NV3X subpixel precision in 3DMark03

[Joe DeFuria]

You win. You got the quote right. How does that change anything, though?

It's more accurate.

What was the point of you bringing up the quote in the first place? I presume that we should have all just expected NV30 path to be "faster" than the ARB2 path? (Even though those original Carmack claims were on NV30 and R300, not NV35 and R350...so again, I'm not sure what the relevance is of you bringing up the quote in the first place.)

Even ignoring the recent HardOCP/Anandtech results, it still appears that the cards are in parity when it comes to Doom3 _coming from the developers mouth_ who we can assume has hardware and runs everything through enough hoops to find this particular cheat.

Again, not true.

The developer said nothing about NV35 and R350. Only about NV30 and R300. And as we know, there are supposed to be tweaks (specifically for Doom3) for both NV35 and R350 relative to their eariler spins, so we really have little idea how they are "supposed" to match-up.

And this is ignoring the fact that still, R350 is essentially a "purely faster" R300, where as the NV35 is slower in some respects than the NV30 Ultra (lower fill rate).

So now Carmack is involved in the conspiracy?

No, never said that. In fact, my mentioning the ARB2 path was for the purposes of pointing out that EITHER:

1) There are some issues with the ARB2 path...no consipracy here at all, just issues that may be present. This can negatively impact R350 scores, even if unintentional.
2) For some reason, this build of the nVidia drivers don't work with the ARB2 (slower?) path, when they worked OK before, at least on NV30 hardware. This supports the presumption that nVidia is doing all that it can to show NV35 inthe best possible light, by not ALLOWING ARB2 to run on NV3x.

(Though, I think Mr. Pabst was doing his benchmarking on an alpha release of Doom3--not sure though and not sure how exactly that'd affect a cards ability to run a particular path or not)

First I've heard of that? Where is it indicated that Pabst's test/build/demo is anything different than what HardOCP and Anandtech had?

 

[RussSchultz]

There was that leaked copy floating around (the one supposedly leaked by ATI) and people have been doing benchmarking on it. I assumed that was what Toms Hardware was doing.

 

[Joe DeFuria]

There was that leaked copy floating around and people have been doing benchmarking on it. I assumed that was what Toms Hardware was doing.

But why

http://www6.tomshardware.com/graphic/20030512/geforce_fx_5900-10.html#doom_iii_special_preview

NVIDIA and id Software (http://www.idsoftware.com/) were kind enough to give us access to a preview version of id's upcoming game DOOM III and let us benchmark it on a system furnished by NVIDIA. ....

Same as HardOCP / Anand...

 

[RussSchultz]

Ah. I never read Toms hardware, and nobody was talking about that incident (it was always HardOCP/Anandtech). Mea Culpa.

 

[Joe DeFuria]

Ah. I never read Toms hardware...

Can't say I blame you.

 

[jb]

And this map was the one used for the results on HardOCP and Anandtech? No.

You stated that they would not have time to do any sort of this type of optimization. I informed you that they have build and the tools to make a test map way back in Nov. And since it May I think that would give them enough time to do some sort of optimizing yes?

 

[RussSchultz]

And this map was the one used for the results on HardOCP and Anandtech? No.

You stated that they would not have time to do any sort of this type of optimization. I informed you that they have build and the tools to make a test map way back in Nov. And since it May I think that would give them enough time to do some sort of optimizing yes?

The cheat alleged by extremetech requires prior knowledge of the map, and the path through the map in order to have the developer outline which driver cull. Without those two items, if NVIDIA is still increasing its scores due to culling its not a cheat but a bonifide optimization.

 

[Althornin]

The cheat alleged by extremetech requires prior knowledge of the map, and the path through the map in order to have the developer outline which driver cull. Without those two items, if NVIDIA is still increasing its scores due to culling its not a cheat but a bonifide optimization.

well, can you imagine a tool that runs through the benchmark once, and then creates static clip planes based on that?

Then you feed the output of that into your driver, and bam! If you have the right tools, it would only take a short time.


NOTE: The above is merely a possible senario - not nessesarily a plausible one, nor one that i think is "exactly how it works". Dont pull a "Russ" on me and nitpit it to death to try and avoid the possibility of something LIKE it being possible.

 

[AzBat]

The cheat alleged by extremetech requires prior knowledge of the map, and the path through the map in order to have the developer outline which driver cull. Without those two items, if NVIDIA is still increasing its scores due to culling its not a cheat but a bonifide optimization.

Well, if NVIDIA is able to predict the path through the map in 3DMark03 without the tools available to beta members, I'm sure they would be able to predict the path of the preview version of Doom3 in which they created the custom map and timedemo. Simple logic if you ask me.

Tommy McClain

 

[PaulS]

Er, no.

3dMark has been available to the public for quite a while now. They don't need access to the internal versions to "cheat" in the public one. If you notice, they (everyone not nVidia) actually had to use the internal one to find out cheating was going on at all. Without it, no one would have been any the wiser, since it looks identical from a fixed camera path.

The Doom3 timedemo was made by iD themselves - They specifically didn't use the nV supplied one. With no prior knowledge of what the camera would be showing, the possibilites for this kind of culling within D3 are limited, if not impossible. Just some general knowledge about the engine isn't enough.

 

[Ostsol]

The only thing I can think of is that NVidia could have the driver report and keep a record of all polygons that aren't ever in the view frustum. This record would ship with the driver and the next time that benchmark/timedemo is run, the driver would always ignore those polygons.

 

[antlers]

Er, no.

3dMark has been available to the public for quite a while now. They don't need access to the internal versions to "cheat" in the public one. If you notice, they (everyone not nVidia) actually had to use the internal one to find out cheating was going on at all. Without it, no one would have been any the wiser, since it looks identical from a fixed camera path.

The Doom3 timedemo was made by iD themselves - They specifically didn't use the nV supplied one. With no prior knowledge of what the camera would be showing, the possibilites for this kind of culling within D3 are limited, if not impossible. Just some general knowledge about the engine isn't enough.

As long as NVidia had access to the timedemo before they had to supply a driver to the participants, (doesn't necessarily have to have been very long before, as discussed above) they had the opportunity to generate a cheat driver. If they wanted, they could have precalculated shadow stencils for every frame (although that would probably make the driver suspiciously big )

 

[PaulS]

Your point would be valid, were it not for the fact the demo was recorded on the very same day as the benchmarking took place.

Clearing the fastest driver development team ever 8)

 

[Althornin]

Your point would be valid, were it not for the fact the demo was recorded on the very same day as the benchmarking took place.

Clearing the fastest driver development team ever 8)

please show me how a scenario such as the one i discussed above is impossible.
thanks.

 

[PaulS]

Not sure which one you're referring to, but i'll plough on regardless.

Whilst nVidia, given sufficient time, could create a cheat driver for the D3 benchmark, you're missing the cruicial point - They didn't have access to the timedemo at all. It was recorded that very day. Do you think nVidia would tell the likes of [H] to hold off benchmarking for 12 hours whilst they go hack a driver? Do you realise how suspect that would look? No doubt nVidia had Doom3 optimised drivers, but i'm as certain as i can be that there was no aggressive culling going on (a la 3dMark).

Also, your post previous to that was full of random speculation, based on no real facts at all. You talk about how they possibly could have done X, how they might have done Y, how they could maybe do Z. So i'm not sure what you want me to comment on there.

Again, i'm not disputing it can be done. I'm disputing that it was done for this particular Doom3 benchmark.

 

[Althornin]

Not sure which one you're referring to, but i'll plough on regardless.

Also, your post previous to that was full of random speculation, based on no real facts at all. You talk about how they possibly could have done X, how they might have done Y, how they could maybe do Z. So i'm not sure what you want me to comment on there.

Again, i'm not disputing it can be done. I'm disputing that it was done for this particular Doom3 benchmark.

MY post above contained just ONE speculation. The rest of it was detailing HOW it was speculation. sorry if you didnt get that, perhaps you could read it again? And as for which one i am refering too, it would be "the one i discussed above ".
Thanks.

You have done nothing but attempt to invalidate with random speculation. i suggest you be fair and not discard others as easily as you dispense it.

 

[Tahir2]

See this is what happens when you cheat and cry wolf too many times.
No one believes you even when you are due some credit.

NVIDIA I salute you for being seriously stupid.

 

[antlers]

If the timedemo was recorded that very day it would be quite difficult for NVidia to cheat.

 

[PaulS]

Althornin - Really sorry, somehow got you mixed up with Antlers4

In response to YOUR point (), i don't have much of a problem with what you said. I'm not sure they've done it, or that it's that easy to implement, but i guess it's possible. But my point about not believing nVidia did cheat with that particular benchmark still stands.

Once again, sorry for any confusion. I feel like such an ass right now

 

[Althornin]

Once again, sorry for any confusion. I feel like such an ass right now

eh dont.
its just a forum easy to get confused.
I agree that it seems really unlikely - but it does seem possible - just improbable.