A different perspective on Unity development.!

May I take a guess?

Because Highend PC Games is the least of their problems. PC Games work fine, and on decent hardware, even an engine that is not top of the line like Unity will perform decently with some optimization.

Unity Stock, with some slight massaging and rewrite of some of the more important part will look close to what you achieved there (not trying to devalue your work, looks really stunning)... not exactly the same, but close enough for 90% of players and usecases.

In the meantime, Unity is getting bogged down supporting many more platforms, some of which have far more constraints when it comes to the platforms median hardware (mobile), or have way more visibility when it comes to players (with consoles still beating PC games AFAIK, at least in total).
Some of these platforms Unity has still some catching up to do (don't know how many Console games are actually built with Unity)...
In the meantime new versions of DX and OpenGL will only worsen the amount of Graphics middleware Unity has to support.


Unity tries to be everything to everyone...

Which is something I noticed with Unity and asset stores assets a long time ago... if you want the latest and greatest tech, Unity is the wrong place to look for it.
But it will end up there, eventually. And if you can wait some more, it will also work and perform.


Which brings me back to the first point. Unity might fight a loosing battle trying to support every platform under the sun, every version of every Graphics middleware out there, with all the different renderer types imaginable.

But I kinda like their wait and see approach when it comes to new tech (if it would be a willing decision, and not just the devs not having enough time to catch up). Oftentimes, I was attracted by other engines that DID implement the latest and greatest out-of-the-box... only to find out that the latest and greatest wasn't so great outside of showcases and proof of concept projects.
Temporal Antialiasing and Unreal 4 anyone? Yeah, that never worked for me. Either you got insane motion blur when fully cranked up, or AA was so weak that FXAA worked better when dialed down several notches.
I will not get my hopes up for Unitys implementation behaving any better, but I think Unity had other fights to fight at the time when Epic released that not-so-awesome-as-advertised tech. And I am glad they didn't waste their dev time on THAT.

At some point, somebody will make TXAA/TAA work. That implementation hopefully gets opensourced. Unity, or one of the Asset Store devs will pick it up and bring it to Unity. Maybe it will be 6 months after a working solution is found... I don't care about that.
I care even less if its 3 years after epic tried using TXAA to sell their new engine... because the initial implementation was not really working over a wide range of configurations.

It could be very well be different for CryEngines SVOG... might be working fine, in actual games, WITHOUT needing CFX'd or SLI'd highend cards to reach acceptable framerates.
Still, my stance on that is the same. The fact I found CryEngine to be a bitch to work in, full of bugs and with a dated editor tells me heaps about where CryTek is going wrong with THEIR development priorities (or was going wrong, they might have stepped up their game in the last year or two).


TL;DR: I rather have Unity work on the important stuff fast (like fixing lightmap baking), than have them try to implement latest-and-greatest tech that is not that great yet. Even if the tech is great, I don't mind the wait as long as Unity is running fine else.

 

And then once you dig a little deeper the really weird stuff - you start to wonder why it allocates something like 20k per frame... and good lord, that cloud plane shader. 1500 lines of no ifdefs.

 

You forget that CryEngine is build from ground for performance and open world unlike Unity

 

This isn't AAA obviously, but it's pretty interesting regardless, since it's running on an iphone.

http://forum.unity3d.com/threads/the-secret-to-great-terrain-on-mobile.305899/

Stock shaders and stuff, pretty solid friggin results. Really good work.

 

Which is moot when devs then take the engine, throw unoptimized art assets into it, write horrible code and call it a day. Result is guaranteed to run way worse than a well optimized Unity game.


Anyway, I do hope Unity concentrates more on important stuff in the future (so it will also be "built from the ground for performance"... which means yes, even Unity will have to re-engineer some stuff from 2007)... asking for more fringe high-end effects like Voxel Based GI most probably detracts from that though. Unity should expose enough of their engine so others could do that efficiently (for example by open sourcing it, and allowing people to rewrite parts of the engine to either let Unity roll the changes back into the main branch, or sell the changes to the engine as a product)... they shouldn't waste their time doing that themselves.
Yes, I am happy about the new "standart image effects"... no, I don't think this was a very efficient use of Unitys limited inhouse dev resources.

They are not Epic, nor are they CryTek. As long as they are not developing for PC first and foremost, and are not a game developer at the side too, they clearly have to take a different route.

 

Well, they do. It's just that we PC focused people aren't too thrilled about that route ^^.

 

Artist use composition
It's very effective!


There is a reason I'm constantly making throwback to old graphics instead of hitech redacted images. Tech won't solve your image problem. Big company that have great visual still use tricks inherited from the psx/n64 and build FROM that to get performance/great image. More than half of those trick are composition first.

For example I'm still seeing people tossing crunchy 4k textures to get a better images on full HD when actually it make it worse! The ratio of the texel density vs the screen density create aliasing artefact, in fact to get crisp image you need to actually BLUR your texture so the texture sampling projected on screen appear crisp! That's counter intuitive but it make sense once you understand what happen.

DS games is the best to study because they have the limitation of old with the skills of now. But profane people will just bounce back from low tech graphics ...

 

Yes, because they try to be too much hands on at the moment. That is what IMO consumes too much of their time. It causes subpar stuff to be released, while old components are not rewritten for far too long.

IMO:

1) concentrate on core development tasks. New core functionality like velocity vectors or instancing: yes, please! Writing their own new Image effects: Nonono... just no*
2) Make sure others can improve Unity as efficiently as possible. Instead of counting on people to hack around your engine to replace old crap or implement missing functionality, concentrate on a plugin architecture that is EFFICIENT enough for a realtime engine.
Better yet, release the source code for the FULL engine and handle what people can do with it. Make it possible to change parts of the engine, and sell them to give incentive for innovation.
2) Use the resources freed to work on the engine core more often. I am pretty sure Unity 4 was still the same engine as Unity 3.5, with just a few components rewritten. That most probably still holds true for Unity 5.
Yes, the engine doesn't need to be rewritten from scratch every 2 years. But make sure to not drag along old stuff longer than needed.


* I am not against new stuff available for everyone, but:
Knowing Unity, their implementation is most probably subpar to what the asset store devs could do (and will do in the end).
The only REAL gain in having Unity do it would be for them to implement it deeper inside of the engine, similar to how it is done in other engines (so no single effect image effects tacked on to the cameras, but deeply integrated effect stacks that do all postprocess effects in a single pass). Of course that is more work. Of course they do not have time for this, because they need to micromanage new image effects.
Writing a new framework to run the effects in that works in the end similar to what Unreal offers (A single stack postprocess option), while giving asset store devs the needed Interface to develop their single effects which their customer could just slot into the framework and have them run together with other effects form other developers, managed by the framework to run them in a single(-ish) pass, THAT would be what I wished Unity would spend their time on.

Though I understand telling people "we have developed an awesome new framework that will allow you to run all those future effects that still need to be developed by other people in one go... and we don't know by when you can expect a TAA implementation" sound less sexy to most people than telling them "WOOT, new TAA effect. just tack it on to your camera!" (even though the end of the line should be "... along with 20 other image effects and see your performance be consumed by the overhead")

 

I agree @neoshaman.



Composition is very important. But technique is extremely important also, especially when targeting low end hardware. It's important to learn how things work, or else you end up running at 5fps, and all you got are screenshots.

 

I don't think that's a totally fair assessment. E.g. I think they've stated that there are plans to combine multiple image effects into a combined and optimized version, but they want to get them working in isolation first. Also I think the new tonemapper is better than for example Scion's and I'm using the new SSAO instead of SSAO Pro. People have been harping on forever, about how they want decent post out of the box and now they start to deliver. You might not have faith yet that they'll be any good, because their stock fx from I don't know how many years ago are kinda meh, but so far it all looks very promising to me. We'll just have to wait and see. Imho there's no point in arguing about whether they are on the right track with this or not. We could both spend our time better.

 

I side with:

Although the new tonemapper is f'n awesome. I really wish they focused on rebuilding the post stack. It's the right thing to do in the long run.

 

Also all transparent textures, apart from UI, are cutout transparent. Looks a bit uglier, but gives you 50% more performance in such a crowded scene.

 

They are.

 

Source?

 

Tim C in one of the threads. Said it'll be combined / optimised when its finished or something.

 

No like rebuild the pipe so that you don't need to combine post anymore. As in like, it all works in a single pass at the engine level.

That's what we're talking about here I thought.

 

I want to precise, I don't diss on tech, It might seems as much, tech do give you more freedom. I'm just pointing that there is a lack, at large, of discussions about the role of composition. People don't pay attention because tech give immediate and understandable graphic impact, it looks like magic, therefore people consider it as such.

Composition is subtle and has subjective quality, so even though solid foundation are universal people don't see the impact immediately and are confused. Since tech talk are covered I'm promoting composition to be sure it stay in people's mind.

But more importantly, maybe I should focus more on how tech allow to achieve composition strategy. For example ssao makes an image better, but why? The truth is that it help, in realistic and chaotic composition, to make shape pop out from background to avoid them blend together, and it allow to ground objects together by adding contact cues.

If you look at ff12, they didn't need the shape poping because they use value range to separate objects and background (they aren't on the same range allowing for clear contrast), however ssao would help objects being grounding together because there is no easy solution to contact cue with the tech and method they had at their disposal.

In modern realistic game, objects and background tend to be on the same value range (bad art direction generally, movie control value range very carefully), ssao and other trickery help a bit out of the box, but value range can still be controlled and some game do it, tech don't replace proper planning and tech contribution is generally giving a rather arbitrary tacked on composition.

I'm often put in a difficult situation, where people like an image for being good, because "tech" ... The thing is they might like how a fog give an eerie atmosphere by silhouetting a forest into plane. The problem is that it's accidental composition, it still not very balanced and not well distributed across objects or visual interest ... Same for every other tech.

And when I point to good composition, it seems that it's mostly colorful image or exaggerated composition (limbo for example), so people think that color = art direction = composition, which is not the point. But it's true that the tropes used in colorful are design to make silhouette pops and have visual interest, it just seems people use them naturally because they are exposed to them, and use pure imitation, rather than understanding and mastering the principle. You still have game like left for dead who have a very traditional rendering yet a strong composition understanding (see valve paper on their graphics, contrast to team fortress 2's paper).

Being able to analyse an image, along its visual problematics, should help tech shine even more, using them where they are actually needed, and avoid overkilling by stacking accidental expensive patch, to help badly planned visual elements to begin with.

 

@neoshaman

Things is when I talk about "tech" I presume the fundamentals of "composition" use of colour / tonality / placement etc. etc. is already taken care of, because if you don't understand that bit then tech with it's hundreds of little tweaky bits is only going to make things worse, not better.

If you're going for specific types of games though, tech (well lighting composition elements mainly) is very important. Again though, you've got to understand what the problem is to define whether said tech is any good in the first place.

 

I see, well I'm talking about combining post FX so redundant work is not there any more. I don't understand how hardcoding it into the engine vs having it part C# and part shader would actually make it a single nanosecond faster, because the engine would also have to do this.

In Unity's case it has done so, it has altered the engine to accommodate better motion blur, antialiasing and so on including using MRT to further optimise. Isn't this what a "combined stack" would do or am I misunderstanding?

There is no single pass either, these effects do require multiple passes, I've probably misunderstood what you mean. In short I don't think it gets any faster than just combining the cinematic effects into an optimised whole since they would do the same job in the source.

 

@ShadowK
I hear you, but I don't think you can separate the two, you can't take care of one separately from the other, tech is your composition tools, so I'm worried when I see performance talk without related composition talks to accompany it. I mean talking about composition IS defining the problem! So why there is very little mention of it

For example we have talk about foliage a lot, mostly in term of pushing a lot of grass mesh irrelevant of the composition beyond it "looks good"... So let's look look at a low tech game (xenoblade) who have dense grass and great draw distance relative to its hardware (a soup up ps2). In term of tech it's 10x weaker than a ps3 and therefore 100x weaker than a ps4. SO wouldn't that mean we should achieve 100x time the same result?

I'm using hirez emulated rendered version, so flaws (and trick) should be exposed easily by looking at it. It was design for low rez.


It's decent! how did they achieved it at great performance on such a weak hardware? There is certainly instancing but there is a lot of visual composition trick too.

- First the grass is really 4 textures.
- You have the "background grass" texture that comes into 2 variants, one that is plain grass and one that is barely distinct wheat style stuff, they have narrow value range but the wheat stuff break shape in an imperceptible way to avoid monotony, despite their low value range they also have a light streak that contrast with their base green (also compensate with the lack of lighting beyond ambient time of day), creating a high frequency pattern that is self masked by each other, and since it's moving (the grass oscillate but not all in the same direction at the same time), it give an impression of overall details and parallax that give depth, ie it create an idea of density.
- The base grass is design to work with the ground texture such as at a distance they blend in visual noise together, which allow to seamlessly fade together when the grass in remove, the visual potency being the same as the ground.
- Then you have flower to break up the value monotony and provide visual interest with resting place for the eyes, then you have darker brush that does the same with a twist.
- Flowers fade much later than base grass because they have higher visual potency but not much as their visual contrast is low, they end up mixing the distant visual noise anyway. The bush however who have a high contrast but is much more sparse, keep being rendering way longer in the background without fading, because he has visual potency, fading would be visible, but also it maintain the illusion of continuity of the grass field much longer! And we are using high rez screen so it low rez, for which it was design for, it's much more effective!
(edit: there is also some mip map trick too, but hey)

Why it's important? Well in early example given of foliage terrain, the grass texture where either too busy yet monotone or too plain still monotone, which lead to shallow visual interest and stack of effect (ssao to regain the parallax, fog to regain shape, polycount to regain density).

That's this observation that prompt me talking about composition. We need context to talk about tech, in this case composition was the context. It's not just overall composition like "framing" like the example of @frosted but balancing the parameters. Lighting was important in that xenoblade foliage too, but they took another way to make it good. Hundred of little tweak isn't just tech, you need to take care of both at the same time.

I'm sure this observation about xenoblade grass can help push better grass on unity WITH the current terrain performance. And on top we may achieve even better visual by adding the tech where it matter, but only after talking about the visual and composition problem. I'm just promoting that, explicit the composition problematic first then show how tech solve it, put back the context!

 

@neoshaman,
I think part of the problem is that very few of us are making a game. Fewer still are actually worried about building actual playable content yet. And finally, fewer still have any idea about how to approach things like visual composition from anything but a "hey this kinda looks good!" way.

Most of us are tech guys. For a tech guy, I think I have pretty good instincts for stuff like composition, but I am not a trained artist. When you post a detailed 'composition' analysis, I definitely appreciate it. In fact they're some of the only high detail posts I bother to read here. I've gone over a few of them more than once, that's saying a lot.

I would ask questions, but honestly, I don't even know what questions to ask. So you're going to have to just take the initiative on the 'composition' discussion.


@hippocoder - my understanding is that UE has a post pipeline that doesn't operate on 'chaining' and can reduce the number of samples and the like required.


Finally, I just wanted to share this little video of my terrain editor... it's gotten completely friggin insane. I ended up spending 2.5-3 weeks on it. I really shouldn't have, but... good god it's F***ing cool.

Here I'm generating terrain for that screenshot. I sped up the video so it's only a couple minutes long.

 

Yeah that's what Unity is doing.

I like your editor, should totally asset store it while you can. some kind of $25 job anyone can afford for an alternative to gaia and friends.

 

Yeah, honestly I don't like gaia very much, it's too ... slow? I like being able to see what I'm doing while I'm doing it. Obviously the stamp idea was taken from gaia the rest were taken from 'terrain auto material'

So the tool basically does what auto terrain material does (brush based approach, combine multiple details and multiple splat textures into a 'material') and gaia's stamp idea and merges them. It's really good. But I gotta stop working on it, and I didn't get to some of the extra stuff like trees, etc.

Reality is, I should have just bought the auto material thing.

 

I managed to get around 160 FPS (w / out vsync) and a draw distance grass wise of about 300.. The quality isn't quite as good, but I don't believe the trade offs are that bad??

 

God, I need to buy a new machine. I don't think I get 160fps in an empty scene!

Nice work dude! Any useful tricks?

@neoshaman I've been thinking about the composition discussion - could you post any notes you have on how I could improve some of my screenshots? Mistakes, missed opportunities, etc? Might be useful to actually use my work since I can take some of your suggestions, make edits and we can look at the 'before' and 'after'. I will not take offense if you are brutally honest.

 

@frosted

Yeah I keep talking about it, to introduce those concepts to people who might not be aware of them, and maybe will start learning and looking about them, to make cheaper and better looking art, without being an artist.

I don't think these concepts ask for finesse and actually tech minded people can turn them into process and even code. Afterall concept like "value ranges" (basically brightness ranges) is really just a range of values! Contrast is really just the difference between too value, and you can use that to evaluate a high pass filter to find variation of texture, it's computable!

 

@frosted

Okay I can look into them, but I have to be frank, before the discussion I had no idea about the xenoblade stuff, in the process of the discussion I discovered how it works lol, context give focus. I'm not sure I can currently do paint over edit though. The truth is that I'm a game designer, I didn't do proper art in ages. But to rebound on that proposition after I have seen @ShadowK post:

I wonder, if there is possible to have some shorter different grass JUST at the transition from grass to the path? Same value/colors than the normal grass. Just something I was wondering.

That looks good too!

 

I took the speedtree billboard picture for the grass, painted some details on it in photoshop.. Then tweaked the living poop out of the quality settings.. Also doing things like modifying the alpha cutout, the less opaque pixels the less the DR has to do the better.! Or so I'm told.

@neoshaman, it's nothing more than a test.. A lot of it will be mesh based, I just need to know if I do decide to splurge on a little bit of foliage it won't cripple me like it does in UE4.!

 

That edging features to place thing at transition is a stroke of genius! Does it place texture or also objects?

 

@frosted

I gotta ask, why do the normals on the terrain look so damn flat? I mean it's like it's not even rendering them.!

 

It works by looking at the splat, then placing everything else against it. I build some 'custom' splat for things like road, so it knows these are special and does not overwrite them with base material rules. So the edge of a road or something can have rocks, or special grass or whatever. Everything is driven by the textures on the terrain, so it will also work with edits made using old unity terrain editor.

You have good eyes. I disabled a bunch of rtp features for the video and forgot to re-enable them, so it wasn't doing all the fancy pom magic... or even plain old normal maps. Whups!

 

I have NM's on mine and it still looks flat???!

 

Honestly, it's been a while since I worked with stock terrain shader. Could be ansio levels if you're viewing at steep angle. I noticed that if you don't use "force ansio" in quality settings that the terrain engine will ignore grass ansio levels so it could be some weirdness there. I would definitely recommend you use an alternative terrain shader. There are a ton that are good, heighblending at minimum is really important and incredibly useful for stuff like tiles or stones in villages and the like.

 

Nice.

 

Well, it might turn out their new effects really blow all the thirdparty effects out of the water. I of course would highly welcome that. After all, more competition is always good, and I might not have too much faith in Unity delivering something good out of the box, they DID fix some things over time, and Unity 5.X starts shaping up to be a quite good version after a year of fixing that horrible 5.0 release.

It might also be that they WILL combine their effects versions into a single stack in the future. Also good.

But: Why push out single effects now, instead of using that time to combine them? Yes, combining the old and tired effects will not interest anyone.
But why not try to go further than just combine some effects into a single stack, like what I have outlined?

At the moment Unity is playing a game of catching up to others without really thinking outside of the box. I would rather have them spend their time thinking outside of the box, and creating something which brings value above and beyond a single effect stack.


What if that effect stack is pushed out in a year, but by that time an asset store dev wrote a way better SSAO implementation, another one an extremly cool looking new AA posteffect, and there is an opensourced implementation of volumetric lighting floating around the forum that also is using a posteffect?
With their current plan, you again will have to use multiple effects again, the value of the stack is diminished (maybe you can use DoF and the Tonemapper in one go, but you still have a seperate SSAO, AA, and Volumetric Lighting Posteffect).
With my idea, and with the asset devs really following the new posteffect framework structure, you could slot them into a single stack, no matter who wrote the effect. That is an innovative approach that would benefit Unity users for the next few years, not only until something better comes out and Unity has to play catch up again.

And while everyone can write good posteffects for Unity, not just Unity themselves, only Unity could write such a framework that has to be integrated quite deeply into the engine, AND give that new frameworks API enough weight so enough asset devs will support it.


Just to make it clear, that single stack postprocessing framework is just an example I came up with. I am sure there are better examples. The point is that I think Unity currently is catching up, doing micromanagement writing small building blocks, while they should be working on bigger things, innovating and moving their whole engine forward.
One of their biggest strengths is their community. It was this community of Asset Devs and people opensourcing their efforts that drove Unity forward, especially in the PC high-end space, while Unity was busy juggling all those new platforms during the 3.X and 4.X days.
Maybe Unity should leverage the power of their community yet again, but provide them with better tools to drive the engine forward (like instancing, velocity vectors, or my proposed postprocess framework).

Really? I always had the impression every postprocess effect worked in isolation, thus incurring some overhead, while combined effects (like what Unreal does) might not work "in a single pass" in the literal sense, but at least can share some things by clever caching, and have way less overhead.

Might be that I completly misunderstand what Unity does behind the curtains, and blowing something out of proportion...


But MY point was something different (see above)... the question of the single stack effects and how to best achieve that was just an example of my point of view.

 

If it won't, do you plan to move your game to Unity?

I like the screenshot you've shared. Personally, I would try to extend dynamic range a little bit.

 

Y'know it completely depends, I have two options in front of me.. I could either go completely mesh based which adds on a fair amount of time to the art portion or I can use Unity and go down the terrain route which adds a lot of time in tools dev. Ultimately both Unreal and Unity have issues in areas and irrelevant of what I choose it's going to be a slog in one area or another.. If I go completely mesh based (as in limiting heavily foliage translucent materials) without a TOD system it's irrelevant what I choose as it'll work great in both.

Whilst doing the artwork portion I keep checking up on both engines to see where they're at, both are improving and Unity this time around has been favourable if not a lot of work.

 

I see, make complete sense. Did you check out Cryengine for your use case yet?

I'm sticking to Unity at the moment but plan to devote some time for R&D where I want to learn and prototype some things in CE and UE4 just to broaden my horizons. I know that CE has this common belief that it's extremely hard but I've seen some of the new tutorials from Collin Bishop and I really liked what I've seen. Besides I believe that once you know one game engine it is a lot easier to learn the next one.

Anyway Unity is awesome, is progressing a lot and I own lots of stuff from Asset Store. I also love Unity services, it is something that helps in the business side of things and is something not yet seen among the competition.

 

Y'know I've said before UE does have some majorly cool stuff in it and one of the biggest things you take out of all of this is play to the engines strength's..

Like @neoshaman's idea of a desert based setup is a great idea, could do a springs style setup for the game and make it mainly mesh based which would increase performance dramatically.! By default with UE4 in a mainly mesh based scene my full rendering / CPU cycle overhead is 8.33 ms @ 120FPS with thousands of meshes.

All I'd have to chuck away is the terrain textures / generate a new world machine setup. Might also try that in Unity too.!

 

Desert based setup?

 

But you'll loose all the apeal of grass everywhere and trees swinging in the wind

 

Well I've been messing about anyway, if nothing else performance is amazing!.. I mean bouncing off the limiter.

Managed to crank the grfx cause of it.! Excuse the "makeshift" artwork I did in like 30 mins, it's just a test.

Also, good news everyone.. Epic know there's a problem and they are adding a forward rendering path.>!

 

Where did I talk about desert based setup?

 

Oh sorry it was billy ..!

 

I was afraid I wasn't following myself lol, I do sometimes forgot so I was worried it happen again.

 

Hey guys, Thought i'd post here since you were discussing my game that that I've released a trailer for it and our store page will be up on steam this week. Also the games being showcased at Indie Prize this week at Casual Connect.

 

Good work. I see basic directionnal light colors and over exposed bloom, why didn't you used some LUT and subtle light coloring with tone mapping ?

 

One possibility is that he was simply too busy making an actual game.

Still... it would be interesting if you graphics gurus can help him to quickly tweak it up a level. A good experiment.

 

Hey Folks!

Thought I'd post it here that my game is finally out. Released it on it's 4 year development anniversary. Check it out!

Store Page:
http://store.steampowered.com/app/486020/

 

Congrats. I saw your email right before I went to sleep last night so didn't have a chance to do anything. I'm looking forward to grabbing the game from Steam tonight.