nVidias Dawn now works on R3x0

[Luminescent]

By no means is Radeon 9800pro only 15% faster than the FX 5800 utlra or non ultra, running pixel shader 2.0; it is about 200% faster.

 

[Joe DeFuria]

I'm not sure that sentence makes sense. Shouldn't the NV35 be almost twice as fast as the NV30 at running 2.0 shaders, which I assume Dawn is? If the 9800pro is 15% faster than the NV30 (and thats a big "IF"), then wouldn't the NV35 be a lot faster than the 9800pro?

first of all, it has not been established that NV35 "runs shaders" twice as fast as NV30. Initial tests by Dave here at B3D indicate along the lines of 20% clock for clock or so IIRC.

NV30 Ultra is of course 10% faster clocks than NV35, so that would make the total pixel shading advantage in absolute terms, more like 10%.

Second, Dawn is not "just" pixel shaders. There's "other stuff" going on, so even if NV35 is 20% faster at shaders than NV30, that doesn't mean that Dawn will be 20% faster on NV35.

In short, No, we cannot assume that NV35 should be "that much faster" than NV30 when running dawn. Particularly because it's using NV proprietary extensions, which can presumably work around many shading limitations that NV30 has in the first place.

 

[martrox]

I'm not sure that sentence makes sense. Shouldn't the NV35 be almost twice as fast as the NV30 at running 2.0 shaders, which I assume Dawn is? If the 9800pro is 15% faster than the NV30 (and thats a big "IF"), then wouldn't the NV35 be a lot faster than the 9800pro?

Umm...check out "shadermark" results from the nVidia slanted evil [H]OCP preview......

http://www.hardocp.com/article.html?art=NDcyLDEy

Maybe it's because the NV35 is STILL slower - in some cases much slower - that a 9800Pro at running 2.0 shaders.......

 

[kid_crisis]

I'm not sure that sentence makes sense. Shouldn't the NV35 be almost twice as fast as the NV30 at running 2.0 shaders, which I assume Dawn is? If the 9800pro is 15% faster than the NV30 (and thats a big "IF"), then wouldn't the NV35 be a lot faster than the 9800pro?

Umm...check out "shadermark" results from the nVidia slanted evil [H]OCP preview......

http://www.hardocp.com/article.html?art=NDcyLDEy

Maybe it's because the NV35 is STILL slower - in some cases much slower - that a 9800Pro at running 2.0 shaders.......

I agree that IF we were running a pure 2.0 shader there is a chance that the 9800pro will be faster. That is not the case here, this is a specialized demo running on proprietary Nvidia extensions.

Creates higher quality images than the original due to the normalization being done in a fragment program (dp3/rsq/mul) instead of in a normalization cubemap which the FX extensions does directly in hardware.
The OpenGL wrapper adds more overhead, as it has to interperet code calls for Nvidia extensions and map them to ATI/ARB extensions, and yet it still runs faster on the ATI card, due to its more sophisticated pixel shader engine.

It's not an apples to apples comparison here, I'll admit that. Hence comparing Nv35's shadermark scores to 9800pro's is not directly relevant. However, comparing NV35's to NV30's shadermark scores probably is relevant, and as I recall the NV30 had about half the shadermark scores of the NV35.

 

[Marc]

Runs fine here. Performance is excellent, even with 6xFSAA and 16xAF. The hair is a little messed up and some of the lighting seems a bit off.

http://homepage.ntlworld.com/paul.toler/dawn.jpg

Can't complain overall though.

(using 97P w/9800 Cat 3.2 hack)

MuFu.

With the "Ultra" demo you don't have the artefacts.

(9700 Pro with 3.2 or 3.4)

 

[Joe DeFuria]

I agree that IF we were running a pure 2.0 shader there is a chance that the 9800pro will be faster. That is not the case here, this is a specialized demo running on proprietary Nvidia extensions.

Right...and there's no evidence that I'm aware of, that NV35 is anything near 2X the shading performance when using proprietary extensions. . If anything, NV may have worked around the limitations that NV30 has when running DX9 compliant, high precision shaders.

In all liklihood, Dawn, since it was designed to show off the NV30, doesn't utilize shaders in a DX9 "compliant" manner.

It's not an apples to apples comparison here, I'll admit that. Hence comparing Nv35's shadermark scores to 9800pro's is not directly relevant. However, comparing NV35's to NV30's shadermark scores probably is relevant, and as I recall the NV30 had about half the shadermark scores of the NV35.

But if we assume that NV30 to NV35's shader marks are relevant to Dawn performance, then that means that shadermarks for R300/R350 are also relevant. In which case, the R350 clearly beats the NV35 in most cases.

You can't have it both ways.

Like I was implying originally, in all liklihood, you just can't use the performance delta between NV30 and NV35 on synthetic pixel shading tests, to base a performance delta for the Dawn Demo between NV30 and NV35.

There is nothing contradictory about R350 being 15% faster than NV30, and also being "faster" than the NV35 when running the demo.

 

[MuFu]

Runs fine here. Performance is excellent, even with 6xFSAA and 16xAF. The hair is a little messed up and some of the lighting seems a bit off.

http://homepage.ntlworld.com/paul.toler/dawn.jpg

Can't complain overall though.

(using 97P w/9800 Cat 3.2 hack)

MuFu.

With the "Ultra" demo you don't have the artefacts.

(9700 Pro with 3.2 or 3.4)

Yes - you're right, cheers. The 3.4s seem to have fixed the problem in the non-Ultra demo anyway.

MuFu.

 

[Brent]

I'm not sure that sentence makes sense. Shouldn't the NV35 be almost twice as fast as the NV30 at running 2.0 shaders, which I assume Dawn is? If the 9800pro is 15% faster than the NV30 (and thats a big "IF"), then wouldn't the NV35 be a lot faster than the 9800pro?

Umm...check out "shadermark" results from the nVidia slanted evil [H]OCP preview......

http://www.hardocp.com/article.html?art=NDcyLDEy

Maybe it's because the NV35 is STILL slower - in some cases much slower - that a 9800Pro at running 2.0 shaders.......

hey! i'm not nvida slanted, or any other ihv slanted at all! and i don't think the review was THAT evil lol


btw, i thought dawn was OpenGL

 

[Mintmaster]

Wouldn't it be interesting if this wrapper could work with Doom3 so that the NV30 path could be run on R300? I don't know if you'd get anything useful out of it, or if the NV30 path even uses a fragment program instead of register combiners.

Do you think this wrapper would work for other NVidia demos and developer examples?

(Sorry if someone else mentioned this)

 

[martrox]

Brent...guess I should have used the , ok? Because I was being a bit sarcastic.....DOH! However, IF there is a bias in your article - and many do feel that way - it was a bit nVidia slanted: Doom3 anyone? And, If that is true - not saying it is - then IF ATI has shown some abilities above the NV35 - shadermark - then maybe those that show nVidia bias - not saying anyone does - can accept it........

Then again....probably not!

EDIT:Mufu...your are right, what little Brent ran on 3DM03 showed the 9800 as a bit(a very little, tieny wieny, bit) better. I have edited my post to be politically correct.....well, just correct!

 

[MuFu]

Brent hardly used any 3DM tests though. That's their policy.

MuFu.

 

[antlers]

ATI should throw those MIT guys a party.

To reiterate what Joe said above, Dawn was written to the NV30's strengths, so the NV35 may not be much faster than the NV30, clock for clock, on Dawn. If the 9800 Pro beats the NV30, it's entirely reasonable to suppose it beats the NV35 as well.

 

[martrox]

Wouldn't it be interesting if this wrapper could work with Doom3 so that the NV30 path could be run on R300? I don't know if you'd get anything useful out of it, or if the NV30 path even uses a fragment program instead of register combiners.

Do you think this wrapper would work for other NVidia demos and developer examples?

(Sorry if someone else mentioned this)

That is a capital idea....... I'd love to see this! Excellent idea!

 

[antlers]

btw, i thought dawn was OpenGL

It is OpenGL but it uses lots of proprietary NVidia extensions. The MIT guys wrote an OpenGL wrapper for ATI's OpenGL that provides the NVidia extensions (something that ATI is not legally allowed to do).

 

[Joe DeFuria]

The MIT guys wrote an OpenGL wrapper for ATI's OpenGL that provides the NVidia extensions (something that ATI is not legally allowed to do).

Heh: "MIT Guys" = "Covert ATI Interns"

 

[Colourless]

Ok, people, Dawn doesn't like my computer. It just crashes on startup. Strangest thing though, it's crashing due to in invalid instrustion... it's attempting to use SSE code on my old Athlon.

 

[martrox]

Mostly ATI Tech??????

 

[Geo]

The MIT guys wrote an OpenGL wrapper for ATI's OpenGL that provides the NVidia extensions (something that ATI is not legally allowed to do).

Heh: "MIT Guys" = "Covert ATI Interns"

And are the "MIT Guys" legally allowed to do so? Will Reichsmarshall Burke be kicking down doors in Cambridge, screaming "Drop the Fairy!!"?

 

[Joe DeFuria]

And are the "MIT Guys" legally allowed to do so? Will Reichsmarshall Burke be kicking down doors in Cambridge, screaming "Drop the Fairy!!"?

Actually, I don't know if it's legal or not. Not sure how a "wrapper" can be illegal. This question was raised back in the "Glide Wrapper" days too (before Glide went OpenSource)...and I don't know what the outcome is/was.

I could definitely see it being illegal if you were required to install nVidia drivers, but that's not the case here.

IF there's anything possibly illegal IMO, it might be more of a licensing agreement breach by those running the demo. I'm not sure if there are conditions that the demo "must" be run on nVidia hardware (or if such restrictions themselves are legal or not.)

In any case, I would almost expect Rage3D / MIT / ISP host of Rage3D to be getting e-mails about "threats" of law-suits...whether or not such law-suits would have any legal merit, as an attempt to get them to pull it. We'll see how this plays out over the next day or two.

 

[KimB]

To reiterate what Joe said above, Dawn was written to the NV30's strengths, so the NV35 may not be much faster than the NV30, clock for clock, on Dawn. If the 9800 Pro beats the NV30, it's entirely reasonable to suppose it beats the NV35 as well.

The key thing to recognize here is that it's the floating point shader power that is lacking in the NV30. Dawn uses quite a bit of integer precision in its shading (not all, but a fair amount).