Hi,
I was wondering if there is any rule of thumb on what settings to chose for the BSP tree.
Maybe based on render resolution, scene face count, object count, etc.
It works pretty well regardless of those quantities. The only things you might want change are the max. depth and the faces/level coefficient and they have a direct influence on the render times vs memory usage.
If you are using Embree, it ignores these settings and is fully automatic.
Best regards,
Vlado
Okay, cool. Didn’t know it doesnt influence anything when using embree. On a side note - is there ANY argument against using Embree apart from not having intel CPUs?
Embree is a bit lower precision so very rarely, there are artifacts. There is also a slight instability in the official builds if for some strange reason an invalid ray direction is generated. Usually, none of these is a problem though. The second issue is fixed for the SP, where we have also extended Embree to handle V-Ray proxies and instances (and hopefully hair, but this is not certain yet as that part of Embree is still under active development).
Best regards,
Vlado
Interesting. Presume this means that in the near(ish) future we should see some further speed boosts using Embree and proxies and instances? I use it on all projects currently - never found a reason not to.
Any estimate on the SP 
I’m eagerly waiting for GGX :o
This is the case, indeed
Best regards,
Vlado
It’s coming…
Best regards,
Vlado
Wow okay, my scene is pretty much ONLY VRayProxies at the moment.
Does that mean that basically the Embree doesn’t do anything? 
Definitely nice to have it work with VRayProxies!
Unfortunately in the official builds, Embree is not used for proxies yet.
Best regards,
Vlado
As I didn’t know this I did some tests.
Number 1: 2m56,6s : Array of a VrayProxy (=1 teapot) + EMBREE On
Number 2: 1m31,0s: Array of a Teapot mesh + EMBREE OFF
Number 3: 0m56,8s: Array of a Teapot mesh + EMBREE ON
BF + LC
AA 1-6 and 0.02 noise threshold
Min shading rate at 16
BF, materials, VraySun at 8 sbdvs
DMC multiplier at 0 and noise threshold at 0.015
Number 2 and 3 are as expected. But I expexted number 1 (with proxies) to be between 2 and 3 not twice as long. In the beginning of the LC it took quite some time to show more or less image. Maybe a nice thing to add at the end of a render is some statistics. How long the precalculations took, the LC, the BF, total. What settings. Some other usefull stuff, with some graphs. How many pixels needed the highest amount of samples etc. And this could be compared to other renders of the history.
So do we conclude from your test that, currently, if you are using proxies you should definitely turn OFF Embree? Looks like it…
The only test you didn’t run was Number 1 but with EMBREE Off. It would be good to compare that against the EMBREE On.
No, it only means that you should not use proxies just for instancing Regular instances are definitely faster than proxies. Proxies are only meaningful when you have a lot of geometry in your scene and you are looking for ways to reduce memory usage.
Best regards,
Vlado