Deferred shading - will it become commonplace?

[Munky]

Originally posted by: xtknight
Why don't you think it will be used in the future? Tough to implement? Maybe a future API such as an update to DX10 (or DX11) or an extension in OpenGL 2.x will make this more straightforward to do by providing the developer with classes or structures that ease the pain? Or perhaps, they will reengineer the GPU pipeline to make it more feasible.

IMO, DS will be used increasingly, but many games are built on top of older engines which use traditional rendering techniques. U3 will most likely use DS, but neither the Source nor the D3 engine support it, and I don't think Crysis will support it either. So the number of games using DS will remain a minority for the next year or two. But new API features and newer HW with support of those features will certainly play a role in the dev's decision on whether or not to use DS in future games.

 

[jim1976]

Originally posted by: Nightmare225

Originally posted by: jim1976

I'll try 158.19 and Vegas and test whether AA it works or not with them..

Let us know, I'm curious as to this too...

From a brief check I think it's obvious that it works with these drivers. To what extent every sample is applied correctly quality and quantity wise needs some time to be tested. But I can say with certainty that it works.
Very good news

 

[CaiNaM]

Originally posted by: xtknight
Need some background info. What exactly does deferred shading do? Has it been implemented well in games lately?

in layman's terms it renders a scene into a buffer and applies the lighting (or shading, as in R6:V and all U3 tech games) and applies shading lighting/later in 1 pass.

the benefits are it calculates in 1 pass and only renders what is actually visible.

edit: found a better explanation:

"The idea of deferred rendering (of all kinds) its too do some operations late compared to conventional renderers. By doing it 'late' i.e. after depth hidden surface removal (i.e. in screen space), you do those deferred calculations only once per pixel instead of (with a conventional renderer) once per z-buffer pass (think worse case where you render back to front a hundred object, you have to do these calculations a hundred times with only the first front most one actually being visible, the other 99 are wasted)." ~SOURCE

 

[CaiNaM]

Originally posted by: apoppin
in the *context* of that question - Using STALKER - it is hard to say if DS will be "commonplace" by using a 7 year-old - totally unique - Engine as the "prime example"

well, U3 engine uses deferred rendering to calculate shadows. epic has made huge money licensing their tech to other developers, and i doubt it change now, as evidenced by the fact 2 games have already used it. that and the fact ut3 will be big, and gears of war will be coming to pc might hint that adoption may come sooner than you speculate.

GRAW uses deferred lighting, and i believe crytek will use a deferred rendering method as well. the question isn't really when it adoption will come; it's already here.

 

[Nightmare225]

Originally posted by: jim1976

Originally posted by: Nightmare225

Originally posted by: jim1976

I'll try 158.19 and Vegas and test whether AA it works or not with them..

Let us know, I'm curious as to this too...

From a brief check I think it's obvious that it works with these drivers. To what extent every sample is applied correctly quality and quantity wise needs some time to be tested. But I can say with certainty that it works.
Very good news

Well, performance is still horrible. I can't play at 1680x1050...

 

[apoppin]

others do - with similar specs

perhaps the drivers

Originally posted by: CaiNaM

Originally posted by: apoppin
in the *context* of that question - Using STALKER - it is hard to say if DS will be "commonplace" by using a 7 year-old - totally unique - Engine as the "prime example"

well, U3 engine uses deferred rendering to calculate shadows. epic has made huge money licensing their tech to other developers, and i doubt it change now, as evidenced by the fact 2 games have already used it. that and the fact ut3 will be big, and gears of war will be coming to pc might hint that adoption may come sooner than you speculate.

GRAW uses deferred lighting, and i believe crytek will use a deferred rendering method as well. the question isn't really when it adoption will come; it's already here.

again... all of that appears to be true ...
-- but it has absolutely nothing to do with my reference to STALKER as perhaps NOT being *the best example* for projecting if DS will be commonplace or not

. . . and *obviously* DS has been around a long time [as evidenced by its use in STALKER] ... --and it isn't commonplace - yet

perhaps that lack of commonness is an issue with DX9's "limitations" ... but again, that is *way beyond* the scope of my simple interjection that "stalker engine is 7 years old"

 

[PhatoseAlpha]

Eh, it started development a long time ago, though from the look of it they've been updating it constantly as time went on.

At any rate, it looks like I got some answer to the question. I suppose I'll just have to keep an eye on the benchmarks if and when this hits, see if the version of AA used here has the same dynamic as the older ones. Thanks guys.

 

[Drayvn]

Originally posted by: munky

Originally posted by: xtknight
Why don't you think it will be used in the future? Tough to implement? Maybe a future API such as an update to DX10 (or DX11) or an extension in OpenGL 2.x will make this more straightforward to do by providing the developer with classes or structures that ease the pain? Or perhaps, they will reengineer the GPU pipeline to make it more feasible.

IMO, DS will be used increasingly, but many games are built on top of older engines which use traditional rendering techniques. U3 will most likely use DS, but neither the Source nor the D3 engine support it, and I don't think Crysis will support it either. So the number of games using DS will remain a minority for the next year or two. But new API features and newer HW with support of those features will certainly play a role in the dev's decision on whether or not to use DS in future games.

UE3 has around 40 to 50 games already being developed with it now. I think its definitly the majority of games that are using it.

 

[apoppin]

but do we know that UE3 will *use* deferred shading in it's games ?
-- it is just one of many features Game Devs can choose from


there are evidently some disadvantages and advantages to using DS ...
it is gonna take me some time to plow all thru it ... and i am bookmarking most of it ... for now
[back to Shivering Isles ... i missed a *locked* place no walkthru mentions]
:Q

anyway ... start with the links from here:

http://en.wikipedia.org/wiki/Deferred_shading

and the dev info from nvidia

http://developer.nvidia.com/object/6800_leagues_deferred_shading.html

 

[CaiNaM]

Originally posted by: apoppinagain... all of that appears to be true ...
-- but it has absolutely nothing to do with my reference to STALKER as perhaps NOT being *the best example* for projecting if DS will be commonplace or not

. . . and *obviously* DS has been around a long time [as evidenced by its use in STALKER] ... --and it isn't commonplace - yet

well, my comments had nothing to do with STALKER, rather the use of deferred rendering becoming commonplace. in the last 6 mo. there have been more major title released using it than not, and major upcoming titles such as CRYSIS and any UE3 titles will use it, so it certainly is not uncommon.

as deferred rendering in its various implementations offer performance advantages (not to mention coding advantages) that are too great to ignore. g80 architecture is designed to run this very efficiently, giving devs more reason to use it.

it's already here, and is already becoming commonplace. i don't see how you can say it isn't.

perhaps that lack of commonness is an issue with DX9's "limitations" ... but again, that is *way beyond* the scope of my simple interjection that "stalker engine is 7 years old"

it's not really a limitation of DX9, rather a limitation of MSAA. there are already "workarounds", but the expense of AA when using deferred methods is simply too expensive for the current hardware most ppl have.

Originally posted by: apoppin
but do we know that UE3 will *use* deferred shading in it's games ?

yes. that's the way it is designed, as mentioned in by Tim Sweeney in various interviews.

 

[apoppin]

so no DS + AA?

--until DX10

did he mention workarounds?
.. i guess i haven't been keeping up with Unreal

EDIT: can you give some links to Sweeny on DS in U3 ?
... i can't find anything

and finding info here is ... i gave up
http://www.unrealtechnology.com/html/technology/ue30.shtml

 

[CaiNaM]

Originally posted by: apoppin
so no DS + AA?

--until DX10

did he mention workarounds?
.. i guess i haven't been keeping up with Unreal

well, R6:V is DX9 & UE3, and it appears the 159 drivers allow use of AA - at a very high cost in performance.

supposedly DX10 will allow for other methods not as expensive, but that remains to be seen.

at any rate, i didn't say he said anything about AA; you asked if we were sure UE3 used deferred shading (actually that's not really an accurate term; deferred rendering would be better as it applies to multiple techniques); i just replied TS had stated UE3 uses deferred rendering for shadows (not sure for what, if anything else.

websearch i ran across this @ motherboards.org:

"The High Level Shading Language of DirectX 10 has a few new features as well support for constant buffers, view constructs, Integer and Bitwise instructions, and the switch statement. DirectX 10 also adds support for Transparency anti-aliasing, shadow map filtering and using MSAA (Multi-Sample Anti-aliasing) in deferred graphics engines. The Geometry Shader is a shader that allows the graphics card to create geometry effects. Effects like stencil shadows, dynamic cube maps, and displacement mapping which in previous hardware relied on the CPU or on multi-pass rendering can be done faster with the Geometry Shader. Two new features, vertex amplification and vertex minimization are introduced. Vertex amplification allows the Geometry Shader to output 1024 vertices for every vertex fed into it. Vertex minimization outputs a lower number of vertices than input into it."

again, not sure how well it does it; just that it does.

EDIT: can you give some links to Sweeny on DS in U3 ?
... i can't find anything

and finding info here is ... i gave up
http://www.unrealtechnology.com/html/technology/ue30.shtml

heh.. i have no clue where i read it... might have been on nvnews.. if i get some time later i'll see if i can find it....

EDIT here's one

 

[apoppin]

thank-you

his last email answer was interesting

Will UE3.0 support predicated tiling to make use of 4xAA on Xbox 360?

Sweeney- Gears of War runs natively at 1280x720p without multisampling. MSAA performance doesn't scale well to next-generation deferred rendering techniques, which UE3 uses extensively for shadowing, particle systems, and fog.

 

[1Dark1Sharigan1]

Originally posted by: BFG10K
I would like to see a screenshot comparison with and without driver AA. Just because he's seeing a performance hit it doesn't mean he's getting driver AA.

This is compounded by the fact that the game is so damn blurry which masks aliasing.

0xAA
4xAA

AA seems to work with the 158.19 drivers, but cut my framerate in half in some places.

 

[BFG10K]

Deferred shading is a method of accelerating complex lighting.

On paper perhaps. In practice the games using it thusfar tend to run very poorly and don't even look better than games using traditional methods which run far faster.

Deferred shading games just tend to give us overdone bloom and blurry mush.

Couldn't they just do SSAA, send that to this "fixed buffer" and then do deferred shading on the bigger image or would that not work/be too slow?

AFAIK that's already what such games do. They implement a basic form of SSAA done on the shader.

The problem with this is (apart from the performance hit of course) is your GPU's AA modes are useless and you're at the mercy of the developer's implementation.

0xAA
4xAA

Those screenshots aren't side by side by any stretch of the imagination. Post of up something in identical positions and then I'll be convinced.

It's good news if the driver can force AA in such games but I've seen so many bogus claims about forcing AA in that game I need better proof.

 

[CaiNaM]

Originally posted by: BFG10K
On paper perhaps. In practice the games using it thusfar tend to run very poorly and don't even look better than games using traditional methods which run far faster.

how is that? you're comparing apples to oranges; there's no "non-deferred rendering" example using the same complex lighting. who's to say that if the same quality of lighting was used via a "non-deferred rendering" method the performance wouldn't tank, or that it's even possible?

can you show me an example of a game which doesn't use some type of deferred rendering, but renders multiple light sources in realtime where the sun, campfires, torches, other npc's using flashlights, etc. will all interact with the environment and cast multiple, complex shadows which constantly change as the light sources move, or as you move relative to the light sources?

if you have STALKER, run an underground level and compare static lighting to dynamic lighting. the difference is amazing. here's an example.

there's obviously a reason why dev's are going in that direction, and i doubt tim sweeney for instance implements it in UE3 because it performs poorly and looks like crap...

Deferred shading games just tend to give us overdone bloom and blurry mush.

obviously your opinion, but that's hardly fact by any stretch of the imagination.

The problem with this is (apart from the performance hit of course) is your GPU's AA modes are useless and you're at the mercy of the developer's implementation.

certainly a valid observation, but we've been saying that all along - it's too "expensive"... at least with current hardware.

Those screenshots aren't side by side by any stretch of the imagination. Post of up something in identical positions and then I'll be convinced.

so open 2 browser windows, one with each example and compare them side to side. it certainly would take a lot less effort than what you are asking him to do...

It's good news if the driver can force AA in such games but I've seen so many bogus claims about forcing AA in that game I need better proof.

it's there if you choose to see.

 

[CP5670]

Does anyone know if Deus Ex Invisible War used this method? That was another blurry looking game that didn't seem to work with AA. (it actually has its own multisampling slider in the options, but it never had any effect for me)

 

[BFG10K]

you're comparing apples to oranges; there's no "non-deferred rendering" example using the same complex lighting.

But there are games that use the feature and games that don't and so far games that use said feature have demonstrated no advantages (or even inferiority) over games that don't.

who's to say that if the same quality of lighting was used via a "non-deferred rendering" method the performance wouldn't tank, or that it's even possible?

I've seen better lighting on non-deferred implementations (e.g. Doom 3, Fear) that run faster and also have working AA too.

can you show me an example of a game which doesn't use some type of deferred rendering, but renders multiple light sources in realtime where the sun, campfires, torches, other npc's using flashlights, etc. will all interact with the environment and cast multiple, complex shadows which constantly change as the light sources move, or as you move relative to the light sources?

A game that uses a unified lighting system like Doom 3 should be able to manage that.

Edited response above due to misreading the quote.

if you have STALKER, run an underground level and compare static lighting to dynamic lighting. the difference is amazing. here's an example.

You want me to to compare static lighting to dynamic lighting? Why? The game's implementation of static lighting is inferior on purpose to increase performance. Static lighting is not supposed to offer the quality of the dynamic path.

there's obviously a reason why dev's are going in that direction,

If there is I can't see it. All I see is poor performance and broken AA.

obviously your opinion, but that's hardly fact by any stretch of the imagination.

Are you claiming R6 isn't blurry? If you are you must be blind.

but we've been saying that all along - it's too "expensive"... at least with current hardware.

Except I'm not talking about a workaround for GPU AA, I?m, talking about the game implementing its own hack job AA. We have hardware MSAA for a reason and it isn't for game engines to try to implement SSAA.

That's backwards and it's also reinventing the wheel.

so open 2 browser windows, one with each example and compare them side to side. it certainly would take a lot less effort than what you are asking him to do...

The screenshots are in two different places and the areas with aliasing in one aren't even available in the other. What exactly do you think you can see there?

it's there if you choose to see.

So is placebo and delusion.

 

[BFG10K]

Does anyone know if Deus Ex Invisible War used this method? That was another blurry looking game that didn't seem to work with AA. (it actually has its own multisampling slider in the options, but it never had any effect for me)

I think it does but you can disable bloom and it'll revert to a standard rendering path and you can then force driver AA.

It also looks much better with bloom off (something many games have in common).

 

[Matt2]

Originally posted by: 1Dark1Sharigan1

Originally posted by: BFG10K
I would like to see a screenshot comparison with and without driver AA. Just because he's seeing a performance hit it doesn't mean he's getting driver AA.

This is compounded by the fact that the game is so damn blurry which masks aliasing.

0xAA
4xAA

AA seems to work with the 158.19 drivers, but cut my framerate in half in some places.

I see it.

 

[BFG10K]

It looks like there might be a difference on some parts of the gun but that could just be due to the different positions.

The places with the visible aliasing (such as the table edges and the plastic bin) are not in a position to show aliasing in the second shot so it's not conclusive.

 

[1Dark1Sharigan1]

Originally posted by: BFG10K
It looks like there might be a difference on some parts of the gun but that could just be due to the different positions.

The places with the visible aliasing (such as the table edges and the plastic bin) are not in a position to show aliasing in the second shot so it's not conclusive.

Well the difference is pretty obvious (abet subtle) in-game. I'll post up some pics of the same position later.

 

[jim1976]

Originally posted by: 1Dark1Sharigan1

Originally posted by: BFG10K
It looks like there might be a difference on some parts of the gun but that could just be due to the different positions.

The places with the visible aliasing (such as the table edges and the plastic bin) are not in a position to show aliasing in the second shot so it's not conclusive.

Well the difference is pretty obvious (abet subtle) in-game. I'll post up some pics of the same position later.

Yes it is.. AA is there.. I'm so absorbed by C&C 3 TW that I didn't have the chance to post some pics.. I'll try post them tomorrow

 

[schneiderguy]

Originally posted by: BFG10K
It looks like there might be a difference on some parts of the gun but that could just be due to the different positions.

The places with the visible aliasing (such as the table edges and the plastic bin) are not in a position to show aliasing in the second shot so it's not conclusive.

the scope on the gun is smoother in the second picture than it is in the first, and that's in the same position in both shots

 

[BFG10K]

Well the difference is pretty obvious (abet subtle) in-game.

Sure but I don't have the benefit of being in-game because I don't have said title, that's why I need legit screenshots.

I'm not trying to be nasty - like I said there've been a lot of bogus claims about driver AA in that particular title so I need concrete proof before I believe it.

the scope on the gun is smoother in the second picture than it is in the first, and that's in the same position in both shots

If the game has gun bob like most FPSes then we don't know whether it's in the same position or not. The only way we would know is if the main scene was in exactly the same place in both screenshots.