Hi
Is there any way to avoid generating all the volumetric lightmap data that appears well above the lightmass importance volume when using a landscape based map?
Aside from generating samples I wouldn’t expect the camera to visit during gameplay, I suspect processing these is contributing to the 4 hours light building time I experience in 4.18, compared to a 20 minute light build time in 4.17.
Screen grabs below will hopefully illustrate the problem.
Thanks for any tips,
Gavin
Does anyone have any thoughts on this?
Hi, this is exactly my problem too.
Basically there are is too much unneeded volumetric lightmap data generated from my custom scaled light mass importance volume.
It is both a waste of memory, and a waste of time building light for this.
Have you manage to solve it yet?
Technically I cannot suggest anything. However, what you could do is do a full test with the volumetric lightmap in place and then screen capture the settings to reuse at a later date. Is it important that you keep the volumetric lightmap in at this stage of development?
Does your lightmap have to be that size? What do you lose with it being smaller on the vertical axis?
No, not solved yet.
My understanding is the developer who implemented this technology suggested it should be possible to change the engine, so the top level of probes were only generated where needed (say within the lightmass importance volume), then only sub-divide those to create all the lower sub-divisions. I don’t believe there’s been any further consideration for implementing this in the engine at this time. There needs to be more people with this issue for it to become a priority.
The intention would be for the data only to be collected for within the Lightmass Importance Volume. However, there is no setting to change the vertical axis as you suggest. The method as implemented captures a cube shape of probes, which is fine in small levels, or test maps, but isn’t great for real cases, such as open world maps.
Hi Gavin,
I have the same problem as well. After a few hours of online search, I finally found out why volumetric lightmaps won’t work for my scene…
“Does not currently support Level Streaming” (from Unreal Documentation - Volumetric Lightmaps)
Maybe this concerns you as well?
No, not using Level Streaming, so nothing to do with that.
Thanks.
I believe the engine just isn’t considering optimising the first level of sample points, to those closest to a surface or volume. I believe this should be possible, as a post elsewhere (by the developer) suggests such could be done - on the first of samples only. I guess there needs to be enough people needing the feature for it to be considered important enough to fix.
According to this - Volumetric Lightmap Samples bug (Reproduce-able) - Rendering - Unreal Engine Forums - it’s expected behaviour, although the VLM samples could be better fitted in the lightmass importance volume. Also top level samples, that seem to populate most of the wasted space, are not supposed to have a big impact on performance.
“are not supposed to have a big impact on performance.”, yes runtime performance maybe, but baking is a disaster, as indicated in my question above.
This is an issue the developer (as you found) indicates an improvement could be made (by limiting the top level samples to near surfaces), but unless there’s enough demand, likely won’t become a priority.
