acceptable tree renders...again

[Nickh]

Hi,

I have an animation to do around a reasonably small built up area. Currently there are 87 Vray proxy trees, each with 100,000 polys in the original mesh.

A frame at 1024x576 is taking just over an hour (from precalculated IRR + LC) AA is DMC 1/6/0.009. The scene is not finished yet and I dont think these setting are even high enough, but am already exceeding my render budget.

I have tested the demo of VrayScatter and this does not seem to have much effect on actual render times, and I dont really need the scatter capabilities for this scene anyway- just need quicker renders.

The trees do have opacity leaves, but the filtering is set to none. I need them to look as good as poss as we get pretty near the.

8 is the max subdivs in the whole scene, and it still takes over 30 mins with a global basic white material on anyway. Spec is dual quad core 2.5 with 8GB RAM.

I've trawled through the forum and cant find anything else to help.

Does anyone have any other tricks, or do these render times just sound acceptable ad there is no magic cure?

Cheers
Nick

[VISUAL_BYTES]

I would suggest that the opacity leaves are the problem. I recently did a 5 building complex with a lot of trees, just geometry though, and render times for PAL frames (720x576) did not exceed 20 min with reasonable GI settings, they were actually between 5-20min and the long times were due to semi-glossy/semi transparent surfaces.

[MikeeMax]

the render times you mention are incredible. and you have a good machine. that's strange.
thos render times would be totally not acceptable for me if i want to make an animation. 10 mins/frame is the max i allow myself. my last anim was in 1280*720 format, thousands proxies and tranlucent leaves evrywhere.it never exceeded 10-12 mins/frame.
i'm curious about your settings. could you post them ?

[Nickh]

Thanks.

I'm just going through the scene looking for any other culprits at the mo, but here's the render settings. GI is using precalculated IRR + LC.

Cheers
Nick

[Nickh]

ok there was a sneaky few high res maps in there, so have reduced them. But the biggest time saver so far has been taking the AO element out. I hadn't appreciated how much time it was adding - about 30% in this case.

Its down to about 20 mins a frame, but AA settings still need to up.

MikeeMax, To get these as noise free as poss what did you have your DMC min\max & clr threshold set to for those 1280x720 frames?

Visual_Bytes, re opacity maps on leaves, following advice on the forum have found that it doesnt make a huge difference when bitmap filter is set to none, and worth the extra time. Although still need to make a final call on that when the scene is compete.

Cheers
Nick

[sea2stars]

I'm not sure if i will help with your render times, but you have 8GB of RAM and have your Dynamic memory limit at 1000MB?

[Nickh]

Thanks sea2stars, I have been advised in the past that having it set to over 1000MB would not make any difference. I will experiment with this though.

What would you set it to with 8GB RAM?

Cheers
Nick

[sea2stars]

Heh. I'd be curious to know if there's a rule of thumb, or best practice, too.

I have 4GB and in heavily proxied scenes I'll up it to 4000MB and switch to Dynamic.

[Nickh]

Wow. That's more than halved it again! Am now down to 9 mins, although still need to bump AA up, but thats getting good now.

Would be interested in a rule of thumb too. Render times on different dynamic values are:

1000MB - 21:41
4000MB - 8:58
8000MB - 9:15

So will go with 4000MB for now, unless there is a better setting that I can get to from a rule.

Thanks a lot.

Nick

[joconnell]

If you're using the same tree model a lot too, try using regular instances instead of proxies - proxies are intended as a way to render massive amounts of geometry when you don't have the ram for it - if your machine can handle the trees as instances without proxies you'll get a speed boost too.

[Nickh]

Thanks, I'll bear that in mind and give it some tests.

It is a shame there is no quick way to convert all the proxies back to meshes, as the new SP3 import meshes seems to just put it to 0,0,0 and ignore all applied transforms.

Cheers
Nick

[beestee]

I generally start with 3 frames, each one using 40% 50% and 60% respectively and move in increments of ten percent in the faster direction until I get a slower render time. That is usually enough to satisfy me, but if you are really trying to crunch times you could increment smaller once you have found the best 10% threshold.

[3DMK]

Tree maps

Hi

Another little trick is to duplicate the tree material but without alpha channels and put it on any trees that are far enough away to not notice it.

Also try and make sure your leaf covers as much of the bitmap as possible as the more transparent areas the more it will slow down render times.

Cheers