Svogi

General Feedback & Requests
81

When, could mean 1,2…3 years. And you wouldn’t know if it would support mid-high spec platforms, what we know is that simpler existing techniques already require high-end.

If they had released SVOGI from the start and kept on optimizing it + the time it takes to release a high-end pc game, hardware wouldn’t be a problem.

82

UE4 LPV

gygu_screenshot_(292).jpg
gygu_screenshot_(292).jpg933×479 51.6 KB
&stc=1

Cryengine SVOGI

6lbg_enes77-no9.jpg
6lbg_enes77-no9.jpg1000×563 246 KB
&stc=1

please bring svogi for us

83

That’s kind of an unfair pair of comparison screenshots there.

VXGI is a faster version of SVOGI, with better results and perf.

84

Though it’s not a native feature, not supported by epic, and performance is 10 times worse than SVOTI (improved SVOGI) and basically isn’t a production ready feature. :frowning:

85

I’m a noob in this topic, but can someone tell me what’s the difference between SVOGI, VXGI etc and why it is superior to LPVs?

I had decent results with LPVs.

86

And VXGI is closed source too.

87

SVOTI does large scale AO as well. I’m not sure if that’s done in SVOGI as well. But it’s like having both VXGI and VXAO working at the same time (which is not possible) with very nice performance.

88

Basic overview, as far as I understand it:

LPV (Light Propagation Volumes)
Pros:

  • Cheaper than other solutions.
  • Good for small outdoor scenes.
  • Old, but proven tech. Has shipped in a fair few games.

Cons:

  • Lots of light leaking, making it poor for indoor scenes in particular, and especially mixed scenes.
  • Works only with Directional Lights.
  • Single-Bounce (probably okay for most scenarios)
  • Very limited draw distance, octree is pretty basic.
  • Used to support emissive materials, but IIRC this has never worked in UE4 and is UE3 only (docs might be out of date, since I thought this was fixed).

VXGI (Voxel Global Illumanation - Made by NVidia)
Pros:

  • Easily the best visual quality. You can get lightmass-like results if you push it.
  • Extremely tuneable / scalable. The UE implementation is actually really nice.
  • Multibounce
  • Supports all material types and static & dynamic objects.
  • Easy to build a source version of engine with it already integrated.
  • All surfaces contribute to GI etc.
  • Takes advantage of newer hardware / drivers.
  • Lots of other nice features, like AO, reflections, specular etc - if you want them.
  • Temporal filter for artefacts.

Cons:

  • High performance cost, takes a lot of tuning to get it working nicely.
  • Windows only IIRC.
  • Cost depends on scene complexity, since entire scene is voxelized each frame.
  • Has a few rendering issues at the moment, in my case all lighting dissappears entirely after a certain distance.
  • Older hardware struggles to keep up, but again, it’s tuneable.

DFGI (Distance-Field Global Illumination) and HFGI (Height-Field Global Illumination)
Pros:

  • Would be a first-class engine feature.
  • Early results are very promising and performant (target is PS4 I believe)
  • Reuses a lot of data from mesh distance fields.

Cons:

  • Requires Distance Fields to be generated (DFGI only, has some extra memory cost)
  • AFAIK, only works with static meshes or objects that generate distance fields.
  • Currently no timeframe for it being finished, or even worked on. Epic has switched focus to rendering performance overall.
  • HFGI worked well in 4.8 but had a performance regression in 4.9.

SVOGI (Sparse Voxel Octree Global Illumnation)
Pros:

  • The demos of it looked cool as hell.
  • Would have been a first-class engine feature.

Cons:

  • Literally never going to come back, renderer has changed far too much. Wasn’t even available to public beta testers (hello).
  • Several years old now, not exactly cutting edge.

SVOTI
Pros:

  • (Probably) faster than VXGI & SVOGI, but with considerably less toys to play with.
  • Native engine implementation.
  • Results look promising.

Cons:

  • It’s CryEngine.
  • Bounce lighting can take several frames to compute, resulting in ghosting.
  • Aliasing and light-leaking.
  • Doesn’t work with everything.
  • AFAIK, ONE game has shipped with it.
  • Not as ‘dynamic’ as you’d like to think, I believe the voxels are pre-computed from static geometry. Could be wrong.
89

UE4 LPV has 1 bounce i think ]

cryengine svogi has octree 8 bounces

for games with day night cycle we need **dynamic **lighting - we cant use baked ones

please improve lpv

vxgi is very slow

also bring vxgi to ue4 as plugin if its possibe - for launcher version

90

I hope Epic has some plans for dynamic global illumination.

91

Interesting stuff:

http://docs.cryengine.com/pages/viewpage.action?pageId=25536008

92

I would like to point out that both CryEngine V and Lumberyard are using SVOGI now. It seems to be working for them, I’m not sure why it can’t in Unreal. Personally, even if it’s a bit more expensive I think it’s a better solution, at least in my case.

93

Neither are using SVOGI, and Lumberyard is basically Cryengine.

94

Yea, many confuse SVOTI with SVOGI. It would be very nice to have something like SVOTI in UE4, even if it’s limited.

95

From what I know LPV don’t work on 4.18 ? Well before the performance wasn’t good compared to other engines same implementation but still 2018 there is no solution yet in this engine for realtime GI.

By 2017-2018 Godot, Blender, Unity, Cryengine and all the AAA private companies game engines got it already :(.

By the way ty for the link and info didn’t know about that ^^

96

It works.
Maybe someone forgot to put r.LightPropagationVolume = 1 in their config again after updating engine.

97

Oh i didn’t used but the people here says don’t work plus is reported and Backlogged:

https://forums.unrealengine.com/unreal-engine/feedback-for-epic/1396203-4-17-4-18-lpv-is-completely-broken-4-19-what-s-next

98

I would say LPV is de facto deprecated at this point. If you have a feature that has core functionality regression for 3+ versions (and currently backlogged) then you should never ever develop ANYTHING under assumption it will work/build/whatever.

100

By Bruno Willian:

101

Nice, but how long is going to take to calculate offline…? You’ll be in the same boat as you are now with lightmapping. I’d imagine that it will probably take even longer to calculate than lightmaps currently do. But hey, it looks great, right?