That was the biggest selling point for UE4 in 2012, all the other engines now have all the features, CE has better shaders and Unity5 is getting better lighting, which puts UE4 behind in terms of graphics. (all this just to please the consoles)
The game that brought Global Illumination back to UE4 is being made for Xbox One (Fable Legends). UE4 also has support for mobile phones (which are weaker than consoles).
Dynamic Global Illumination has been the holy grail of hopes and dreams of what to achieve during real-time rendering. Lighting is very much the most important factor for gaming engines at the moment and coming up with a solution to calculate indirect lighting that has a minor perf cost and good results is a major achievement for current engines. When UE4 first debuted, SVOGI was hands down the most talked about feature and the one everyone was focusing on having for next-gen. Last gen was more about post processing where this one is very much all about getting accurate lighting simulations. It is a huge selling point if any engine can do it right and efficiently and if you bothered to pay attention to the many threads and discussions about GI, you’d see that it is extremely high up on the list of popular most-wanted features for UE4 not too mention, all the articles by the media that specifically highlighted this feature more than anything else when the engine was first shown off to the public.
I would have to agree with the various posters here: SVOGI was THE selling point of UE4 when it was debuted in 2012. I don’t think that can be argued. Now we don’t have SVOGI, because it wasn’t the best solution for the engine, but at this point UE4 needs a good solution and soon. With UE4 gaining pretty significant adoption rates, Epic is going to have to really deliver on the dynamic lighting front soon, because Unity 5 and CE3 will be way ahead if they don’t. Now, don’t get me wrong, I think that blueprints are the biggest and best new addition with UE4 (along with Code Hot Reload and other usability features, plus of course, PBR), but many people in the public will not see most of that. They will see dynamic lighting get highlighted this generation and they will see the visuals. At the moment, I would argue that UE4 is the best rendering engine for indoor spaces and relatively small play-spaces, but when it comes to the big open worlds and all that, CE currently blows UE4 out of the water.
I think Epic will be doing something about this; they know that they have to. So now we just need to sit back, continue discussing, and wait for them to blow us away with whatever they choose to go with.
The current situation being Epic will further improve the LPV GI solution is better than if they would spend resources on integrating some fancy but costly GI solution like SVOGI or VXGI. If Epic can provide a full-fledged and improved LPV integration in the upcoming months (this would make for an Christmas present :]), they are in a good spot compared to CryEngine and Unity I think.
Don’t you mean Lionhead?
Three weeks ago I asked this:
“I wonder what the timeline is on improving the LPV integration at the moment (like the improvements Martin mentioned in the rendering update stream from April 10th)”
Answer from DanielW:
“I asked Martin what the improvements were he had in mind, will see soon.”
Ideally Lionhead will provide Epic with their latest LPV code and they will take it from there, I hope the new Fable won’t stand in the way of this. Like, it has to be released before Lionhead is allowed to share any code and stuff like that…
Actually I read in a few staff posts that just a few things are missing from code shared by Lionhead (and what is missing will probably be in 4.5), so looks Epic will work in a custom solution now to improve the LPV or implement other system.
I don’t think LPV is a path that will lead to a method that works the way that people want
O noticed that in 4.5, RSM step takes about 10ms(!!) for LPV. I don’t know what is exactly included in it (only generating RSM, light injection, etc)m but that’s just way to much.
Porpagate step is another 5ms.
In the end, it’s about 15ms, for GI, that doesn’t look that good to begin with, where light propagation have short distance, thus producing artifacts, and in which specular reflections are not as good as, as with cone traced solution.
That being said, technique have it merits, but there is something broken in current release, and it could be really changed to use voxelized scene repesentation for better results, as I posted before.
I’m not sure if something that can take 12 ms per frame on a 770 could be called the “biggest selling point” of a game engine.
It’s a cool technique, but good luck getting any acceptable framerates with that.
Hmmm…interesting. In 4.4 its also eating up 12ms at least (in our game), so this might not only be broken in 4.5. Do you at least have more features for LPVs in 4.5? Like skylight injection or so? Because in 4.4 its still not very usefull anyways since everything that should be sky is just a yellow/greenish blurry mess. An there is nothing that looks close to what Lionhead showed in their blog…and that was still not feature complete but already way ahead of the current implementation.
Also what I would definitely recommend is, to exclude small geo from the LPVs. I was able to gain 8ms in the office by just removing everything from contributing to the LPV thats not walls, roofs, big things like cars, beds, wall units etc. So all small stuff was completely removed for contribution and that already saved our heads in regards to running on consoles. Despite the fact that the current LPV integration only really works for PC and XBox One…PS4 is something completely different (guess Lionhead didnt bother with that since they are developing for MS^^)
No, it’s pretty much the same.
Well funny thing is, that most of my scene is composed from walls, with wood beams. It’s hardly complicated geometry. I say pretty standard cottage medieval houses.
I would be nice to get detailed break down of that RSM stat mean in stats.
Basing on what saying about smaller objects, I think voxeling geometry in scene, and then injecting lights, would improve performance. Somwhat, as voxelization step would get rid of all small geometry, and left only fairly simple data structure to inject light.
Anyway, my point is. If Voxel technique will eat similar amount of ms, and look incomparably better… That is no brainer to what is of better use.
12ms, yikes! I still haven’t tested LPV myself as I’ve heard there is an issue where if you have Material Layers with a layer that acts like a glass that calculates refraction, that it can crash the engine. I extremely want the need to drop Lightmass at this point and go fully dynamic. The new game I am developing is primarily based in outdoor environments so going with a good GI solution has been my ideal goal once I can take advantage of it. At the moment, is there a thread that already lists all the current issues when enabling LPV? I’ve looked and I’ve found scattered results.
I also agree, I do think Epic is going to be looking real hard at what the right approach is on this feature. VXGI is being advertised as only working on 9xx series but since this is part of GameWorks, one could hope that this would be scalable enough that it could exist on more than just PC and GTX 9xx or above. Like Ray Davis mentioned, Epic probably needs some more justification to handle the implementation themselves if platform support is too limited. I personally haven’t looked into it but I am not sure if the other engine’s GI solutions are multi-platform. If so, then Epic needs a solution that is cross-compatible as well if they really want to compete with the ones currently available. Also, if I’m not mistaken, isn’t UE4 the only next-gen engine that doesn’t support SLI?
So, just to pick up some things you mentioned:
First of all, you can prevent the engine from crashing because of translucent materials if you do these two things 
Actually, even if the name for the second thing (check the box…the not checked box is the default^^)doesnt sound like doing this (because it actually makes indirect light from emissive) but its the recommended way of disabling a translucent material for the LPV:
So…as far as I know, there is no whole thread covering good solutions for dynamic lighting just right now, however, I would say that I could give you quite some good insights on this in a more fitting environment (than just spamming this thread :P) if you want to^^ But hey…we could also start a thread like this and discuss what people do to achive certain results dynamically in UE4 right now! 
That sounds actually quite cool^^ (Might do that…but I think weekend would be better for this…have to collect all resources and screens first^^)
Gut feeling about what I would start out with if I were you, I would definitely still recommend LPVs with the mentioned optimization for you (remove small stuff) in combination with a skylight (but then it gets more tricky if you use a dynamic one, because you dont wanna pay DFAO with LPVs together^^).
Yeah…just some thoughts on this, and now I´ll be quite with stuff thats not really thread related
Cheers!
In addition to Daedalus’s tips you may want to use a skylight with little intensity to give life to translucent materials(especially glass) since they dont receive any light or reflection with LPV only. Also LPV looks pretty well in a GTA/Watchdogs sort of environment. Check out the video in my last post here: Vanishing Point (Not the '97 remake!) - Film, TV & Animation - Epic Developer Community Forums
I still got to try the console command Daedalus mentioned in 's thread and see if i can handle interiors(shops and stuff) together with this setup. Using PP volumes for interiors to disable LPV and using reflection captures should work i believe. We’ll see…fun stuff!
Having done some voxel cone tracing stuff myself, this is indeed the current problem, and NVIDIA’s own implementations have been consistently disappointing in terms of performance. I wouldn’t bet on anyone actually using their provided library.
Here is a recent talk by one of the developers () of NVIDIA’s clipmap based voxel cone tracing technique (which is presumably what they call VXGI in the Moon landing article). It looks quite promising because they have implemented it in a very scallable way ranging from only 12MB to 2.5GB of memory usage. One of the slides shows that GI only requires 12.9 ms on a GTX 770 and allegedly allows for better quality AO than screen-space techniques even on low-end hardware. At lower resolutions there is a lot of light leaking, however. Light leaking generally remains a problem for thin objects, but I could imagine that’s something artists can work around.
.jpg)
If you believe this table VXGI is new library.
Yes!!! That’s I guess going to suffice but you’ve given me hope on at least being able to transition or give LPV a shot. BTW, that thread doesn’t sound bad, and considering how popular GI is cough that would be helpful for others. Also, what are your impressions of DFAO so far? From what I’m hearing, it sounds like its a great addition to LPV. But thanks for the heads up on disabling anything that is translucent. Another thing is translucent shading really needs an upgrade and is pretty limiting. I know it’s been on the Roadmap for a while but I really hope Epic is considering getting in improvements or other alternatives/shading models for translucent materials. Right now, I have a big problem with ordering with translucent materials even down to the polys on a StaticMesh. I’ve tuned the materials for the time being so you don’t notice the issues but hopefully improvements are in the pipe sometime soon. Thanks for the reply bud
Thanks for the tip Jacky on making the translucent materials have a little more depth! BTW, I saw your thread a while ago and that’s a really Challenger with some pretty killer materials, great job! What’s the link to inisides’s thread that you mentioned? That’s a interesting switch off between using LPV for the outside and going with PP Volumes for the inside. If the console command hopefully doesn’t provide a jarring feeling when transition from LPV to PP Volumes, that could be sufficient. Great idea!