Doom3 -- hopefully another Carmack interview with me

[Reverend]

Okay, I've played quite a bit of the game and my "conclusion" regarding the graphics of the game is that it is... Good. Perhaps the most impressive graphics aspects of the game, to me, are the character graphics.

Now that many of you have the game and hence seen the game in all its glory, perhaps now is a better time than ever to ask John about certain things regarding the game's graphics. From the simple (why the low rez textures?) to the slightly more advanced (the shader for see-through glass is reasonably big... why?). Sort of like a post-mortem.

Post your questions here, I'll pick 'em. 3 days from today before I send them off to John. Please be nice to me and help me save time -- no discussions here, just questions.

Of course, there's no guarantee John will answer every question!

 

[Dave Baumann]

Actually Rev, now that D3 is done and dusted, I'm more keen to know about his next generation engine, seeing as its been said that he's already looking. Can you see if he's willing to give up any details on that - what he's looking at in terms of requirements, how it differs from the Doom3 engine, etc., etc.

 

[DarN]

Will you be including Humus' tweak (or something to that effect) in a future patch, perhaps taylor the engine more to current hw specs?

 

[AlNom]

*fixed*

just trying to point that Carmack does talk about stuff... so people can adjust their questions if it's already been covered. Forget it.

 

[Neeyik]

Alstrong - Rev specifically requested NO DISCUSSION. Please can people just post questions for the interview...otherwise the evil mod pixie might have to come out to play.

 

[Reverend]

Will you be including Humus' tweak (or something to that effect) in a future patch, perhaps taylor the engine more to current hw specs?

Already asked him that regarding Humus -- he never replied.

 

[Reverend]

Actually Rev, now that D3 is done and dusted, I'm more keen to know about his next generation engine, seeing as its been said that he's already looking. Can you see if he's willing to give up any details on that - what he's looking at in terms of requirements, how it differs from the Doom3 engine, etc., etc.

Hehe... that's a big ask

This is something he'll (likely) reveal in a future .plan update, not in an interview with a media outlet. I can try but I know where I stand with John.

I think we know roughly where he's heading or where his concentrations lie in a purely graphics sense (look at all his .plan updates from Quake3 to the latest) but it's unlikely he'll tell me much (if he even would at all) in terms of specifics -- he may want to go back on his word and patent some of them before Creative does

ps. Sorry... now I'm doing the discussion!

 

[thop]

1.) How does the Doom3 engine handle open space? I mean real open space, where you can see the horizon, not the little outdoor snippets in D3. And depending on the answer: Was that one of the reasons why there are no real outdoor levels or was it really a game design decision?

2.) What exactly did delay the game so much from its original release date (2003)?

 

[AlNom]

[ question ] In your keynote address shown at Quakecon, you mentioned shadow buffers as a possible route for your next engine and using "jitter samples" to get rid of the resulting aliasing. Will the next engine still have a unified lighting model, and will stencil shadows not be used at all?[/ question ]


( Can self-shadowing be enabled in Doom 3? Does the engine support it? )


Thoughts on the recent finalization of OpenGL 2.0? Are you satisfied? What do you want to see?

 

[Richard]

( Can self-shadowing be enabled in Doom 3? Does the engine support it? )

Yes, there was a mod released days after the game came out that enables self-shadowing for everything. It was turned off for many objects because of quality reasons.

Anyway, my questions for you Rev:

a) You talked about renderers being art asset independent, do you see the renderer you're working on now for your future game to be "backwards compatible" with D3 art?

b) During D3's development you stated your next engine after that would use a high level shading language and that IHV would just have to handle it. Have you changed your mind? Which language are you leaning towards?

c) In D3 there are some experimental HDR stuff, do you expect the new renderer to fully support it through fp buffers and/or fragment programs?

 

[digitalwanderer]

Evil pixie strikes...

 

[Scali]

I would like to know why all paths use a CPU solution for skinning and shadowvolume generation, instead of using a vertexshader-based solution for the paths that run on hardware with vertexshaders that are powerful enough?
I ask this because a system like mine (XP1800+/Radeon 9600Pro) is unable to run Doom3 at a decent framerate during combat, while 3dmark03's Battle of Proxycon clearly shows that my Radeon 9600Pro has enough vertexshader power to handle similar scenes with shadows and skinning at interactive framerates (so no answers like 'vertexshaders have to skin multiple times' please, or 'the physics require it', since you don't have to skin all triangles every frame for collisions, and even if you did, you wouldn't have to pump them over the AGP bus every frame, and the GPU wouldn't have to wait on it), let alone modern hardware, which has 4 times as many vertexshader units as a humble 9600Pro.
It seems that the Doom3 engine will be unnecessarily CPU/AGP-limited now, which means that any game based on the Doom3-engine will have to have relatively low-poly objects. Why is this choice made, instead of making use of vertexshaders, which should scale better in performance over the next few years than CPUs?

 

[pc999]

Why he dislike multi-cores CPU, and (if possible) next-gen consoles in geral, it looks like there is no other option by 2/5 years?

Anything for his next engine would be great too...

 

[jvd]

Ask him what he feels about longhorn and ms requiring dx 9 lvl hardware as the standard for video games with it .

 

[DeanoC]

With the benefit of hind-sight, is he happy with the way he implemented the shadow system in Doom3.

 

[Diplo]

Was there any technical reasons why co-op play wasn't included seeing as this was in the original game?

 

[Luminescent]

How is the default turboshadow stencil shadow implementation set by the game (r_useturboshadow) different from standard stencil shadowing? What makes it faster on modern hardware?

 

[AlNom]

is r_jitter related to your research engine with shadow buffers?

(there's another cvar with "jitter" in it.. I forget what it is though)

 

[Sigma]

I would like to know why all paths use a CPU solution for skinning and shadowvolume generation, instead of using a vertexshader-based solution for the paths that run on hardware with vertexshaders that are powerful enough?
I ask this because a system like mine (XP1800+/Radeon 9600Pro) is unable to run Doom3 at a decent framerate during combat, while 3dmark03's Battle of Proxycon clearly shows that my Radeon 9600Pro has enough vertexshader power to handle similar scenes with shadows and skinning at interactive framerates (so no answers like 'vertexshaders have to skin multiple times' please, or 'the physics require it', since you don't have to skin all triangles every frame for collisions, and even if you did, you wouldn't have to pump them over the AGP bus every frame, and the GPU wouldn't have to wait on it), let alone modern hardware, which has 4 times as many vertexshader units as a humble 9600Pro.
It seems that the Doom3 engine will be unnecessarily CPU/AGP-limited now, which means that any game based on the Doom3-engine will have to have relatively low-poly objects. Why is this choice made, instead of making use of vertexshaders, which should scale better in performance over the next few years than CPUs?

Impossible to create silhuette for shadow volumes on GPU and the same goes for skinning when you need to reconstruct tangent orthobasis. It must be done in the CPU. Skinning on the GPU is possible for simpler per vertex attributes...

QUESTIONS:

1 - Commitement to OpenGL? Feelings regarding the GL2.0 update and GLSL.

2 - What kind of solution for shadows? Anything like deep shadow maps?

3 - Fluid interactions that lack in Doom3.

 

[DeanoC]

Impossible to create silhuette for shadow volumes on GPU and the same goes for skinning when you need to reconstruct tangent orthobasis. It must be done in the CPU. Skinning on the GPU is possible for simpler per vertex attributes...

Its not impossible, just very hard. Actually calculating the silhuette is fairly easy but scatter-gathering the results into something you can use efficently is. I was going to do an article for GPGPU2 but I'm not happy with its efficiency yet, so it will wait until another book I guess...
Recreating tangent basises should be trivial (with VS3.0), no more difficult than computing smooth vertex normals on GPU.

Sorry for the discussion Rev, if anyones wants to continue this discussion lets move it to a new topic.