show any bearing of advantage. Was it done so intentionally? One has to decide this for themselves, but given the overwhelming number of examples in the past, it does strain credibility.

 

[DemoCoder]

Nobody said that, whats the point of rendering a scene when you are getting 10 fps (PS1.1 multipass) vs 25 fps or 30 fps (singlepass)...what does 3Dmark put in the bottom right hand corner (framerate is life).

The point is to render a scene on two cards and see how fast it runs. HOW it is rendered is frankly, irrelevent. The final output is what matters.


With Doom3, no two video cards are going to render it in the same number of passes. But gamers don't care how many passes it takes, they care how fast it runs

So in your opinion, 3Dmark should simply put up an error message if you try to run it on a GF3 saying "Sorry, you can't run this"? Even though it COULD run it? I mean, don't you want to know how much more efficient a PS1.4 shader performs compared to a PS1.0 2-pass shader? The fallback simply makes the GF3 LOOK BAD, why are you complaining?



I'll remember your comment Doomtrooper when NV30 specific super-long shaders come out and you are arguing for R300 multipass rather than just failing to work at all!

 

[Nick[FM]]

No point in calling it advanced then is it ...PS 1.4 could do the same with future..but that would be too easy.

Well, IMHO, yes. It is more advanced than the other pixelshader test we have/had in 2001.

PS1.4 could do the same with future - Sorry but not sure what you mean by that?

Woha back up the truck PS 8.1 is a DX 8.1 feature, just because ATI was the only card supporting you are not showing any favortism in any way, but showing full DX 8.1 support...BTW reverse that and your same mentality can be used for PS 1.1 since their was only one manufacturer that supported on 2001's release

Again, I'm sorry but I don't understand you. I mean, what you mean.

Anyway, when 2001 was released, we had 1 DX8.0 Game Test - true. Was it our fault that there was only one company with full DX8.0 compliant hardware on the market when MS released DX8.0 - certainly not. You can't blame us for that! We wanted to do a DX8 benchmark, and we did. It's not our fault that not all companies release new hardware when a new DX version is out.

If we would now release a test based on, let's say DX11 ( 8) ), you probably would call us biased towards that company which releases DX11 compliant hardware first. Am I correct? At least that seems to be your logic.. :?

Ummm ok, you have the DX 8.1 Feature Test stuffed in with EMBM, DOT 3 bumpmapping...are you calling DX 8.1 in a whole a feature ??

Ok now I see. Again, we had a couple of choices. To leave out PS1.4, which would have been dumb, or then make it as a feature test. As I said 1001 times earlier, it was not an option to change the scoring system. To add 1 more game test would have done it, and editing the code (to implement PS1.4) would/could have done it. We didn't want that to happen.

Why is there is a difference in thinking at futuremark whenever you guys see fit, there was a significant change in DX 8 going from 1.1 to 1.4 (PS 2.0 is built off PS 1.4) yet Futuremark felt it doesn't deserve to be counted into scoring, or in other words you support whatever you feel is necesarry, tell that to Microsoft.

Edit: I re-read my post and it didn't make that much sense.

Ok, A new try. PS1.0/1.1 was introduced in DX8.0, and was considered a big leap forward in real-time graphics. Never had graphics looked so good. Or, it was now possible to make better looking, and more real graphics. It was pretty big stuff back then. Some time later, MS released DX8.1, which introduced an updated version of PS1.1, the PS version 1.4. It didn't have such an impact on the public. Why? Because it didn't actually bring that much new. Ok, I was told that some shaders are only possible with 1.4, and not with 1.1, but most are. 1.1 can do them, but in 2 passes. That affects both performance, and the IQ accuracy.

This brings me back to the same thing again. When DX8.0 was out, there was only PS1.1. We released a game test with DX8.0 PS1.1. Later when DX8.1 was out, we wanted to do something with it, but the only possible way was to do it in a feature test. We didn't want to affect the score.

Am I making any sense now? :?

 

[DemoCoder]

Taken to childish name calling eh ATIFood?

On the FUD and nonsensical NVIDIA propaganda posted by Democoder-

What's "advanced" about a pixel shader should be the level of complication of its shading algorithm, not some contrived program that is supposed to be "1.4" only, since it's unlikely you'll be able to come up with one that can't be trivially converted to 2-pass

That is the point. The Advanced Shader Test is nothing of the sort. It wasnt written with the complexity level you have described to show any bearing of advantage. Was it done so intentionally? One has to decide this for themselves, but given the overwhelming number of examples in the past, it does strain credibility.

Interesting. You call my comment nonsense (that "advanced" should mean "advanced lighting algorithm" not "Pixel Shader version = X") but implicity agree with it.

The fact is, someone should take a relatively advanced shader like "Procedural WOOD" and code it up in PS2.0 and 1.4/1.1 if possible. That's how to benchmark a shader.

Starting out with the premise "I will make a shader in PS version X that runs better than version PS Y" is a bad way to benchmark. You've already biased your result.

You should use real world shaders, possible converted RenderMan shaders, and simply see how fast each card can run them. What PS version you coded them in is irrelevent. You should use the most optimal rendering path for each piece of HW.

Anyway, extraordinary claims require extraordinary evidence. The burden of proof for your conspiracy-nut crap is on you, since you are the one continually running around these forums making such absurd claims in the face of the developer themselves telling you otherwise. Go ahead and fine the last time I posted such similar nonsense against ATI or some ATI software, review site, or benchmark.

 

[Sharkfood]

With Doom3, no two video cards are going to render it in the same number of passes. But gamers don't care how many passes it takes, they care how fast it runs

Way to make a completely irrelevent example! We are talking a synthetic benchmark here, not a game. Unless you are stipulating that people are online playing 3DMark and enjoying it as a game.

With synthetics, it's absolutely crucial the comparisons are apples to apples, not only in methodology, but in output result. If either/or are out of sync, then it's not a truly useful tool for comparison purposes.

That's another problem with the Advanced Pixel Shader test- it's allegedly not running the same passes depending upon the IHV being tested, therefore it's resultant scores aren't a good basis for apples->apples comparison. Add to this how PS1.4's main benefit is single pass where lesser PS1.0/1.1 is multiple passes, yet wasnt designed to symbolize this reality and it all makes pretty common sense.

Remember- nobody is argueing that a PS1.4 shader cannot be implemented in a backwards compatible method with PS1.0/1.1 to look identical. The argument is that such an approach should yield higher performance with a single pass versus more passes, not the other way around (all things being equal). Unfortunately, the APS on 3DMark2001SE was specifically designed in such a way to strategically pick an example and methodology that would not yield this reality with PS1.4. The end result is- it illustrates a fictional outcome: one of multiple passes performing the same or in excess of single pass PS1.4.

I'd simply say anyone that doesn't question this reality, especially in light of ATI's comments (and the comments of anyone that has coded in DX8.1), truly gives away the true purpose of making such stipulations.

 

[DemoCoder]

With regards to that ATI message, I'd like to see some specifics. I saw that claim before with regards to per-pixel phong, and it is simply false. Ghost of Envy had a per-pixel phong demo using register combiners up on B3D before the 8500 even came out. The fact that ATI coders can't imagine how the GF3 can run it is different than it being physically impossible. It's hard to prove what's impossible to do once you bring multipass/render-to-texture into the equation.

 

[nAo]

In the spirit of this forum I believe one should give a proof of some PS1.4 shader that can't be done on PS1.0-1.3 in one or multiple passes, or a proof that all the shaders one can write on PS1.4 hw can be written also on PS1.0-1.3 hw.
Quoting 'proofs' from NVIDIA or ATI doesn't seem a good way to improve this discussion.
And about single pass PS1.4 shaders being faster in the general case versus multipassed PS1.1 shaders..IIRC John Carmack doesn't seem to agree on this. But I can't find the quote at the moment..

ciao,
Marco

 

[Sharkfood]

Starting out with the premise "I will make a shader in PS version X that runs better than version PS Y" is a bad way to benchmark. You've already biased your result.

I'm glad we agree on this. This is exactly how the APS in 2001SE was designed. It's results are clearly evident of such.

You should use real world shaders, possible converted RenderMan shaders, and simply see how fast each card can run them. What PS version you coded them in is irrelevent. You should use the most optimal rendering path for each piece of HW.

I totally agree here also. This is much better than hand-picking the most utopian example imaginable for one particular IHV. (cough.. advanced pixel shader test *cough*).

The burden of proof for your conspiracy-nut crap is on you, since you are the one continually running around these forums making such absurd claims in the face of the developer themselves telling you otherwise.

Scroll up to ATI's developer comments above. Also, go read any of John Carmack's commentary on single-pass versus multiple passes. That is truly ignoring the claims in the face of the developers themselves telling you as such.

 

[DemoCoder]

Shark,
you're simply unhappy with the result of the benchmark. You want them to code an example that shows 1.4 beating 1.1. But multipass isn't always slower than single pass. It depends on a number of factors.

In that regard, the APS test is bad. They didn't pick an example to show 1.4 advantages in some scenarios. But to call this NVidia bias?

Do you think the programmers sat around like this:

Prog#1: Let's create a DX8.1 PS1.4 test to make ATI look bad!
Prog#2: Muahhaha! Great idea. Ok, how bout this little water shader. It runs fastest on PS1.1
Prog#1: Wow, ingenous! We'll say it's an advanced shader, and GF3 will run it faster! Hahahahha!
Prog#2: Call Nvidia for the paycheck! I'm retiring early!


OR, do you think it was simple laziness/incompetence?

I mean, DX8.1 just came out. How many programmers know how to write shaders to truly exercise that pipeline? You think it's trivial to write a shader that runs faster in single pass than multipass?

It's not your criticism of the APS that I disagree with. It's attributing this to some deliberate pro-Nvidia plan that I find ludicrous. Maybe if John Carmack or Tim Sweeney wrote 3dMark, I'd feel this way, but there are way too many poorly coded 3D engines out there to attribute this to deliberate intent. Most likely is that 3dMark is just a poor benchmark and by an accident of history, it favors NVidia (written for DX8, Nvidia had the first DX8 hardware out)

It might also be likely that 3dMark2003 is ATI biased because ATI's had the first DX9 hardware out for 6 months. Another accident of history.

 

[Doomtrooper]

So in your opinion, 3Dmark should simply put up an error message if you try to run it on a GF3 saying "Sorry, you can't run this ?" That's just blatant misinformation. I'm shocked it was even published. For example, the ocean water screen saver has two shaders: water, clouds. Neither can possibly be done on a GF3 (or pre 1.4 pixel shader part) regardless of how many passes you break it down to. The math simply cannont be expressed without 1.4 pixel shaders. We wrote them, we know.

The statement of 1.4 being a small step is just plain wrong. The 8500 can do 12 texture samples and 16 generalized math ops in a single pass. The GF3 isn't even close. Not a single developer I've spoken to doesn't think ps1.4 is leaps beyond the GF3's pixel pipe.

Alex V, ATI"? Even though it COULD run it? I mean, don't you want to know how much more efficient a PS1.4 shader performs compared to a PS1.0 2-pass shader? The fallback simply makes the GF3 LOOK BAD, why are you complaining?

No, exactly where did I say that, I said the Future test could have its fallback to PS 1.1 just like their advanced Pixel Shader tests my preference..don't put words in my mouth.
Another point DX8 > DX7 yet DX7 users had the "Sorry, you can't run this ?", DX 8.1 > 8 ........ maybe you can tell me why its ok for one Dx revision and not another to be 'ok' to state "Sorry, you can't run this ?".
I don't want to hear DX8 was more of a improvement then DX7..yada..yada..a revision is a revision.

 

[Bambers]

Stop pounding out conspiracies. I get more than double the framerate using PS 1.4 than PS 1.1 on my Radeon.
(refers to Ati techdemos, 3dmark 2k1 and some other various non-at-madonion demos)

You sure you haven't screwed something there. I lost a bare 10fps in the APS test with my 8500 using 1.1/1.3

 

[Doomtrooper]

In the spirit of this forum I believe one should give a proof of some PS1.4 shader that can't be done on PS1.0-1.3 in one or multiple passes, or a proof that all the shaders one can write on PS1.4 hw can be written also on PS1.0-1.3 hw.
Quoting 'proofs' from NVIDIA or ATI doesn't seem a good way to improve this discussion.
And about single pass PS1.4 shaders being faster in the general case versus multipassed PS1.1 shaders..IIRC John Carmack doesn't seem to agree on this. But I can't find the quote at the moment..

ciao,
Marco

A test of the non-textured stencil shadow speed showed a GF3 about 20% faster
than the 8500. I believe that Nvidia has a slightly higher performance memory
architecture.

A test of light interaction speed initially had the 8500 significantly slower
than the GF3, which was shocking due to the difference in pass count. ATI
identified some driver issues, and the speed came around so that the 8500 was
faster in all combinations of texture attributes, in some cases 30+% more.
This was about what I expected, given the large savings in memory traffic by
doing everything in a single pass.

A high polygon count scene that was more representative of real game graphics
under heavy load gave a surprising result. I was expecting ATI to clobber
Nvidia here due to the much lower triangle count and MUCH lower state change
functional overhead from the single pass interaction rendering, but they came
out slower. ATI has identified an issue that is likely causing the unexpected
performance, but it may not be something that can be worked around on current
hardware.

The other comment was I believe the famous High Poly bug that was later fixed (which wasn't a high poly bug)

 

[DemoCoder]

Well, too bad, you are going to hear it. Major version changes are usually fundamentally different and revisions are usually bug fixes or small additions. If MS thought DX8.1 and DX8 were sufficiently different, they would have called it DX9. DX8 introduced SIGNIFICANT changes to the DX7 API. DX8.1 didn't add much at all.


If you have a version number like X.Y, usually changes in X signify significant new functionality, non-backwards compatible API changes, and require significant QAing. Changes in Y are bug fixes and minor updates. In my entire 18 years of experience in the software industry, I've haven't seen very many breaks of that paradigm. Some companies go deeper into X.Y.Z formats, or even X.Y.Z.buildnumber, but its irrelevent.