What can we reasonably expect from next gen graphics?

[gongo]

Hear me out guys, i been playing with the Unigine Benchmark on my 4870 1GB. Let say im not particularly impressed. At 720p with 8XAA and 16AF, performance is barely keeping the minimal 30fp. The thing then is the graphics itself does not look very pretty compared to what we have on consoles, sure the textures are uniformly super higher rez, the lighting is smooth, HDR is higher definition, there are dynamic shadows casted on every objects and the geometry is very high. Still i can see shadow flickering, grass pops up, objects just cutting through the distance blur, the worst offender is how the same old tired lighting that makes everything looks "dull and cartoony". This is after we have gone through 2 DX generations and 4-5 stages of GPUs hardware upgrades.

I know those with powerful pc do put up the act of a superior being, im guilty too but TBH the visual difference between whats on Pc and consoles are not the kind of leap. I would say the same about Crysis (it still is a DX9 based engine). The Crysis guys were right to say the next leap will come in 2012 from next gen console development.

So i ask the 3D experts, what can we expect from next gen 3D graphics? Will the HDR lighting finally look near CG quality and not some cartoony washout? Will we have models that are rounded when they are supposed to be rounded?

If we take from the "Predict the next gen hardware thread", the power that should go into PS4/720, is it enough to bring a leap in 3D fidelity that makes our jaws drop? I remembered the very first Project Offset video, and i thought to myself hey they got the lighting and object blur right...for this gen! This gen is almost over and the engine still remains to be seen, and if those later videos released are real realtime, it looks downgraded.....previs BS as usual.

First trailer...skip to the 50s mark
http://www.youtube.com/watch?v=Q9e_SuHeCc0

Gameplay video...looks like Oblivion ...with the same tired cartoony washout
http://www.youtube.com/watch?v=FpdPWVfaBQs

 

[Lidve]

I personaly think that we are allready at least 2 generations ahead of consoles, in every department
Unigine benchmark is there to showcase tessellation and some other features, and its really impressive

 

[Lidve]

Problem is that people mistake "realism" and "graphics"

There are games that use real life footage and you just click button to go to "next level"
would you say they have best graphics out there? I mean they are as real as you can get


Lets say you have 1 model that is as realistic as possible, and fixed background image of real mountain

It might look attractive, but its nowhere near 1 cartoon model in 3D world with volumetric clouds and correct shadows, physic interaction,etc etc

First one might look more real etc. but second one have way better graphics

 

[Neb]

.....So i ask the 3D experts, what can we expect from next gen 3D graphics? Will the HDR lighting finally look near CG quality and not some cartoony washout? Will we have models that are rounded when they are supposed to be rounded?

You are confusing HDR quality/type vs artwork choices. You can have the best lighting system in the world but if you target 'cartoony' artstyle then it will look cartoony. About your model smoothness comment I find it a bit strange considering Uniengine has DX11 support with tesselation. One of the engines highlights.

If you check games around they go from cartoony to vivid to realistic looking. One example might be ArmA 2 and it even does 128bit HDR. Might be rough on some edges but visuals are scarily realistic and very detailed. All in all a mather of good artwork, good design and ofcourse backed up with strong tech. Why Crysis looks awesome but say the Mechwarrior mod for it on very high looks like Crysis on low-medium at quick glance.

I remembered the very first Project Offset video, and i thought to myself hey they got the lighting and object blur right...for this gen!

And back then you had target versions for this gens games and those games ended up looking far inferior making that PO demo still look vastly better and technically way ahead with stuff you'll enjoy next gen if lucky. There is a technicall presentation about engine. Everything was realtime IIRC.

 

[Nesh]

Problem is that people mistake "realism" and "graphics"

There are games that use real life footage and you just click button to go to "next level"
would you say they have best graphics out there? I mean they are as real as you can get


Lets say you have 1 model that is as realistic as possible, and fixed background image of real mountain

It might look attractive, but its nowhere near 1 cartoon model in 3D world with volumetric clouds and correct shadows, physic interaction,etc etc

First one might look more real etc. but second one have way better graphics

What about 3D realistic model in a very realistic 3D background and all other related graphical features?

One cartoon model on a fixed cartoon background wouldnt impress either

 

[Lidve]

What about 3D realistic model in a very realistic 3D background and all other related graphical features?

One cartoon model on a fixed cartoon background wouldnt impress either

that would be amazing

 

[Neb]

What about 3D realistic model in a very realistic 3D background and all other related graphical features?

But then that needs to be acheived without the heavy use of baked stuff and 2D assets as in photomapped textures as that stands out like a sore thumb and breaks photorealism. Funny is that I've read lots of people saying Crysis visuals apart from tech wasn't anything special to them becouse it looked to real. Sureallity was more impressive to them (apart from car games I assume)!

One cartoon model on a fixed cartoon background wouldnt impress either

It could impress as it could not. Just becouse something isn't photo realsitic looking doesn't count it out from looking impressive and even more impressive than photorealistic graphics. Reminds me of how Max Payne 2 at time looked utterly impressive and way more photoreal than FC yet sum for sum FC was leagues ahead in visual IQ while looking more cartoonish.

 

[sunscar]

About your model smoothness comment I find it a bit strange considering Uniengine has DX11 support with tesselation. One of the engines highlights.

Hey Neb - check the top line of his post and you'll see why. He can't enable tessellation/displacement on his card, only a 4000 series. So he litterally is running a demo equivalent to a current gen game at higher resolution, and barely anything more. It's easy to see his confusion that way.

I'll try to displell a few things though.

Shadow flicker aside, and lighting aside, tessellation + displacement mapping should finally give us smoothed, rounded models at rediculous mesh densities. Couple that with virtual texturing shemes and now we're looking at high polygon objects with rediculous texture resolutions, too. SO we could feasibly emulate *some* offline CGI scenes fairly well. We're still likely another console generation or more till we can expect REYES-style rasterizers that display billions of micropolys per frame though, but by that time we may be looking at Crysis-size worlds with offline-level poly counts and texture resolution, and it still may be another generation or two beyond that till we can simulate the whole thing at a per-poly level.

 

[Lidve]

Hey Neb - check the top line of his post and you'll see why. He can't enable tessellation/displacement on his card, only a 4000 series. So he litterally is running a demo equivalent to a current gen game at higher resolution, and barely anything more. It's easy to see his confusion that way.

that makes sense

 

[Nesh]

But then that needs to be acheived without the heavy use of baked stuff and 2D assets as in photomapped textures as that stands out like a sore thumb and breaks photorealism. Funny is that I've read lots of people saying Crysis visuals apart from tech wasn't anything special to them becouse it looked to real. Sureallity was more impressive to them (apart from car games I assume)!

There are people that liked and disliked Crysis for looking the way it looked. But there are also people that dislike "colorful" visuals as well because they dont look "realistic". Uncharted 2 impressed me like no other console game, but there are people that thought it was too colorfoul to meet their tastes. There are also people that like both as long as they look good.

It could impress as it could not. Just becouse something isn't photo realsitic looking doesn't count it out from looking impressive and even more impressive than photorealistic graphics. Reminds me of how Max Payne 2 at time looked utterly impressive and way more photoreal than FC yet sum for sum FC was leagues ahead in visual IQ while looking more cartoonish.

You didnt get me on that one.

Thats not what I said. To make his argument he put an example of realistic visuals in a context where they are not impressive and compared it with an example "cartoon-like" visuals in a context in which they can impress. I pointed to that

Both realistic and unrealistic visuals can be impressive and we need both.

 

[Neb]

There are people that liked and disliked Crysis for looking the way it looked. But there are also people that dislike "colorful" visuals as well because they dont look "realistic". Uncharted 2 impressed me like no other console game, but there are people that thought it was too colorfoul to meet their tastes. There are also people that like both as long as they look good.

True but that was one point it got under heavy critique. Also from what I understood colors in games have been the hotness this gen with all the 'KZ2, Gears etc' color talk or rather 'lack of color' talk.

Thats not what I said. To make his argument he put an example of realistic visuals in a context where they are not impressive and compared it with an example "cartoon-like" visuals in a context in which they can impress. I pointed to that

Both realistic and unrealistic visuals can be impressive and we need both.

I understood it as the complexity of the game(s) also factor in and not just how photoreal it looks hence his comment. amazing graphic doesn't just consist of a nice model on a static scene with strapped on photos on simple geometry and baking galore. A 'cartoony' game could look far better without looking as photoreal simply becouse it has tech, effects and graphics that are advanced and gives 'graphical depth' and exploration areas etc.

 

[Laa-Yosh]

Offline CG quality is a moving target... Just compare Toy Story, Final Fantasy TSW and Avatar.

I'd say we've reached or surpassed the following things in Toy Story:
- Lighting is more complex with all the GI and similar stuff, although it's all precalculated (but who cares for the most part).
Toy Story and actually most Pixar movies up until Ratatouille have relied on manually placing dozens or hundreds of point light sources to simulate area lights, bounced light and such; now they also have raytracing and various GI techs available in Renderman
- Some shading is better when using various tricks, fakes, SH, and higher quality source art (we have relatively good methods for water, fire and smoke, hair, skin and such)
Of course Renderman shaders of today are untouchable by anything in realtime rendering.

There are problems as well:
- Geometry complexity is still way below as PRMan has been able to do micropoly rendering back in '95 already, and, well, offline means you're only somewhat limited by memory and other resources, but not by vertex processing speed
- Shadows have more artifacting because you can't manually finetune bias and other settings. Also have to note that Mental Ray is quite good at fully raytraced area shadows (one of the reasons we've switched to it - huge time saver with lighting).
- Antialiasing and texture filtering and motion blur and DOF are all untouched quality wise, because they are primarily dependent on performance, thus they can only get so far in a realtime application.

FF:TSW is actually quite close to the current visual quality in many ways, but still has vastly superior geometry complexity, a lot of dynamics animation (cloth, muscles, smoke etc) and is free of most rendering artifacts. They couldn't use GI or SSS or other more recent advancements in offline CG tech though; in fact the first few shots of the movie have been rendered with Maya's age-old software renderer instead of Renderman.

Nevertheless, IMHO it is safe to say that under the right circumstances, a current game engine can produce bullshots that reach FF's general quality in the eyes of the general audience. AA/textures/post effects etc. are nevertheless still far too rough in any realtime application to even come close to acceptable levels.

Avatar is untouchable in every way possible and will remain to be so for the foreseeable future. One good reason is that without all the trickery possible despite the 3D, it would literally take billions of dollars just to produce the art assets for a game of such visual quality. And then noone has tried to animate anything yet...
(One interesting thing to note is that Avatar's lighting is based on spherical harmonics, although taken to an entirely new level because of the incredible scene complexity that Weta has built into all the shots)


So the general rule still remains: by it's nature offline rendering is always going to be ahead of realtime rendering.
And the dynamic qualities are very hard to reproduce as well, as all the complex hair and muscle and cloth dynamics for characters, and other stuff for VFX (explosions and smoke, or leaves blowing in the wind) can be run offline for as long as it takes to get perfect results, with last minute manual fixes for the tiny errors that even the most advanced simulations still produce.

 

[Lidve]

I know this might not make much sense but here it is - for me,its more about "immersion in world" than actual graphics

I dont know how to say it, but not so good looking water that you can actualy interact with will always be better for me than some photorealistic water that you cant do anything with

So this tessellation thing may not look as good as some other game, but actual bumps instead of normal maps is big leap from my perspective and i really admire it

Its just matter of time someone talented make good use of this new technology

 

[Nesh]

True but that was one point it got under heavy critique. Also from what I understood colors in games have been the hotness this gen with all the 'KZ2, Gears etc' color talk or rather 'lack of color' talk.

Heh both of these games have been praised and criticized simultaneously actually. It ranged from either "check out that lighting effects and detail" to "thats too much brown and gray".

What we need is good art direction. And good or bad art direction can be applied on both "photorealism" and "surrealism". I think that was the main problem with Crysis in the first sections

I understood it as the complexity of the game(s) also factor in and not just how photoreal it looks hence his comment.

Well I would agree then.

 

[Lidve]

Other example would be Crysis

sure, there might be game that have better models used for trees with better use of maps etc.

but in Crysis you can actualy break tree, move it, shoot it etc. among many other things

So that interaction with virtual objects is what makes graphics stand out at last for me, opposed to some super real image of tree that is just that - image

 

[Laa-Yosh]

As for what to expect, I think the best approach is to review the hardware options.

It's highly reasonable to go with a top GPU from ATI/Nvidia, so plenty of computing power for complex shaders and geometry and such.
Image quality (resolution, AA and AF) and frame rate are however highly dependent on available bandwidth. Don't expect too much on this front, there's only so much wiring you can put on a motherboard with a budget. EDRAM could mix things up, from the developer comments here on B3D I'd still expect at least MS to use it (but possibly Sony as well, in order not to screw with multiplatform development).

As for CPU, it may still make sense to go with a custom solution in which case IBM should be preferred just to break away from Intel's monopoly. 6 to 8 cores are quite likely, but perhaps with functionality limited similar to the XCPU and PS3's PPE. Sony may drop Cell, too, although it makes sense to build upon years of software engineering from their developers...

Memory is cheap enough that 2 GB should be a given. Lots more data, lots more detail, lots more production costs though. Wonder what Carmack would do with so much RAM

Overall I think the visual quality of the DX11 benchmarks should be achievable at 1080p, and with developers mastering the hardware, probably even more. The question is how the hell they're going to find the money to produce all the content...

 

[marcus_rocks]

If the next-gen games can achieve the AA level similar to that of Warhawk then I'm happy.

 

[Laa-Yosh]

That's actually a good point. Adding interactivity is probably cheaper then increasing the resolution and detail (and thus workload) of the assets. The new hardware will most likely have the resources to deal with it, too.

 

[Nesh]

That's actually a good point. Adding interactivity is probably cheaper then increasing the resolution and detail (and thus workload) of the assets. The new hardware will most likely have the resources to deal with it, too.

I would love that as the most impressive thing for me about the so called "tech demos" that Sony presented at E3 2005 wasnt simply the level of realism, but the level of animation, interaction and physics with the environments and characters.

We need a natural flow of interaction.

Thats what I wanted to see this gen from next gen consoles, but only few games approached it. And that only by small portion.

edit: Also I would like to add that during the release of the new consoles after the 128 bit generation, the first game that felt next gen for me wasnt on a next generation system. It was on PS2 and that game was called Shadow of Colossus. Everything else seemed like older games with prettier visuals.

 

[Laa-Yosh]

I would love that as the most impressive thing for me about the so called "tech demos" that Sony presented at E3 2005 wasnt simply the level of realism, but the level of animation, interaction and physics with the environments and characters.

If you mean the KZ and Motorstorm trailers, maybe 8 days, then know that they've all been pre-rendered so all the stuff I've listed as advantages to the approach apply to them as well.

Thats what I wanted to see this gen from next gen consoles, but only few games approached it. And that only small portion.

That's because realtime offers far less resources for the calculations (we offload them to the render farm) and no way to fix problems in 'post', either in the 3d scene by pushing vertices around or in compositing by painting out the problem. Although maybe one could develop image processing algorithms to fix stuff... but I doubt it'd work reliably though. All in all, even with a lot of processing power some of the issues are just not possible to solve. But fortunately the audience is far more forgiving with games