Face/level coef. setting??

[dlparisi]

I'm working on a rather large scene (3 floors of a hospital lobby, xrefs, lots of proxies) and was having trouble getting consistent render times. What was happening was that different machines on the farm were getting wildly different times and amazingly the faster machines were actually rendering slower. For example, my dual quad xeons at 2.5ghz were rendering the frame in ~2:30 (min:sec) and the quad 9550 @ 2.83ghz was doing it in about 1 min. All machines were well under their available ram and were at 95-100% on the cpus.

After a LOT of troubleshooting, I boiled it down to the face/level coefficient and was able to even everything out by dropping it to ".5" rather than 1. The dual xeon went down to about 45seconds and the quad 9550 stayed about the same. I'm not sure if this is actually the problem or just solves the problem caused by something else but it works.

So is this normal that two machines would interact with the face/level coef. differently? I would have assumed that as long as the available memory was OK that they would behave the same but this doesn't seem to be the case. Also, other than trial and error, is there any way that I can start to determine an optimal (or even non detrimental) setting for the face/level value.

Using Max9x64/vray 1.5sp2

[vlado]

It might be related to memory bandwidth issues or something like that; different face/level coefficients will distribute the data in the acceleration structures a bit differently, which may cause different behaviour of the processor caches and memory access patterns.

Best regards,
Vlado

[dlparisi]

Thanks. So is there any way to estimate the proper value for this?

[vlado]

Well, values around 1.0 seem to be best; higher values slow down the rendering but don't save so much RAM; lower values take a lot of RAM but do not make things much faster. That's why it's the default on 64-bit versions of V-Ray.

Best regards,
Vlado