No one is using PS 1.4! See inside for details.

[jvd]

IMO ATI was very forward looking with ps 1.4 capability in the 8500

And similarly in a "IMO" kinda post, it isn't ATI that is "forward looking" by including ps_1_4 in their 8500.

This applies to every IHV with their latest chips - NVIDIA wasn't "forward looking" with the first available DX8 chip (NV20).

It is about who is able to release the "latest tech" during a certain time period.

while it may not show a performance increase in games like ut2k3, 3dmark03 shows that it can have positive effects on performance vs. ps 1.1

ps_1_4 is more than just the capacity and ability to reduce the number of passes for a given pixel/texture program, which can affect performance... but performance-per-se, I am sure, is not the only consideration.

Your comment, as I quote it, is a telling example.

[edit]NV30 = NV20

but who was more foward looking at the same period. ATI with the 8500 or nvidia with the geforce 4. Ati added ps 1.4 nvidia did not . The geforce 4s came out after the 8500. So really after nvidia made the geforce 3 they could have put out ps1.4 in the geforce 3 ti line . Okay mabye the specs weren't finalized at that time. But you can't tell me that the specs weren't finalized when the geforce 4s came out. Esp since the geforce 4s came out after the r200. So mabye its not so much foward looking with ati as it is lazyness with nvidia

 

[Randell]

why are peole insisting on being completely blinkered when considering 1.4 on the 8500 in relation to the Gf4.

The Gf4 is a more powerful/efficient card (ignoring AF for the moment). The PS1.4 improves the performance of the 8500 relative to its 1.1-1.3 performance, not by making it perform better than a Gf4 in 3mark03.

Think how much faster at DX8.1 shaders the Gf4 would be if it had PS1.4.

 

[Grall]

Maybe they are just trying to get the list 100% accurate.

It would not look good to list a game that then turned out not to support PS1.4...

You mean, like saying Diablo2 (a DirectX 6 game) using hardware T&L?

Yes, Nvidia did say that. Not sure if they've fixed their list since, but it was there for quite some time, and it always made me chuckle...!

*G*

 

[Evildeus]

]Evildude..

Next time you write my name, please write it as it should.

The 8500 has hardware issues (in the form of some internal limitations) that hold back its performance from what it should be. PS 1.4 cant save it in some cases.

Oh i see, it's so broken that a 8500 can't make it through?

On the other hand, It was shown here and at Rage3d that the R300 series definetly bennefits from the 1.4 optimizations. Adding about 400 points and 3-4 FPS in the heavy Shader tests over PS 1.1 hardware.

Well 3-4 more FPS when it's 1.66/2 times more polygones when using PS1.1, i don't see it as an "optimization".

 

[Randell]

Well 3-4 more FPS when it's 1.66/2 times more polygones when using PS1.1, i don't see it as an "optimization".

at THG Lars shows it as 5-6fps in GT2&3 which equates to ~12%, so I guess you shouldn't optimise for 12% more performance then? Anyway thats on a 9700Pro - other cards will have different impacts, no one has measured the 8500 yet have they becasue it crashes.

 

[Dave H]

at THG Lars shows it as 5-6fps in GT2&3 which equates to ~12%, so I guess you shouldn't optimise for 12% more performance then?

In Ich's tests the 9700 Pro gets 22% and 23% improvements in GT2 and 3. And I think Lars' results were similar. 12% is the improvement to the overall 3DMark03 score.

 

[Evildeus]

Well i don't say that i don't like the increase of framerate. What i say, is that, IMO, 12% more framerate, doesn't seem to me as something optimised, as the Ati board render 2 times less poly in GT3 for exemple.

I like more FPS, and sure the R300 and PS 1.4 gice us some more, but not much more compare to what PS 1.4 effectively does in the game.

 

[breez]

I compared PS1.1 and PS1.4 with my 64MB Radeon 8500LE (275/290). Unfortunately I can't run Troll's Lair because it always crashes.

GT2 PS1.4: 7.2fps
GT2 PS1.1: 6.4fps

12,5% increase.

Maybe someone with 128MB board could test this.

And don't ask where I got the serial for Pro version I'm going to remove Pro version because I can't submit any results to ORB and I really got this for this purpose only.

 

[Reverend]

but who was more foward looking at the same period. ATI with the 8500 or nvidia with the geforce 4. Ati added ps 1.4 nvidia did not . The geforce 4s came out after the 8500. So really after nvidia made the geforce 3 they could have put out ps1.4 in the geforce 3 ti line . Okay mabye the specs weren't finalized at that time. But you can't tell me that the specs weren't finalized when the geforce 4s came out. Esp since the geforce 4s came out after the r200. So mabye its not so much foward looking with ati as it is lazyness with nvidia

I don't work for a IHV nor do I know what goes on in the minds of Mr Jen-Hsun or Mr Orton. There is always a possibility after much going-ons/meetings between MS, the IHVs and developers that NVIDIA made the decision to forego ps_1_4 until the NV30. Maybe they felt the extra transistors are better saved, maybe they felt that ps_1_4 really don't matter when they foresee (foresaw!) that the NV30 would be out six months ago with ps_2_0. Who knows.

 

[Dave H]

NV25 was a refresh part. Adding PS 1.4 support presumably would have meant more futzing with the pixel pipeline than they like to do for a refresh.

 

[Gollum]

Its hard to say exactly why the decision was made not to upgrade the GF4's shaders to PS1.4 level. I think it was the wrong decision on Nvidia's part, but it probably was more cost- and time-effective for Nvidia to simply ignore it and there was probably an unhealthy dose of pride involved too ("we're market leaders they should do what we do not the other way around").

On the other hand we shouldn't forget that Nvidia isn't the only IHV who chose to ignore PS1.4. Neither Matrox, Trident nor SIS or anybody else I might have forgotten has support for anything above PS1.3. So please correct me if I am wrong, but to me it appears ATi is the only one supporting PS1.4 while Nvidia certainly isn't the only one not supporting it. Constantly singling Nvidia out seems a bit one-sided to me. Also, despite the technical superiority the real world gains of PS1.4, in both quality and performance, seem to be negligible in most cases I have seen reported.

IMHO the importance of this whole issue seems kind of blown out of proportion (Nvidia's fault). Either way, in retrospective this whole issue seems to have been the turning point of the tide, ATI is now the innovator being chased by everybody else and they are seemingly on the better track for the immediate future too. Now lets see what it takes for the innovator to turn into market leader again...

 

[martrox]

I guess the real point here is that while ATI has pushed the envelope, nVidia has been content to sit on it....um....laurels. (Hmmmm....maybe thats the real reason for the comparison to 3DFX ) In fact, in a way, this just parallels the TNT2/Voodoo3 era of debate, only this time nVidia is 3DFX! Remember when 32 bit color was the "thing"?

For those that still harbor dislike for nVidia because of the demise of 3DFX, maybe this can be a point of closure......

Revenge is a course best served cold........( ACK! I sound like a trekkie!!!!)

 

[Snyder]

Madden 2003
Tiger Woods 2003
Nascar 2003
NeverWinter Nights
Doom3
UT2003
Unreal 2 (ucomming patch which will also add PS 2.0 support, dunno if a similar patch is planned for UT2003)
New World Order
Independance War 2

IWar 2? I wasn't aware of the game using DX8 pixel shading at all. Or did they add it with a patch? (Just like glide support with the first one)

Seems I have to dig it out again and give it a try...

 

[Ostsol]

IWar 2? I wasn't aware of the game using DX8 pixel shading at all. Or did they add it with a patch? (Just like glide support with the first one)

Seems I have to dig it out again and give it a try...

I-War 2 has two render modes: DirectX 7 and DirectX 8. The latter uses vertex and pixel shaders, though I've had trouble getting it to work. . . It appears that most people have. . .

 

[Reverend]

IMHO the importance of this whole issue seems kind of blown out of proportion (Nvidia's fault).

I agree that it is a bit too "loud" for my liking. I don't agree that this is because of NVIDIA though. We're all responsible for that!

 

[Tonyo]

Guess what? Doom III and UT2003 both (yes, both) use PS 1.4 or equivalent. Obviously, Doom III is an OpenGL program and there isn't such a thing as "PS 1.4", however, it does use PS 1.4 functionality if supported

I think it's very misleading to say that Doom3 uses ps1.4 functionality when it actually doesn't use any of it and it even runs at the same visual quality on pre ps1.1 cards.
As I understand it, ps1.4 functionality is the need for dependent texture lookups, as this is the only difference (barring minimum number of instructions & temp registers in the shader) between ps1.3 and ps1.4.

Doom3 will run at the same visual quality in a ps 1.4 capable card (ATI_fragment_shader path, Radeon 8500) than in a "ps1.1 or even less" capable card (NV_register_combiner path, GF256 to GF4 and Wildcat VP, dunno about Parhelia).

I say "ps1.1 or even less" capable card because NV_register_combiner allows Doom3 to run full quality on Geforce 256 and Geforce 2 cards, which are not ps1.1 (had Carmack written Doom3 in Direct3D, it wouldn't run on those cards at full quality).

Regarding the number of passes, Doom3 will use 5 passes per light in 2-texture-stage cards, 3 passes per light in 4-texture-stage cards and 1 pass per light in 7 texture-stage cards or cards with 7 texture samplers (this is the 8500's case, as this card supports 6 different textures, but 12 texture samplers).
The final proof of the pudding that Doom3 does not require ps1.4 functionality at all, is that Doom3 could run at the same number of passes and same visual quality in Radeon 8500 (ps1.4, 6 texture stages) and Wildcat VP (ps1.2, 8 texture stages). I say "could" because I don't know if a specific NV_register_combiner path for 8 texture stages exists in Doom3.

You could argue that ps1.4 functionality is needed so you don't have to multipass (which is actually not true, as I've pointed out before), but even in that case and concerning speed, Carmack himself stated that in certain cases multipassing on a GF4 was still faster than single-passing on a 8500: "I can set up scenes and parameters where either card can win, but I think that current Nvidia cards are still a somewhat safer bet for consistent performance and quality"

Or in JC's words (referring to GF4 vs. 8500): "The ATI hardware is a little bit more capable, but not in any way that I care about."

 

[Ostsol]

I think it's very misleading to say that Doom3 uses ps1.4 functionality when it actually doesn't use any of it and it even runs at the same visual quality on pre ps1.1 cards.
As I understand it, ps1.4 functionality is the need for dependent texture lookups, as this is the only difference (barring minimum number of instructions & temp registers in the shader) between ps1.3 and ps1.4.

Nitpicks:

Doom 3 certainly does -use- PS 1.4 functionality, though (as you said) it doesn't -need- it to run. PS 1.4 merely provides the capability to render the same effects in much less passes, which would be more significant if the polygon counts were much higher.

PS 1.4 also allows one to use a single texture coordinate to lookup multiple textures, while earlier versions require that multiple texture coordinates be used to lookup multiple textures, even when the coordinates are the same. A relatively minor issue and it comes only as a result of the ability to do dependant texture lookups, but it's there.

 

[Dave Baumann]

Or in JC's words (referring to GF4 vs. 8500): "The ATI hardware is a little bit more capable, but not in any way that I care about."

http://www.webdog.org/cgi-bin/finger.plm?id=1&time=20020211165445

 

[Ostsol]

Or in JC's words (referring to GF4 vs. 8500): "The ATI hardware is a little bit more capable, but not in any way that I care about."

http://www.webdog.org/cgi-bin/finger.plm?id=1&time=20020211165445

Hmm! Thanks for the link!

That quote was made with regard to vertex shaders only, not the entire card.

Regarding pixel shaders:

The fragment level processing is clearly way better on the 8500 than on the Nvidia products, including the latest GF4.

The only problems were some other bugs that made the 8500 run much slower, which is unfortunate since much of the technology was far superior.

 

[Tonyo]

Doom 3 certainly does -use- PS 1.4 functionality, though (as you said) it doesn't -need- it to run. PS 1.4 merely provides the capability to render the same effects in much less passes, which would be more significant if the polygon counts were much higher.

Sure "PS 1.4 merely provides the capability to render the same effects in much less passes", but you don't need ps1.4 functionality to run Doom3 in one pass and as result I'm reticent to qualify Doom3 as a game requiring ps1.4 functionality at all.
An 8-texture-stage ps1.1 capable card can do the rendering exactly the same the 8500 does it. In fact, if Radeon 8500 supported less than 6 textures, it wouldn't be able to do the rendering in one pass even if it's ps1.4 capable. The important issue is not ps1.4 functionality, but the number of different textures and total texture samplers.

Regarding pixel shaders:

The fragment level processing is clearly way better on the 8500 than on the Nvidia products, including the latest GF4.

The only problems were some other bugs that made the 8500 run much slower, which is unfortunate since much of the technology was far superior.

It would be foolish to deny that 8500 technology is superior to GF4, but that doesn't mean that Doom3 needs ps1.4 functionality, as I think I've proved.

Regarding performance of multipass vs. single pass, I don't have any other information but the overall summary on JC's plan

I can set up scenes and parameters where either card can win, but I think that current Nvidia cards are still a somewhat safer bet for consistent performance and quality.

and Beyond3d's interview

Dividing up a fixed texture cache among six textures might well be an issue, though. It seems like the nvidia cards are significantly faster on very simple rendering, and our stencil shadow volumes take up quite a bit of time.

You are right, though, on one of the quotes referring only vertex programs, my bad