Most people’s GPU doesn’t have very much memory. People go through memory very easily using Lightmass so if you tried it with a GPU you’d need a lot more memory than what cards currently have.
I have yet to try the new lightmass baking improvements that have been implemented from the Intel library, but I heard it does make quite a difference 
Thanks for the clarification Daedalus51 but i have one question when it comes to dead island 2 … The game was clearly designed to be a kind of open wolrd game, like Far Cry, The Division or MGS V etc…
And GI was clearly a big point from what i have read… Why you guys simply didn’t used Directional and SKY light ? What was the reason to try alternatives, scene quality ? Performnace ?
My point about GI is, no matter if we are talking about Lightmass, VXGI, SVOTI etc… It makes a game more pretty and even realisitc. If a good GI solution is not that important, why have you guys tried so many different systems ? Why not making things easy and simply just keeping directional and sky light ?
Just forgot to ask, have you guys at yager every played with VXGI around ?
Hey Adik,
well first of all…I think I never even in the slightest implied that a well made GI solution is not important 
That is the reason why we tried so many things…it totally is! However, sometimes in a production it is just not doable. You think you can make your game with lightmass and then suddently, it turns into an open world game and the tech does not yet support this. You try to find solutions but you fail. This is the moment, at least if you didnt go this route because of a more stylized style before anyways, where you start saying…you know…maybe we can somehow make it happen like this.
Also…at least in our case, before we started working on the AC4 lighting tech, we had almost no graphics coding resources, and shortly after we were almost done, the project got cancelled. So yeah…GI is very important. Its just ****ing hard to do and most people have to fake a great deal. Of course everyone is looking for a holy grail and the idea of it sounds too great. But I dont think it will happen this generation
When it comes to quality…its a mix and match. VXGI produces quite some good results…maybe the best, but its darkening is too strong and it has loads of other issue like bleeding and performance things. Some of the other techniques can produce better results when combined with more traditional techniques like SSDO and other, however the work of an artist is equally important! Also, those more traditional approaches all dont support dynamic geometry. This is really the hardest thing to do…and proper area lighting even from quad sources.
Also, when it comes to resolution, its way more important to have high frequency detail inside compared to outdoor areas. Honestly, if we could combine static lightmass lighting for interiors with dynamic distance field lighting for exterior, we could make some already **** happen. But sadly UE4 doesnt allow to combine the two. There is this problem for outdoor areas with lightmass which we ran into as well with DI2, and thats foliage. You cant afford lightmaps for foliage in a huge world full of stuff 
So if you have quite some foliage, you have to go dynamic, and distance field AO and heightfield GI look great outdoors. But it all falls appart as soon as you go inside 
Well, considering performance, as you have asked, the SH stuff combined with optimized DFAO was what we had in the end. A probe was placed every two meters, one meter above the ground or things. It looked pretty solid and the performance impact of the probes was basically not noticable. However, they took a couple MB of memory and took roughly 2h to bake for a complete area. So I would say for our situation the best we could have done 
Regarding VXGI, yep…I played around with it, but I honestly believe it…well it doesnt suck completely, its just not as cool as I thought for many reasons, including questionable performance, confusing settings, skylighting issues, bleeding and over saturated and too contrasty bounce. Yeah you can tweak the settings and I tried a lot, but achieving good quality while maintaining performance for both interior and exterior areas was basically impossible.
So again, it all comes down to the type of game you are making. I remember Doom3! They were rocking with normal mapping, dynamic lighting (no GI ) and highly detailed characters. If you say well why didnt GTA look that good at the same time? You are taking a lot of things not into account. Doom3…the whole game was tailored towards the fact that they had no proper ambient lighting and their **** was so expensive to render that it was just dark narrow corridors with like 2 or 3 characters on screen at the same time. They just couldnt afford more
So again, GI is supper important, but its the hardest thing to do…kinda, like proper lighting in general I would say. You can make even ****** materials look good if you could render light like in real life
Cheers!
EDIT: sorry for swearing so much
I wonder why though the 3D packages don’t pick up these GI techniques, because something like VXGI can be turned up and get very good quality much much quicker than rendering in Vray or Mental Ray, even if it takes a full second to do a frame that’s way faster.
Because they are relying on 3rd party software, like V-Ray RT or Iray, to come up with a GPU accelerated solution.
Thanks again for clarification Daedalus51, i was like 100% sure that current games have and need GI solutions for Dynamic light… and that UE4 i like the last engine which didn’t take advantage of it 
By the way, it’s sad to hear that dead island 2 was canceled… I heard about that a few months ago…
Those are different types of renderers. They have viewport renderers for 3ds Max and Maya, in some cases it can get the results you need in the viewport, but they should add something like dynamic GI to improve it more.
Hmmm…those are lazy farts I dont know…there isnt a lot of inventive stuff going on besides modo which seems to really do some creative stuff. And although I dont like Maya, the new versions really do some neat things as well. However, when looking at autodesk, they do things their own way and dont seem to care that much. It feels really weird^^ So who knows
Be glad, Enlighten is horse ****
One quick question - what do you think is a realistic “maximum size” an outdoor map with foliage could be and still have baked lighting? I know this is like asking how long a piece of string is, but I am mostly looking for a very ballpark-ish answer like is it closer to 10 square meters or 1000 square meters?
The size is not what matters. Its the density of content and the lightmap resolution. You can use a hybrid approach like what rabellogp does as seen here: https://forums.unrealengine.com/showthread.php?135650-Exterior-Light-Studies
If you combine Lightmass in a smart way with dynamic lighting, you can achieve some really great results. I would argue that, of course having a bake farm too, you could probably populate the landscape of the kite demo with dynamic vegetation and static buildings and still kinda bake it. Not in super highrez lightmaps most likely, but it should be decent.
Of course you would also need to look at you texture streaming pool settings and one of the most important things that would complement this is using layered materials with a shared library to maximize texture usage and share as much as you can. This is something that would need to be carefully evaluated budged wise but I think it could work
Cheers!