VrayFur extreme rendertime

[fAEkE]

We don't have time

[fAEkE]

Ah, the project is finally rendering so I have some time to test fur again.

I have covered the car floor with vray fur. The scene is lit with about 10 arealights, one direct light and a hdri dome.
IM settings are set to Medium -> Custom -4/-2, 40/25. Light cache 500 subdivs and "Use light cache for glossy rays". DMC is set to 0,9/0,01.

Lights have 20-30 subdivs. Resolution is A4 400DPI.

Without fur, lightcache calculates in about 1 second and IM in about 2-3 seconds for the region I'm testrendering. With fur, it takes forever for light cache onl, estimated time 3 hours. Is there some way I can lower the rendertime? Is there something wrong maybe, or do fur take this long to render? I have dymanic memory set to 6000 btw.

[super gnu]

hmm thats quite a demanding scene you are describing.. 10 area lights are quite a killer.. also iirc, lightcache isnt the best choice when using lots of dynamic geometry (fur/displacement/proxies) since it needs to have the whole lot in ram at once. you could try very small bucket size and BF secondaries.. might go faster.. whats your ram look like in the task manager? maxed out? cpu at 100%?

you could also look at the "vraypattern" plugin sounds ideal for stuff like this.

[fAEkE]

Yes, but only 2-3 lights are for the main lightsetup, the rest are for detail lighting.

Ram is not an issue, got 24gig and the scene uses 8-10 when rendering. The cpu tho.. looks like only about half of my 24 cores are working, and the total cpu usage when calculating light cache is 3-4 percent....

I'll try BF, thanks

[super gnu]

ok so id try a) incresing your dyn. memory limit... it sounds like you are capping ram usage at 8-10 and the cpu is starved of data.. also, what have you got your number of passes set to in the lightcache.. should be on 24 to match your cores..

[fAEkE]

24 passes yes. I tried 12gig dynamic memory now and no difference in rendertime

[super gnu]

hmm not really ajn idea then.. if youve got the passes set right, and increasing the dyn. mem limit doesnt help, im not sure what could be stopping the cpus going at full whack..

when you hit render, looking at the ram graph in the task manager, does it show a gentle upward curve, and smoothly level off, or does it reach a point and then go flat like the top was chopped off the curve?

also you might find that its just taking a while to fill the ram (never understood why) before cracking on.. you might find after 5 mins yer cpu shots up to 100% and yer estimated time drops like a stone.)

orr. it could be a many cores issue... maybe its doing something single threaded that needs to finish before it can do other stuff...

[fAEkE]

Looks like it takes a while for something to be loaded into the memory as you said. Light cache begins with about 4-8 hours estimated time, but it took only 5 minutes to calculate. The Memory curve looks like a staircase. Begins at 8 gig and ends at 9.

It still takes long to render imo. Low settings for both LC and IM. But I guess fur is demanding

[super gnu]

yep it certainly is. theres another current thread talking about displacement not using all cores.. vlado mentioned that the actual geometry generation for displacement is only single threaded at the mo. probably its the same for fur, and youre waiting for all them little hairs to be created...

one thing that might speed things up a little... how many knots do you have the hairs set to? if your just doing a carpet you could get away with just a couple i guess.. that should reduce the mesh density quite a lot.

[fAEkE]

Ye I've read it aswell. If fur ain't multithreaded, I'm not surpriced that it takes time to render it

I've thought about that. Got 4 knots atm, and I think thats a good number.

[Deflaminis]

I think I have found the problem, already. But it rises another question.

The problem seems to be that the mesh has a couple of small polys that will get a very large amount of hairs campared to the other faces. Is there a way to fix this without modelling the carpet? The mesh is cad as I mentioned, and some faces are very large while others are very small.. I guessed that "per area distribution" would fix this but I was wrong.

If it's uv mapped can't you use a density map instead of applying it to faces?

[fAEkE]

I thought about that, but it would be very difficult to get the same amount of hair all over the carpet. I need to know how many percent more hair the small areas have compared to the large.. well you get what I mean, almost impossible. But if you keep reading the thread you see that I solved the issue. Thanks for the tip tho!