Which path will NV40 use in Doom3?

[Hanners]

I'd expect it to use the NV30 path, if only for the sake of UltraShadow support.

 

[Zeross]

I'd expect it to use the NV30 path, if only for the sake of UltraShadow support.

I don't think that UltraShadow is tied to the NV30 code path for various reasons, the most important being that UltraShadow isn't supported on all NV3x chips. Moreover UltraShadow is quite simple to activate the extension interface is juste : void DepthBoundsEXT(clampd zmin, clampd zmax); it doesn't deserve a whole code path to support it.

 

[AndrewM]

We dont know yet if there's a specific NV35 path, but it would seem unlikely at this point.

An NV40 path might be interesting too, with FP16 blending. The different paths do a bit more than just switch shader versions.

 

[jvd]

Simple.


Whichever one makes them faster than ati .


If they need to run the nv30 path to do that then they will .

 

[Richard]

I'd be surprised if it doesn't default to the NV30 path for the extra speed (no matter how small).

Would it be possible to add a "soft shadow" path for NV40 & R420 because of their massively increased fillrate?

Won't happen.

What makes you say that? I think you'd get reasonable performance with either of those two cards by making multiple shadow volumes through jittering point lights.

Btw, does anyone know if there's a limit on how many shadow volumes you can have at any given time? (I seem to recall something about 8 volumes).

jvd: You forget of course that even if the nv40 is made to run by default on the nv30 path you'll still be able to use it on the ARB2 path right?

Btw, (this is directed at B3D staff) have you decided what kind of "path-philosophy" you'll use with the DOOM 3 benchmarks (assuming you'll use it on your p/reviews of course)? For instance, will you benchmark NV3x and R3xx using the same path (ARB2) or go with dedicated paths as HardOCP went?

 

[jvd]

jvd: You forget of course that even if the nv40 is made to run by default on the nv30 path you'll still be able to use it on the ARB2 path right?

Yea but thats not what will be the default and benchmarked path by the majority of sites and publications

 

[ChrisRay]

I'd expect it to use the NV30 path, if only for the sake of UltraShadow support.

Thats what I was thinken. I dont see how the arb code path would have Ultra Shadow extension calls in them. Unless theres an NV40 Arb 2 path? who knows.

I have no idea Carmacks Plan that he gave us is so old. So much has probably by this point in time.

 

[Colourless]

I would have though that the stencil volume drawing code would be seperate to the light filling code (which is what the various paths are for). The stencil code can be shared between all the paths.

 

[jolle]

he seemed pretty impressed by NV40 when commenting it..
So that could imply he was happy about it running fine with ARB2..

Using a special NV40 path might expose more of the new stuff on it, unless ARB2 will automaticly do that, not sure how it works..
But it prolly would be extra work, and the game as it is might not benefit at all from getting access to higher instuction limits or what ever it would expose..

So from that i guess ARB2 would be preferable, specially from Carmacks view..

Only guesses tho, cant back any of it up..

 

[Xander]

The big thing that the nv30 path offers over the ARB2 path is the partial precision support.

Since that is STILL going to be a selling point of the nv40 (it can do partial precision when it wants), it is almost certain we are looking at it using the nv30 path.

Unless you think even the nv40 can competitively run FP32 full-time against the r420 only running FP24?

 

[jolle]

How much can the NV40 acctually gain on running 32x0 in the z passes (and what ever other conditions it does) compared to if it wouldnt?
been wondering about that for a while now... as carmack described it, its a fast pass anyhow, so will the speedgain there amount to anything?..
anyone got any ideas on that?

 

[DemoCoder]

No, the Stencil/Z only pass is several passes. One-two for each lightsource depending on technique.

 

[Sxotty]

First jvd I think you are not being to hard on J.C.

Second I don't think NV/ATi would have to pay anything, or I mean they WOULD NOT unless they were trying to sell it, they could mod it for free and give it away. Anyway I think it is likely he might put an nv40/r420 path in just b/c they will both be able to do things that neither the 9800 or fx series could realistically do due to speed issues.

 

[jolle]

No, the Stencil/Z only pass is several passes. One-two for each lightsource depending on technique.

ok tnx, but what i wanted to find out is a rough estimate on how it will affect overall FPS when doing those in 32x0 opposed to everything in 16 pipes?
If the "colorless" passes makes up a very small part of the overall time spent on 1 frame, it sounds to me as if the impact would be rather small..

and with nvidia making that possible, is there any other gameengine that can draw benefits from it?
does it automaticly benefit from it when your using volume shadows?
or do devs have to adapt the rendering to it?

I have trouble grasping why Nvidia would use that appoach if it wasnt useful in more then 1 game engine.. that is why im asking..

 

[Xander]

I have trouble grasping why Nvidia would use that appoach if it wasnt useful in more then 1 game engine.. that is why im asking..

First off, it's not something game-specific. If you tell the driver to do something like a stencil/z-only pass, it will simply do it in 32x0 mode. If you give it some color to work with, it will drop down to 16x1 - no input from the developer required.

It's also hardly Doom3-specific. Doom3 is the first engine to make extensive use of stencil operations (the lighting for this game will be quite something new), but it won't be the last. Further, the Doom3 engine itself will be directly used as the core for many, many other games.

(As an aside - I wonder, when you buy the Doom3 'engine' - what render paths are included? IE., how much of the engine do you GET?)

 

[Dave Baumann]

(As an aside - I wonder, when you buy the Doom3 'engine' - what render paths are included? IE., how much of the engine do you GET?)

As far as I can see the render paths are just different shader models for the unified lighting scheme. If you don't have the different render paths you don't get the lighting scheme - I guess the real question is what is the Doom3 engine without it?

 

[Dave Baumann]

FYI - the most accurate "Doom3 rendering mechanism without being Doom3" test is actually 3DMark03's Game Tests 2 & 3. Look to relative performances in those test to see how what types of differences are achieved. (These don't support UltraShadow, but I have my doubts that gains massive amounts, at least not in comparison to the double z/Stencil)

 

[Richard]

jvd: You forget of course that even if the nv40 is made to run by default on the nv30 path you'll still be able to use it on the ARB2 path right?

Yea but thats not what will be the default and benchmarked path by the majority of sites and publications

That's why I'm interested in knowing B3D's opinion on this. Personally I think they should be benchmarked in the highest quality path (i.e. ARB2) regardless of what path they default to.

Dave, any opinion on this matter yet?

 

[Neeyik]

Well purely on the basis of comparable benchmark results, one would have to use the standard ARB2 path; just in the same way that we normally use projector shadows in Splinter Cell.

 

[radar1200gs]

Well purely on the basis of comparable benchmark results, one would have to use the standard ARB2 path; just in the same way that we normally use projector shadows in Splinter Cell.

Whats the point of using a comparable path if most of the users will run the game in a non comparable path due to performance gains?

The "there is only one right way to run a program" nonsense has to stop. Its bullshit.