The importance of Shader Model 3.0

kinney · Apr 23, 2004

This is a repost of analysis I made in another thread that I thought was good, concise information that most might be interested in reading about Pixel Shader 3.0 and Vertex Shader 3.0.
A bit of this is speculation, but all signs point to the general ideas being accurate.
Unless ATI somehow has SM3.0 support but that is highly doubtful or they likely would have already made that known after the NV40 launch.

I feel the forum needed something like this posted as everyone is so "uncertain" about Shader Model 3.0 and what it does/will do/benefits. I am trying to bring some certain conclusions to be made from available information.

Without further ado..

1. PS3.0 offers great efficiency and will offer much higher performance than PS2.0

2. Effects will be easier to implement, and those same effects less stressing on the hardware through PS3.0

3. PS3.0 offers full precision (32bit floating point), while 2.0 does not (24bit). Anyone who has studied Nvidia's new pixel shaders will know that it has PLENTY of power to use this as well.

4. VS (vertex shader) 3.0 will bring exponentially greater image quality over VS2.0 and where we'll see the biggest IQ improvements. This feature finally brings us displacement mapping that DX9 originally promised.

Simply put, Shader Model 3.0 is a very big deal and will remain the standard until 2006 when Longhorn and DX10 is released.
I should also note, that as I've stated in the past, Nvidia moving to 32FP with the NV30 was a wise move.
Now, the move to PS3.0 was much easier from an engineering standpoint and their 2 years of experience working with 32FP hardware will likely give them huge performance gains over the ATI PS/VS3.0 hardware whenever that gets here.

As my list of games shows, I expect a quick death to partial precision 24FP and PS2.0.
While VS3.0 will bring us huge IQ improvements.

Questions/Comments?
I could have something wrong, as I am but one man. But these are my predictions/analysis after reading multiple sources about the technology.

____________________________
:evil:

RESIDENT FORUM WARRIOR :evil:

I just neutered the cat.
Now he's a liberal.

Vimp · Apr 23, 2004

The thing I'd have to paticularly point out my disagreement with is when you said...

I should also note, that as I've stated in the past, Nvidia moving to 32FP with the NV30 was a wise move.

Just because moving to 32FP early with the FX series gives them a head start in their work of their next gen hardware which actually makes use of it due to some additional experience in no way makes what they did wise. It isn't wise to implement something into a card when the card isn't even capable of making use of it. And then as a result having everyone who buys their products miss out on what it could have been capable of had they gone with 24FP instead of the 32FP which was obviously to early for its time with the FX.
It may have been that they simply messed up and had to go with it since they didn't predict very well how things would turn out in the future. But even if that was the case its still far from calling them wise in having done it.

_______________________________
Canada
Asus A7N8X-X, Athelon XP 2500+ Barton,
Samsung 1gb DDR400, MSI GeforceFX5900 XT.
Aquamark=36077 3DMark03=5322

kinney · Apr 23, 2004

You have a point, it might not have been wise.. but it IS hard to fault a very forward looking company.
It appears to have been one of those moves that ended up appearing wise.. when it was more merely a forward-thinking decision.. that ended up helping them immensely.
I wouldnt classify it a "bumbling mistake" though by any means as pushing the limits can only be commended IMO.

____________________________
:evil:

RESIDENT FORUM WARRIOR :evil:

I just neutered the cat.
Now he's a liberal.

Vimp · Apr 23, 2004

Well actually a "bumbleing mistake" sounds more along the lines of what I was thinking. I mean think about it. Putting 16FP and 32FP in the FX could be likened to providing a medium setting that it can easily do and a super high setting that it can't do and not providing anything inbetween (24FP). So this results in the consumer being forced to use the medium setting which dosn't take advantage of the power of their system and not the inbetween setting that their hardware would have been able to more sufficently been able to do had it been included or the super high setting that goes to far beyond what their hardware can do.

_______________________________
Canada
Asus A7N8X-X, Athelon XP 2500+ Barton,
Samsung 1gb DDR400, MSI GeforceFX5900 XT.
Aquamark=36077 3DMark03=5322

darko21 · Apr 23, 2004

Re:You have a point, it might not have been wise..

Yes it was not wise but a hinderence in performance. still is even today!

Re: but it IS hard to fault a very forward looking company.

Easy when they screw up like that.

Re:It appears to have been one of those moves that ended up appearing wise..

only to a fanboy

Re: when it was more merely a forward-thinking decision.. that ended up helping them immensely.

so if ati released 256 precision but you only got 2fps in doom3 you would be patting them on the back as a great future forward looking company?

Re: I wouldnt classify it a "bumbling mistake"

Funny everyone else does.

Re: though by any means as pushing the limits can only be commended IMO.

yeah we are used to your opinions biased as she gets to the point of meaningless drivel fanboyism.

If I glanced at a spilt box of tooth picks on the floor, could I tell you how many are in the pile. Not a chance, But then again I don't have to buy my underware at Kmart.

kinney · Apr 23, 2004

I dont know man.. I think that 16FP was what shouldve been stuck with if dev's thought performance was too low.

I dont necessarily disagree with what your saying.
As I'm not "against" 24FP.
Theres honestly not that much IQ difference going from 16-24... and not that much IQ difference going from 24-32 IMO.. but my point was that 32FP will be the standard for quite some time (at LEAST 2006) and is what is by all standards of the term, "full precision".
I think its the mark to shoot for, being full precison. And I think that if performance IS a concern.. 16FP will be chosen for best performance as it is best used for ALL of the previous gen cards.
Kinda leaving 24FP in the cold.

Either way, with the NV40 core Nvidia has gotten great performance in even 24FP games and I think will be able to now fully push PS3.0 and 32FP with the power of the NV40.

This is more speculation, and I actually said this a LONG time ago (years) on this forum... but we -might- see the NV30 cored stuff like what you have JUMP in performance if everything goes to DX9C (32FP PS3.0) as possibly some of its OTHER (remember DX9+? :tongue: ) unused features will be unveiled to enable more performance. I really dont know, its a theory I've held for some time and its getting ever more near to either becoming gospel or false.

We'll see.. but as my original post details.. the cards are stacked highly in Nvidia's favor.. from either the bumbling mistakes or their just looking too far ahead. However you want to look at it. :/

____________________________
:evil:

RESIDENT FORUM WARRIOR :evil:

I just neutered the cat.
Now he's a liberal.

Vimp · Apr 23, 2004

Well as you know I'm a fan of Nvidia so I certainly hope things work out for them better this year as you so optimisticly preach. However on the issue of 16FP and 24FP being too little to notice, I was of the impression that the banding and blockiness found in FarCry was due to it being 16FP where as the 24FP Radeon looked smooth and nice. If that is the situation then I'd have to say that 24FP must have a noticable difference over 16FP. However I maybe wrong about that issue and perhaps the banding and blockiness wasn't because of it using 16FP but perhaps it was because it was using 32FP in the partial precision optimisation mode that was perhaps causing it to miss certain marks resulting in 0FP in spots, or something like that. In which case I'd have to agree that 16FP and 24FP are perhaps to close to notice the difference by much.

As far as your speculations of the NV300 series making good use of its having 32FP due to SM3's better performance, I'd have to say that I'm skeptical since SM3 is something that the GF6000 series had to be specificly desinged to make use of which the GF5000 series simply doesn't have. So I don't see that 32FP becoming usefull for the FX in the future if my understanding of things is correct.

_______________________________
Canada
Asus A7N8X-X, Athelon XP 2500+ Barton,
Samsung 1gb DDR400, MSI GeforceFX5900 XT.
Aquamark=36077 3DMark03=5322

kinney · Apr 23, 2004

I'd like to expand on this statement a bit.

This is more speculation, and I actually said this a LONG time ago (years) on this forum... but we -might- see the NV30 cored stuff like what you have JUMP in performance if everything goes to DX9C (32FP PS3.0) as possibly some of its OTHER (remember DX9+? ) unused features will be unveiled to enable more performance. I really dont know, its a theory I've held for some time and its getting ever more near to either becoming gospel or false.

I'll point to some specifics for you since you are one of my favorites.

The original NV30, hated as it might be.. I trumpeted for its forward looking features.
The NV30 Pixel Shaders contained both dynamic flow control and dynamic branching.. these are some of the main reasons why the NV40 will be more efficient than the R300 and all its offspring (including R420), will not have this feature as it will not have PS3.0 support.

This is our "DX9+" that is now revealed better and ratified in the form of DX9.0C.
Predication was also a NV30 hardware feature that is used in PS3.0, it also speeds up shaders.
Arbitrary swizzle is the 3rd formerly unused NV30 features, yet again, and efficiency improvement.
Also, the NV30 had a maximum 1024 instructions (well above the 512 minimum for PS3.0) while the R300 had 32max.

As far as "DX9+" vertex shaders, they also had dynamic branching, predication and dynamic flow control.. all which improve performance/efficiency. As well, it had a maximum of 65K maximum VS instructions (R300+ has 1024max).

So as you can see, the NV30 stuff like what you have has most of the Shader Model 3.0 features.
I believe those older cards will see an extra boost that is unexpected from most who are blind sided by this information. The good thing is, the NV30 cores are missing most of the elements that they wouldnt have the power to handle, but have the Shader Model 3.0 parts that pretty much ONLY improve performance. Not that some of this stuff won't improve IQ as developers will likely push the limits further because of the added efficiency. Or, simply boost all NV30 based scores. But what they are not capable of, that stuff is out of their league.. and better left to the NV40 generation.

But its interesting, and good news for sure for anyone with a NV30 based card.

____________________________
:evil:

RESIDENT FORUM WARRIOR :evil:

I just neutered the cat.
Now he's a liberal.

Vimp · Apr 23, 2004

Well thats certainly an interesting take on it and I guess time will tell. However I certainly wouldn't comend them and recomend their hardware based purely on such speculation since its deffinitly not for certain.

_______________________________
Canada
Asus A7N8X-X, Athelon XP 2500+ Barton,
Samsung 1gb DDR400, MSI GeforceFX5900 XT.
Aquamark=36077 3DMark03=5322

TheGreatGrapeApe · Apr 23, 2004

Well first, Satement 3 is false. You're confusing minimum with an upper bound.

Interesting with all these advanaced feature the FX series still couldn't perform. But of course it's about the features and not about actually making use of them, right?

Checkboxes and potential perrformance don't always offer what one expected. Once again I refer you to the FX5200 series stellar DX9 support.

And dynamic branching while sounding great is a feature that even nV says may not be useful in the GF6800 series due to the fact that it will actually reduce performance for pixel shading implementation. For vertex shading it should offer far greater immediate benifits.

Also remember that many aspects you're talking about, like static/dynamic branching, swizling, etc can be found in the <A HREF="http://msdn.microsoft.com/library/default.asp?url=/library/en-us/directx9_c/directx/graphics/reference/assemblylanguageshaders/pixelshaders/ps_2_x.asp" target="_new">PS2.X standard</A>. Try and keep that in mind if you decide to refine your PR blitz.

The big questions are exactly what areas will be supported by the various hardware, and what performance and IQ benifits will the differences in that functionality bring to gamers. Since you can't answer that right now (neither can I or anyone else here [except maybe Borsti/Lars]) then I don't see any point in even discussing anything's supposed 'importance'.

We'll see when the playing field actually has something to compare to.

- You need a licence to buy a gun, but they'll sell anyone a stamp (or internet account) ! - RED GREEN GA to SK :evil:

kinney · Apr 23, 2004

Well many parts of my theory are certain.. the gains are what are in question but I think theres 0% chance of it negatively hurting stuff based on the NV30 like yours.
If it does, they will disable it for those in game. Simple.

I never said anything about recommending then now due to such information.. how do you guys pull that kind of stuff out of posts? :frown:

____________________________
:evil:

RESIDENT FORUM WARRIOR :evil:

I just neutered the cat.
Now he's a liberal.

darko21 · Apr 23, 2004

Re: I'll point to some specifics for you since you are one of my favorites.

I want to be one of your favorites... that would be so wonderful. oooooohh just the thought of being one of kinneys chosen favorites..

If I glanced at a spilt box of tooth picks on the floor, could I tell you how many are in the pile. Not a chance, But then again I don't have to buy my underware at Kmart.

kinney · Apr 23, 2004

Try and keep that in mind if you decide to refine your PR blitz.

WTF man.. seriously. lol, you crack me up.

Like I said in the other thread, I shouldve been a bit more specific.. as I know about your exact PS2.0 standards.. but over time I have at least become accustomed to the R300+ defining PS2.0.. that might be wrong but hey.. my points still stand.

edit- to make it simpler for you lol.. just do this
1. PS3.0 offers great efficiency and will offer much higher performance than PS2.0

Instead of that add a "MINIMUM" before PS3.0 and "ATI's implementation of" before PS2.0..
there we go.. media blitz refined..
lol

____________________________
:evil:

RESIDENT FORUM WARRIOR :evil:

I just neutered the cat.
Now he's a liberal.Edited by kinney on 04/23/04 03:40 AM.

kinney · Apr 23, 2004

One thing I should add, that the stuff I pointed out is all good stuff... it is clearly not in the best interest of a hardware company to be behind.. certainly the NV40 has the power to swing with whatever R420 brings to the table.. plus more hardware features.

And while being long on hardware features will never HURT you, being SHORT definitely has a chance of doing so.

All in all, I think you'll be pleasantly surprised with your 5900 when games take advantage of PS/VS3.0.
Which will be soon, NV will ensure that. Heck, they've already done <A HREF="http://www.firingsquad.com/hardware/nvidia_geforce_6800_ultra/page24.asp" target="_new">it!</A>

____________________________
:evil:

RESIDENT FORUM WARRIOR :evil:

I just neutered the cat.
Now he's a liberal.

cleeve · Apr 23, 2004

I've seen some more info on PS3.0 patch on the Inq and gamespot:

Apparently, there is no image quality enhancement offered by PS3.0 in the Far Cry case, but there is a "possibility that the 3.0 path might be slightly faster in same cases on some hardware"... like the 6800.

So apparently the majority of out-of-the-gate PS3.0 features will be minor speed enhancements for cards that can take advantage of the feature.

PS3.0 may be capable of more advanced stuff, but it won't be implemented in that capacity for a while because 1) there isn't enough of a user base to bother with PS3.0-only features yet and 2) the first gen PS3.0 hardware is probably not fast enough to run heaps of it.

In an interview with multiple developers at gamespot, some said you can't see a difference between PS2.0 and 3.0 while others said there are some advanced features that may become used that would be a difference.
All seemed to agree that the increase between PS2.0 to 3.0 is not a great as the increase between 1.0 and 2.0.

________________
Radeon 9500 PRO (hardmodded 9500, o/c 340/310)
AthlonXP ~2750+ (2400+ @2.2Ghz)
3dMark03: 4,055

kinney · Apr 23, 2004

Did you read my original post? :frown:

I stated that PS3.0 is mostly about speed (not that IQ gains cant and wont be gotten from it). While VS3.0 appears to have the large IQ enhancing features.

I think you'll see SM3.0 implemented in all TWIMTBP games (most of the good ones), like I pointed out theres already a good list. And everyone will move to DX9C (if just for the PS3 performance enhancements) when its released.. most of the performance enhancing features only require compiling through a different compiler.

Note that the entire SM3.0 is/or already has implemented this on these games at very least. That means both VS/PS3.0!

Lord of the Rings, Battle For Middle-earth
STALKER: Shadows of Chernobyl
Vampire: Bloodlines
Splinter Cell X
Tiger Woods 2005
Madden 2005
Driver 3
Grafan
Painkiller
FarCry

The double edged sword to what you are saying is that while 1.x to 2.0 was a large jump for IQ, it on the same hand took forever to get implemented into games.
From 2.0 to 3.0 on the other hand the titles are hitting the ground MUCH faster than the preceding technology.

Most of this I suspect is due to NV and their huge TWIMTBP program, but in the end it matters not. The end result is SM3.0 is a great feature, and already is assuredly not a checkbox feature.
The arguments good and healthy, but I am afraid the end result cannot be anything but my original posts conclusions that SM3.0 is definitely worth it and definitely the future now that MS has ratified it in DX9C.
And like I said to Vimp, most of the performance enhancing functions were already within the NV30.. who knows what we might see happen there.

____________________________
:evil:

RESIDENT FORUM WARRIOR :evil:

I just neutered the cat.
Now he's a liberal.

cleeve · Apr 23, 2004

Yeah, I read your post...

To me, all things being equal (IQ), if all PS3.0 delivers is a boost in shader speed, then it would come down to which card delivers more frames per second. PS3.0 itself would be irrelevant in itself because in that case it would simply be a feature that increases FPS. (i.e. if the 6800 has PS3.0 but delivers fewer frames per second than the X800, then what good is PS3.0?)

If PS3.0 offers advanced effects that PS2.0 cannot handle, and those effects are quickly implemented into games in development, then PS3.0 becomes far more compelling.

________________
Radeon 9500 PRO (hardmodded 9500, o/c 340/310)
AthlonXP ~2750+ (2400+ @2.2Ghz)
3dMark03: 4,055

kinney · Apr 23, 2004

Well, I suppose that my viewpoint after reading the technicalities on PS3.0 is that it makes ATIs task of trumping NVs speed a much harder task to accomplish without SM3.0.

And like I said, either in this thread or another.. the NV40 is fast enough in brute force that even in a "best case scenario" for ATI hardware (24fp DX9.0) I dont think they'll top the 6800s.
Thats my theory of course.. but just theorizing from the preliminary benchmarks, and assuming from ATIs silence a lack of SM3.0 I just dont see that happening.

As far as what you're saying about PS3.0 making advanced graphics that the R420 cant match with PS2.0, I think that example pales in comparison to the massive graphical quality that Vertex Shader 3.0 with its displacement mapping will bring.

Its pretty tough to match the advanced features of the NV40 with brute force... ATI basically needs to bring a sh*tload of it.
And when NV caught ATI with their pants down on SM3.0 they also successfully surprised them as to the brute force they brought to the table.
Double whammy.

____________________________
:evil:

RESIDENT FORUM WARRIOR :evil:

I just neutered the cat.
Now he's a liberal.

cleeve · Apr 23, 2004

Displacement mapping certainly is tasty, but I'm not sure that VS2.0 can't do that already.

As far as what's going to happen when the X800 comes out, what can I say except we shall see what we shall see.
I'm not making any predictions on this one... honestly, I can see it go either way at this point. But if the X800 PRO is getting 15k in 3dMark'03 like it's being reported to, I'd say the 6800 has a fight on it's hands.

________________
Radeon 9500 PRO (hardmodded 9500, o/c 340/310)
AthlonXP ~2750+ (2400+ @2.2Ghz)
3dMark03: 4,055

confoundicator · Apr 24, 2004

Theres honestly not that much IQ difference going from 16-24... and not that much IQ difference going from 24-32 IMO.. but my point was that 32FP will be the standard for quite some time (at LEAST 2006) and is what is by all standards of the term, "full precision".

Uhhh, I though FP stood for floating point, not full precision, as in a 32 bit floating point decimal number. As far as the IQ difference I have no idea.

He that but looketh on a plate of ham and eggs to lust after it, hath already committed breakfast with it in his heart. -C.S. Lewis

The importance of Shader Model 3.0

Distinguished

Distinguished

Distinguished

Distinguished

Distinguished

Distinguished

Distinguished

Distinguished

Distinguished

Champion

Distinguished

Distinguished

Distinguished

Distinguished

Illustrious

Distinguished

Illustrious

Distinguished

Illustrious

Distinguished

Share this page