You can increase the LPV size over 20000 as I tested, it will work and only cost few frames drop(about 1-5fps drop in real time when I set LPV size to 50000). I am currently testing an outdoor scene which is build with modular pieces. From the result I tested, LPV works best. I get stable 115-120fps @ 1680x1050 with single GTX Titan Black. With DFGI on, I got about 60fps and with VXGI on, I got 80fps. I am using a Nvidia’s GTX Titan black so I think if you are on a Titan X or GTX 980ti, you probably get better frame rate when using VXGI since Maxwell cards utilize better than previous generations. However, I cannot really afford VXGI because I haven’t finish vegetation on landscape so I am going with LPV GI. Surely, LPV suffers the light leak, however, I separate my building into modular pieces and they all have thickness, so I don’t see light leaks. BTW, I am using Nvidia’s VXGI branch, with LPV on, SSAO off and HBAO+ on, the result is pretty good and no performance impact which makes me really happy.
The issue with that is that the larger your LPV range the lower the quality is, because it’s a fixed amount of quality, so you can have a small scene with higher quality dynamic GI, or a large scene with low quality dynamic GI. At a certain point your level can be too large for LPV since you just don’t get enough quality with it at the range you have to set.
Precisely. I’m currently using LPVs up to a size of 40’000, which is still kind of ok, quality-wise. But it’s not always large enough to cover the entire level and increasing the size further completely destroys the quality. Even a size of 100’000 is still good enough for the distance though, which is why cascades would be a great solution.
I’ve messed around with this a lot but spotlight support doesn’t actually appear to work at all. Something I’m missing?
It’s in there but still requires the Directional light to be enabled for the scene. When I tested it I lowered the intensity of the directional light to near nothing and increased the spot light intensity to quickly check it out. I’ve not used it a lot though. Was able to see the interaction in the LPV visualization mode as well.
Hmm, I definitely tried that. Directional light and emissive LPV was working, but no spots. Will try again, I guess.
My bigger problem with LPV right now is in 4.10 (haven’t tried 4.11) it doesn’t work on PS4 at all for me. Build crashes on launch even just having it enabled in consolevariables.
For primarily outdoor scenes, I’ve found LPV to be more than capable. The occlusion features that were added in 4.9 kind of negate the need for DFAO - it is far less detailed (depending on volume size) but there are actually fewer artifacts with LPV occlusion. To date, I haven’t had a problem running DFAO and LPV at the same time, but then I don’t see the point in doing so if the goal is to keep things fast.
Again, for large outdoor levels, there is no other realistic lighting option. Heightfield GI seems to be missing in action, and I’ve seen nothing to suggest VXGI or SVOGI would be better in this capacity even if they were performant…
VXGI is basically amazing if it were performant - muuuuuuuuch better than LPV.
Could you explain how you prevent the problems illustrated in my post above for somewhat large outdoor levels? I’ve literally been trying for years now to come up with a decent solution for the limited LPV size/quality, but so far failed miserably. There doesn’t appear to be a way to cover a large area without reducing the quality to a very unappealing level at the same time. And the level I’m talking about isn’t even that big, about 60000x60000 Unreal units.
Hey guys,
Anyone got an idea of what a LPV Blendable Volume is supposed to do? I can guess but how do you set things up? Docs are sketchy at best.
I do experience this behavior, but I didn’t realize it was so dramatic until I set up my own test scene with default parameters. When I first started experimenting with LPV (long ago), the very first thing I did was to turn up Light Injection Bias to ~3 in order to prevent light bleeding with thin meshes; I’ve stuck with that ever since. It cuts down dramatically on glancing light bounces, and in turn I never had such an extreme brightening effect within the LPV radius to begin with. Also, in terms of it being noticeable, it helps that I have the LPV radius / volume size in sync with where cascaded shadows end (and DF shadows begin).
I have to agree, the implementation of cascades would be brilliant. 
I don’t know why people still stick to LPV, it’s a extremely outdated technology, it has bad performance, bad quality and lots of issues.
Cryengine dropped LPV years ago becasue of the limitations and bad performnace. Since Ryse son of rome, they are using a probe based Lighting/GI solution. Not talking about SVOTI which is a kind of new Lighting/GO system in Cryengine.
For me personal VXGI is so far the best looking solution when it comes to Dynamic GI in UE4, but they really have to improve the performnace…
It’s Nvidia, they won’t. Crytek’s SVOTI is the same Sparse Voxel Octree stuff Epic already abandoned, but much, much faster. I’d guess some multiple importance sampling type stuff? Dunno. Regardless there’s some massive improvements in realtime many lights recently, especially from Square Enix. Would just have to hope Epic gets back to realtime GI at some point, just DFAO isn’t nearly enough, or fast enough at the moment, to rely on.
Cryengines SVOTI is amazing but i would be even happy with a light probe based Lighting/GI System like in Cryengine or Snowdrop. It works on PC’s and conosles and the result is still very nice looking. I mean games like Ryse Son of rome or The Division are looking fantastic. It’s sad to see that Cryengine has two good ligting/gi system, SVOTI and Probebased Lighting/GI and we even didn’t have one really working alternative in UE4.
Hopefully epic will continue the work on figuring something out…
This for sure. Half the stunning looks of CryEngine SVOTI are coming from it too. Without the probes it still looks good but not nearly as impressive imo.
Thanks, that’s good to know. It seems unlikely that Epic would add a major new feature to LPVs, but I’ll just have to keep bugging them about it!
What else could I stick to? VXGI appears to be the only real alternative in UE4, and from what I’m being told performance is worse than LPV and it doesn’t work on consoles. So it’s not even an option for most games. Using only a sky light for dynamic outdoor illumination certainly can’t be the answer either.
I see it the same way like you… with only one difference, LPV is no alternative for me. Working on projects with UE4, i’m always using Lightmass, cause there is no other way for good looks and good performnace.
Thats what i wrote like 1000 times before it other post, i see Dyanmic light plus sky light also as no laternative for me, cause the quality in a scene is not therer without GI.
I mean now days, since like lamost every engine has opened his source codes, maybe someone will figure out to export/import features between the engines :rolleyes:
p
Of course baking always looks better and has better performance than dynamic lighting. But, well, it’s not dynamic then. It’s not a solution for games with drastically changing lighting conditions.
Open world games can look amazing with dynamic lighting if you have a working GI solution, have a look at The Division, Ghost Recon Wildlands, Ryse Son of Rome, Kingdome Come and many more… this games are looking amazing.
They are a lot of Archviz scenes with lightmass, they also look amzing, the problem is… lightmass for large open world games, very difficult if not inpossible… especially when you like to have day and night cycles.
You’re completely right, but that’s kind of the point I was trying to make. Lightmass just doesn’t work for every type of game, so what choice is there apart from using LPVs?
I guess you could argue that UE4 is simply the wrong engine for making an open world game, but I don’t think it has to be. LPVs may not be the last word for real-time GI, but they are definitely viable. Or at least they would be with cascades! 