Yet another shadows question

I was told on this forum by reliable sources such as EsH that the doom 3 rederer only renders what light lands on. Therefore darkness make things run faster, because the engine renders less.

However, in the illegal alpha apparently (i dont have the alpha so i dont know) its possible to *optimize* by turning shadows OFF!

Why doesn that make the engine run SLOWER?

Couldn't help ya there. It's really too unfortunate that the Doom 3 engine won't be even slightly illuminating the really dark areas. It would just make the world look so real. However, hardware rendering limits are what holds that back from happening. Next Carmack engine perhaps.

Wobbo said:

I was told on this forum by reliable sources such as EsH that the doom 3 rederer only renders what light lands on. Therefore darkness make things run faster, because the engine renders less.

I think that was a theory rather than a statement of fact.

Hi.

While Doom3's lighting engine only displays what is lit, a lot of the light shading stuff is still done for the unlit fragments-- it's just not written to the frame buffer. Some cards (I believe) do have an early stencil test, that will prevent the fragment from even being calculated. However, you have to remember that drawing the shadows takes up a huge amount of fill rate in itself. For example, if you have four large zombies standing off screen to the right, and a light shines overthem, across the screen, onto some wall on the left, you are updating the screen somthing like 3xnum_zombies, even though niether zombies or shadowed wall is even seen! This is because the shadows are sort of like a separate model, which has to be rendered. In addition, calculating the shadow volumes takes a huge amount of CPU power. I don't believe I ever said that objects in shadow will be rendered (significantly) more quickly, but if I did, I appologize-- it's just not true, for the most part.

About the totally unlit areas-- Carmack has said that there is an ambient sort of light that can brighten up shadowed areas. For what it's worth, there is a paper floating around out there detailing how soft shadows can be done using "wedges". It looks like a robust way of going about things-- no pixelation, casts on arbitrary surfaces, etc. Requires some of the more advanced 3D hardware, though, so a bit out of Doom3's tech target.

Well in the alpha you do have the option of using a gunlight to illuminate the darkness. This is key, because of the way the shadows look really and spooky, especially when you go into a dark corner and zombies start chomping your ass when you turn the light on

oh i get it, i didnt understand stencil shadowing, thanx esH

IIRC shadow volume's take place after the scene has been lit up. Then the shadows are applied to the scene, contrary to other shadowing methods. So I beleive Doom's engine does render everything whether or not its "covered by shadows", just because the shadows are the last thing to be rendered

The engine renders the whole scene unlit and then goes through it light by light.
Depending on which gfx board you have it'll take more or less passes.

It's basically true that pixels which lay outside of any light volume or inside a shadow volume which does not ovelap with another lightvoluem do not get shaded any further. They're just pitch black hence less pixels so shade per light.

That's some sort of optimization, but still the shadowing takes up 50% or so of the rendering time, so it's not _the_ factor to get good performance.

To get the best perfomance you should disable the shadows and the bumpmapping, but in this case I'd suggest to simply install Quake3 instead.

Radea` said:

IIRC shadow volume's take place after the scene has been lit up. Then the shadows are applied to the scene, contrary to other shadowing methods. So I beleive Doom's engine does render everything whether or not its "covered by shadows", just because the shadows are the last thing to be rendered

With the alpha? If so, I suspect that's merely something left unoptimized.

I post this here, beacuse I don't where to post it. And don't want to start a new topic.

On the messageboards of polycount I saw this(scroll down) WIP model to be used in Tenebrae2. Uses the same tech as Doom3. Generated High poly model Normal map to low poly model stuff.

Seems cool.

AndrewB said:

With the alpha? If so, I suspect that's merely something left unoptimized.

No. Thats the rendering technique for shadow volumes in general. If its not done that way, its not known as shadow volumes.

Take a look at the trites screenshot, the shadows off the trites' legs are overlapping, they are not completely dark, so that ambience lighting must be correct.