[3dsmax 2016 & Vray 3.6]
Two renders of the same setup - I might have slightly adjusted the settings trying to resolve the missing sections of the beams of light. One render was all on a local machine while the other was via Backburner/DR. Seems like region renders, or rendering at different resolutions will produce different artifacts.
Setup: VrayEnvironmentFog, 18’ high and 10’ distance. Disc vray lights.
Thoughts on fixing this?
Four things come to mind:
1) Lightcache. if it’s used, results will be different between region and full screen (just a different region, eh.), or at radically different resolutions. You may need to increase the LC subdivs a bit to compensate.
2) Adaptive lights and light cutoffs: as ALs depend on the LC, the above applies, possibly.
3) Intensities: brighten the lights and see if the issues are solved that way.
4) Min AA: try 2 or 3 subdivs, see if those find enough light in the volume to warrant further sampling towards max AA.
Poking in the dark though, so take with a pinch of salt.
Thanks @lele -- These are all super helpful suggestions.
Doubling the lightcache subdivs already looks to have resolved it. Going to do a full size render to confirm.
I had already tried noodling the adaptive lights without success. Both changing it to full evaluation and doubling the adaptive number. Also tried changing the lights cutoff to 0 from .001, didn’t help.
The min AA might be what I go after next if I’m still seeing issues.
Ended up being a combo of doubling LC and enabling full light calculations - which I might not have needed to do if I deleted all the extra lights not being used in the VrayEnvironmentFog. But it worked!
Could you kindly send me a cutdown scene, maybe just the lights and fog setup, along with simple untextured geo?
I’d like to run a few tests for tech which isn’t yet released (but may come for SP2), which could well sidestep the issues you had. 
@lele got an email I send you a link to a zipped up scene? or send to support?
emanuele dot lecchi at the chaosgroup domain.
Sharing a bit of good news publicly: with SP2 the issue with missing buckets will be fixable with just a min AA of 2 subdivs, and the Adaptive Lights sampler, thanks to a few new and improved techs (largely transparent to the user too. Defaults Magic!).
For a scene with these many lights, the performance difference is quite substantial when compared to full light eval, nearing one order of magnitude.
For tiny lights going entirely amiss, but not generating artifacts, raising the LC subdivs and lowering its sample size will still be the way to go (again, no need to change light sampler to full eval!).
Below: the first is a a full light eval with 2-32, 0.25, and LC at default, rendering for 11 minutes.
The second is a 2-32,0.05, LC with 3k subdivs, and sample size at 0.005, rendering for 9 minutes.
Is this the way to go then for scenes with environment fog?
Well, it’s nothing strikingly new: when there are sampling issues, with very fine details not being picked up, raising minimum samples in the AA helps, as does making the LC samples smaller in screen-space size, and oversampling it.
In this scene some of the lights were well below the pixel in size, and behind plenty of fairly dense fog, so difficult to pick up with just one sample per pixel.
If the base sampling finds nothing, the adaptive routines won’t kick in, and the pixel will come out empty.
What’s good in SP2 is the new light sampler for lights inside volumes (still only for GPU, but with ample chance to be ported at some point, likely for SP2. Notice the tentative sentence construction, here.), and the improved LC which will work heaps better for animation and varying resolutions, without the kind of issues we have today in those situations (and with more features to boot, but we’ll leave those for a better time.).
With the new (GPU) sampler, there is even no more need to raise min AA or change the LC settings, as things will just work in these kind of situations (lights in volumes).
I didn’t know this was a thing.
Is there a way to make the light cache kinda automatic? So it does change the subdivision based on resolution etc?
