FX and PS 1.4, DX9 tests?

[Ante P]

So now that the card is out can we have the final verdict on this?
To me it seems as if the card doesn't support it considering how much better results the 9700 Pro gets in the "Advanced Pixel Shader" test in 3D Mark 2001 SE.

BTW I don't understand why anyone didn't use ShaderMark 2.0?
ATi and nVidia also has lots of DX9 "generation" tests (demos) and then there's the DX9 SDK, to put it short: why the hell haven't we seen any DX9 benchies?
I sure would like to see some Doom 3 benchies, personally I dunno if I'd dare to publish them myself though.

 

[LeStoffer]

ATi and nVidia also has lots of DX9 "generation" tests (demos) and then there's the DX9 SDK, to put it short: why the hell haven't we seen any DX9 benchies?

Beats me too. I thought that nVidia would release at least some kind of demo test that would show off NV30's shader ops power, but instead things are awkwardly eerie on the DX9 front.

 

[Ante P]

ATi and nVidia also has lots of DX9 "generation" tests (demos) and then there's the DX9 SDK, to put it short: why the hell haven't we seen any DX9 benchies?

Beats me too. I thought that nVidia would release at least some kind of demo test that would show off NV30's shader ops power, but instead things are awkwardly eerie on the DX9 front.

http://www.tommti-systems.com has a DX9 Pixel Shader benchmark based on Atis treasure chest demo. Of course it's an ATi demo but I dunno how optimized it is for ATi hardware, considerin git's DX I think it would be a pretty fair comparison

 

[Brent]

ATi and nVidia also has lots of DX9 "generation" tests (demos) and then there's the DX9 SDK, to put it short: why the hell haven't we seen any DX9 benchies?

Beats me too. I thought that nVidia would release at least some kind of demo test that would show off NV30's shader ops power, but instead things are awkwardly eerie on the DX9 front.

http://www.tommti-systems.com has a DX9 Pixel Shader benchmark based on Atis treasure chest demo. Of course it's an ATi demo but I dunno how optimized it is for ATi hardware, considerin git's DX I think it would be a pretty fair comparison

i'll check it out...

 

[Humus]

I'd love to see results of running my phong demo on both GFFX and 9700. It's not a benchmark, but at least is has a framerate counter
I'm most interested in how fast it is in running floating point shaders, todays games are less interesting.

 

[Joe DeFuria]

I'm most interested in how fast it is in running floating point shaders, todays games are less interesting.

Heh...maybe for you. Tell that to everyone who's bought into the "future proof" argument before.

Of course, I'm interested in all aspects of these cards -- current and "potential future" performance.

 

[demalion]

You know, I wonder how fast the GF FX would run Humus' Phong demo?



Maybe some reviewer will listen eventually.

Say, what was that you were saying about thinking you could get some benefit out of "128-bit" precision for your mandelbrot set generator?

 

[Humus]

Heh...maybe for you. Tell that to everyone who's bought into the "future proof" argument before.

Well, I'm not like everyone else who do stuff like playing games. I waste my time coding instead

Say, what was that you were saying about thinking you could get some benefit out of "128-bit" precision for your mandelbrot set generator?

That's something I'm very interested in too.

 

[Brent]

You know, I wonder how fast the GF FX would run Humus' Phong demo?



Maybe some reviewer will listen eventually.

Say, what was that you were saying about thinking you could get some benefit out of "128-bit" precision for your mandelbrot set generator?

Hailing Frequencies Open

 

[Ante P]

I'd love to see results of running my phong demo on both GFFX and 9700. It's not a benchmark, but at least is has a framerate counter
I'm most interested in how fast it is in running floating point shaders, todays games are less interesting.

I'll be glad to use it for my review.

btw I just realized that ShaderMark isn't a very good benchmark at all, scores fluctuate WAAAAY too much

perhaps the author can make it more reliable, I can't remember his nick here at B3D though

 

[DegustatoR]

PS2.0 aren't working in 42.63 Dets. And VS2.0 is buggy. But all DX8.1 functionality (including PS1.4) is there.

 

[Ante P]

PS2.0 aren't working in 42.63 Dets. And VS2.0 is buggy. But all DX8.1 functionality (including PS1.4) is there.

I wonder why not a single review even tried the newer 42.70 drivers. I don't mean for the whole review but just "for fun".

So PS1.4 is working properly then, good to hear. (Especially since I've been called an idiot for claiming that the NV30 supports it )

 

[Brent]

PS2.0 aren't working in 42.63 Dets. And VS2.0 is buggy. But all DX8.1 functionality (including PS1.4) is there.

I wonder why not a single review even tried the newer 42.70 drivers. I don't mean for the whole review but just "for fun".

So PS1.4 is working properly then, good to hear. (Especially since I've been called an idiot for claiming that the NV30 supports it )

My initial impressions on DX9 support were wrong

Using the DX9 SDK I opened the Caps Viewer to take a look at the caps the 42.63 drivers expose:

As you can see they do have PS/VS 2.0 support in DX9

 

[Ante P]

PS2.0 aren't working in 42.63 Dets. And VS2.0 is buggy. But all DX8.1 functionality (including PS1.4) is there.

I wonder why not a single review even tried the newer 42.70 drivers. I don't mean for the whole review but just "for fun".

So PS1.4 is working properly then, good to hear. (Especially since I've been called an idiot for claiming that the NV30 supports it )

i just tried them, no dx9 support yet

How long can you hold on to that sample of your. There's other fish in the sea even though we're small ya know. just kidding

well no DX9 supports kinda sucks
does it support that functionality in Open GL then?

 

[KimB]

PS2.0 aren't working in 42.63 Dets. And VS2.0 is buggy. But all DX8.1 functionality (including PS1.4) is there.

I wonder why not a single review even tried the newer 42.70 drivers. I don't mean for the whole review but just "for fun".

So PS1.4 is working properly then, good to hear. (Especially since I've been called an idiot for claiming that the NV30 supports it )

Interesting, but apparently the low performance is also there. I wonder if there's any possible way to force PS 1.0 use in the 3DMark2001 advanced pixel shader? Just for kicks, I would like to see how the FX fares with PS 1.0 vs. 1.4, at least in that test (most particularly compared to how the Radeon 8500 and 9700 score in the same two tests...)

 

[Ante P]

PS2.0 aren't working in 42.63 Dets. And VS2.0 is buggy. But all DX8.1 functionality (including PS1.4) is there.

I wonder why not a single review even tried the newer 42.70 drivers. I don't mean for the whole review but just "for fun".

So PS1.4 is working properly then, good to hear. (Especially since I've been called an idiot for claiming that the NV30 supports it )

Interesting, but apparently the low performance is also there. I wonder if there's any possible way to force PS 1.0 use in the 3DMark2001 advanced pixel shader? Just for kicks, I would like to see how the FX fares with PS 1.0 vs. 1.4, at least in that test.

you could always force PS1.0 on the 9700 Pro to see if that evens out the performance differences
PixelShaderVersion: 10
in the registry will do the trick

I dunno if nVidia has a similar settings?

 

[KimB]

Actually, Ante P, I was more thinking that the 8500 and 9700 might both show performance increases, while the FX doesn't, or shows much less of a performance increase when using PS 1.4. I think that would make the most sense...though it will depend heavily on the architecture and drivers of the respective cards.

 

[Fred]

I do buy the future proofing argument somewhat. GF3s still perform at satisfactory levels in most apps. I'm sure the 9700 will be just fine for Doom3, and the next 2 generation of games thereafter.

The Kyro argument comes a little to mind, even though it had a fillrate advantage in some of the games of the time over the MX, the latter still can perform at 640*480 with some eyecandy turned on in say certain UT maps. Those users who went the latter route, are benifiting from the card now. It seems a bit treasonous for tech heads on this board who have to have the newest, shiniest vid cards, but a lot of end users only upgrade when they get new computers (eg 2-3 year upgrade cylces).

Granted, in this day and age, at least some level T&L is now a defacto requirement of nearly any entrance level card , so the future proofing argument holds a little less weight than it once did with vanilla rasterizers like the V5 and Kyro.

 

[KimB]

Granted, in this day and age, at least some level T&L is now a defacto requirement of nearly any entrance level card , so the future proofing argument holds a little less weight than it once did with vanilla rasterizers like the V5 and Kyro.

I don't think so. I think that in the coming years, we'll see even more of a split between DX9 and sub-DX9 parts. Remember that back with the original T&L cards, the CPU didn't have much of a problem keeping up. I don't think that's remotely the case today. There are many things that you can do now in DX9 (mainly requiring higher precision) that were simply impossible to do satisfactorily in DX8. And as for the vertex part of the pipeline, well, the GPU really is tons faster than the CPU is now, so there's no longer the argument about just offloading the calculations (which can be done more effectively with DX9), but of drastically increasing overall geometry throughput, even when the CPU is not occupied with anything else.

 

[JD]

Someone needs to write or find hw independant shader code and let gffx run it. Existing benchmarks like 3dmark, nature, etc. might be bias towards one ihv. Basically, provide an open source shader code.