r420 may beat nv40 in doom3 with anti-aliasing

[Xmas]

nv40@400mhz 12.8 gpixels/sec stencil fillrate while fsaa
x800xt@520mhz 16.6 gpixels / sec stencil fillrate while fsaa

Just to get the terms right:

NV40@400MHz - 12.8 gigasamples/sec color/stencil/Z fillrate with FSAA
R420@520MHz - 16.6 gigasamples/sec color/stencil/Z fillrate with FSAA

The special thing about NV40 is that it can get 12.8 gigapixels/sec stencil/Z-only fillrate without FSAA.

 

[Joe DeFuria]

The special thing about NV40 is that it can get 12.8 gigapixels/sec stencil/Z-only fillrate without FSAA.

Make that 12.8 Giga-Zixels/sec.

 

[Dio]

1.21 jigowatts????

 

[Hellbinder]

Especially since the developers have sold their souls to Nvidia.

Well, recent interview with JC mentioned that NV3x rendering path was dropped and ARB2 is used instead.

He also mentioned that he CHANGED the ARB2 path. He dumbed it down which just so happens to make it perform better on Nvidia.

 

[Evildeus]

Where did he say that?

 

[Snarfy]

Where did he say that?

he gives out these things called .plan files, which i saw a long while ago, but havent seen since...

any1 know where we can find the .plan docs?

 

[Johnny Rotten]

Where did he say that?

I also would like to see where he said that or the basis for Hellbinders 'dumbing down the ARB2 path' comment. However I expect we will be left wanting...

 

[anaqer]

any1 know where we can find the .plan docs?

Just finger him.

 

[Johnny Rotten]

any1 know where we can find the .plan docs?

Just finger him.

John hasnt updated his .plan in about a year. His most recent comments regarding Doom3 and its rendering paths (including the decision to drop the NV3x path) can be found in Reverends mini interview found on the front page of this site (which btw has no reference to any sort of 'dumbing down' whatsoever).

 

[christoph]

Where did he say that?

I also would like to see where he said that or the basis for Hellbinders 'dumbing down the ARB2 path' comment. However I expect we will be left wanting...

well he did say something here:
http://www.beyond3d.com/forum/viewtopic.php?t=12006

Also, based on information you provided to the public (via your .plan files as well as your interviews with us), has there been any significant changes made to the ARB2 path where quality is sacrificed for the sake of performance?

I did decide rather late in the development to go ahead and implement a nice, flexible vertex / fragment programs / Cg interface with our material system. This is strictly for the ARB2 path, but you can conditionally enable stages with fallbacks for older hardware. I made a couple demo examples, and the artists have gone and put them all over the place...

 

[Joe DeFuria]

AFAIK, Carmack never said he "dumbed down" the ARB2 path. He may have indicated that he added PP hints which may not have originally been in there, but that's about it.

However, he DID say that he believes nVidia is "dumbing down" the requests that he makes in the ARB2 path via their drivers. (Some form of app detection, or prior knowledge of empirical data to make behind the scene chages). Carmack has noticed that making seeming small / innocuous changes to his code has from time to time "broken" performance on nVidia cards...that is, they revert to the ARB2 performance before the NV30 path was removed.

Again, my only problem with what's going on, is that it's obvoius to me that the ARB2 path will no longer be able to be used for any kind of meaningful "apples to apples" comparison between ATI and nVidia cards, and yet, I'm sure the majority of reviewers /testers will believe they are the same.

 

[Bjorn]

Again, my only problem with what's going on, is that it's obvoius to me that the ARB2 path will no longer be able to be used for any kind of meaningful "apples to apples" comparison between ATI and nVidia cards, and yet, I'm sure the majority of reviewers /testers will believe they are the same.

I don't see a problem as long as the IQ are the same. And we would never have seen an strict apples to apples comparision anyway since it's FP24 vs FP32.

 

[SteveC]

I may be wrong but at some point didn't JC say that there would be NO benchmarking feature in D3 or that frame rates would be capped at 60? I thought I read that somewhere.

Steve

 

[Joe DeFuria]

I don't see a problem as long as the IQ are the same.

As I said, it is a problem from a standpoint of trying to evaluate technological strengths and weaknesses. I had hope that we would be able to use Doom3 as another synthetic tool, but that's not going to be possible. As far as "which card runs the Doom3 game best", I agree...as long as image quality is largely comparable, it's not a problem for making that evaluation.

And we would never have seen an strict apples to apples comparision anyway since it's FP24 vs FP32.

FP24 = FP23 in that they are both "full precision" by GL standards, AFAIK.

 

[anaqer]

And we would never have seen an strict apples to apples comparision anyway since it's FP24 vs FP32.

Depending on how liberal he's been with _pp hints, could be FP16 vs. FP 24 as well.

 

[Bjorn]

As I said, it is a problem from a standpoint of trying to evaluate technological strengths and weaknesses.

That's true. And i would hope that they remove these optimizations for the NV4X. Which i'm guessing is likely since they might even slow it down.

FP24 = FP23 in that they are both "full precision" by GL standards, AFAIK.

That might be true but they're still not doing the same work. And according to Carmack, there were small quality differences. I'm guessing that that was before he added PP hints though.

 

[Joe DeFuria]

That might be true but they're still not doing the same work. And according to Carmack, there were small quality differences.

Quality differences between FP24 and FP32? Or FP24 and FP16? I remember comments on the latter, but not the former.

 

[bizzoshaker]

SO WHICH CARD WILL PERFORM BETTER RUNNING DOOM3?

 

[Joe DeFuria]

SO WHICH CARD WILL PERFORM BETTER RUNNING DOOM3?

WAIT FOR DOOM3 TO BE RELEASED, AND THEN WE'LL KNOW.

 

[anaqer]

SO WHICH CARD WILL PERFORM BETTER RUNNING DOOM3?

Whichever has a hardware accelerated, fully programmable floating point caps-filter.