Deferred Rendering: Please No!

I liked everything about the latest livestream very much, except when you showed off deferred rendering. Deferred rendering is very bad and has unfortunately become a fad in engine work. You lose antialiasing you lose translucency you lose render to texture like mirrors and whatnot. You lose so much and gain so little. I haven't seen a deferred game that looks good yet. It's a seal that guarantees a game will look like **** and run like ****.

An aliased mess full of messy shaders, poor light falloff, lack of shadows and technical limitations on the artstyle is the order of the day in the land of deferred. It looks good in super sampled bullshots but nowhere else. In practice it becomes blurry and confusing because of the screen space AA that has to be employed.

There are other ways of doing whatever you need to do. I'm not a graphics programming genius and I'm sure you know what you need far better than I do.
I know it's very fancy to be able to have hundreds and hundreds of lights in the scene, but the cost in image quality and performance isn't worth it. It's never worth it.
Especially in an RTS where clarity and subpixel detail is paramount.

Considering the Aesthetic and scale of PA, I think it's not nearly as big of a deal as you make it out to be.

Mike

So let me get this straight, you've no clue what you're talking about, but you think you can tell experts what they shouldn't use?

Not being able to have hundreds and hundreds of lights in a scene, when we're talking about up to a million units in game?

I think it would be pretty necessary to have those hundreds of lights.

Oh but again, if you bothered to read I think it's super important exactly because of those things. Deferred precludes the clarity and detail that you'd be looking for in order to be able to clearly pick out small units from a distance in an RTS.

I know about deferred rendering. And I know there are many many different ways to approach lighting in a 3D engine. I can't say which approach is right for PA specifically but I can say deferred is never right for anything.

I'll try to address your concerns on deferred rendering. I'm no expert and not a dev but have done some graphics programming.

Deferred rendering, or some derivative thereof, is used in most AAA games made in the past year or so. Many of these have excellent examples of anti-aliasing, great shaders, lots of high quality shadows, and stunning art. Crysis 2 and Starcraft 2 are a couple of examples.

It is true that traditional methods of anti-aliasing don't work with deferred rendering. This has been remedied by new algorithms such as MLAA and FXAA. These algorithms can be more expensive but they also look nicer than traditional methods.

Shaders don't change a whole lot with deferred rendering except to require more memory to add additional shaders, which isn't *much* of a problem on a computer.

Mirrors are handled slightly differently with deferred shading but I haven't seen many problems implementing them. At least, not many problems that aren't already present with forward rendering.

Transparency becomes a bit of a problem with deferred shading but, as noted on the live stream, the devs don't need a whole lot of transparency in this game and if they do need more there are several great algorithms to solve this, such as inferred shading.

Shadows are something that actually improves with deferred rendering because you can have a lot more of them. The problem here lies with memory.

I get the feeling that you are used to games on consoles where memory is extremely limited. On a computer the blocky shadows and limited shaders you see on consoles rapidly disappear.

As for art, that entirely depends on the artist. From what I've seen PA has some great artists.

Again, I'm not a dev and don't know the details of the games architecture, but I hope I have addressed you concerns somewhat.

Tyler Seacrest

FXAA performs better at the cost of overall detail/sharpness and it isn't as effective as MSAA + SGSSAA. However the latter configuration cuts the framerate in half, but that can at least be solved by throwing more power at it.

Here's an example in Deus Ex HR.

FXAA http://cloud-2.steampowered.com/ugc/903 ... 656096ABA/

4x MSAA + 4x SGSSAA http://cloud-2.steampowered.com/ugc/903 ... 3C057FDE8/

Now you might make a case that PA doesn't have much transparency, but here's an example of beam weapons in Sword of the Stars 2, which has a deferred rendering system and only FXAA: http://cloud-2.steampowered.com/ugc/920 ... 73FBEE30E/ Some of the beams are so aliased they almost disappear. You can add a SMAA injector but even that doesn't help much with aliased thin lines at a medium to far distance.

Of course performance versus image quality is a tough trade-off, you probably don't want to repeat the "You will need a NASA computer to run Supreme Commander" mantra.

Yes SGSSAA is impressive especially on the geometric AA

I forgot to mention there's one brute-force alternative if everything else fails in respect to AA: downsampling from x times the native resolution. Unfortunately this hack needs a lot of VRAM and it's a bit complicated to set up.

Planned if memory serves

We have that already

will be of use in this... how?

I think tylerseacrest just schooled bmb.

This is patently untrue. FXAA is not AA, it's edge smoothing. What you forget to mention is that how Crysis 2 actually looks because of the "AA" is that it is blurry, still heavily aliased, just the jaggies are blurry now, and has tons of unsightly ghosting. It's a mess to look at and performs terribly. This is generally true for all deferred games if they even bother with "AA" at all.

There are a few techniques for implementing reflections in a deferred renderer but they are not very good.

Brute forcing AA in a deferred renderer is extremely expensive to the point of uselessness.

Even for drawing lots of lights, deferred is not actually very good because while you can do it in one pass you still get an insane amount of overdraw.

All of the RTT/reflection/AA/resolution scaling issues come from one thing really, which is the crazy amount of data you have to store to do something as simple as light the scene.

There is no benefit to deferred, only downsides.

Funny fact many people are raging over blur from AA, but then play with maxed Motion blur or DOF.

Have you ever taken a photograph with a DSLR?

this game is blocky if you watch the pitch what unit is sphere and if you see the water was transucent

Dunno about others but I can't see any difference between medium or high settings with and without AA and vsync and other stuff. It all just looks the same to me so I don't understand whats the big deal about this

Perhaps I need some glasses :ugeek:

BMB, y u so abrasive?

And when PA comes out with no antialiasing, magically reverted transparency, and a pressing need for nonexistent render to texture you can talk. Till then, you have no real base to complain.

But then it's too late to change the rendering system. You need to plan ahead by weighing the pros and cons. And there is proof that there are both pros and cons to deferred rendering.

I mean, that's great and I see that, but running around claiming you CANNOT get anti-aliasing, transparency (the frikkin livestream had loads of transparency rendering), and render to texture (...without even explaining why we would need this) on an engine already displaying a bunch of new tech just seems pointless IMO.

I'm the only one that still does not use AA or vsync in the games?

I think if the choose deferred rendering if because it has more upsides than downsides.