plates vs mesh

[mikeh]

here are two exact same trees rendered in the exact same environment. one is made up of plate leaves with a leaf image mapped to them that has opacity. the other is a tree with the leaves made up of a mesh that has the leaf image mapped to it...no opacity. There is a pretty large difference in render times. The plate version has the map in the opacity slot. I tried popping the opcity map in the refraction slot but i could never get the leaves opacity portion to work. If there is a secret to that, please let me know.
Thanks.
mh

plates...3:37 min
mesh.....1:19 min

http://xs.to/xs.php?h=xs225&d=08141&...objects696.jpg

http://xs.to/xs.php?h=xs225&d=08141&...-plates823.jpg

[MoonDoggie]

What happens if you turn off filtering (on the opacity map) in the plates (opacity) version and turn the blur to .01? I'd be interested seeing your results, because others have reported faster (albeit not that much faster) results when doing this.

[mikeh]

big difference...any other speed tips?
mh

render time 1:00

http://xs.to/xs.php?h=xs125&d=08141&...ter-off594.jpg

[joconnell]

That's the main one really - the issue comes up where vray hits the anti aliased edge of an opacity map - eg a grey value. A White value for the leaf is totally opaque so vray bounces back all light rays happily. A black value is totally transparent so vray shoots right through it. A grey value is nasty since you need to do both bounced rays and refracted rays and if those rays go on to hit another leaf with a grey opacity value you end up with a tonne of rays being bounced. Knocking off the filtering and using quite harsh opacity maps is your best bet.

[RErender]

I wonder, how do the tests scale? If you have 20 trees in each test what would the times do?

[MoonDoggie]

That's a remarkable difference. Let's see some more tests.

[mikeh]

i just did a test with ten trees for each. The difference between the trees is that the plates trees come from Onyxtrees standalone software and are exported out as 3ds files and then imported into max. These cloned with no problem, very quickly.

The mesh trees are treestorm trees and still associated with treestorm. (the leaves are mesh sub-objects)They cloned slower and crashed max after 6 or 7 trees were cloned so i had to start over cloning them and save after every 2 clones.

Also, both types of trees are diplayed as boxes. When the 3ds trees were cloned, they held their display properties. The treestorm trees didn't. I was required to stop after each tree clone and redo the object properties. I eventually discovered that every time i cloned a treestorm tree, it added a set of leaves. (ie original is one leaves and one trunk, clone is one trunk, 2 leaves, clone of that is one trunk, 3 leaves and so on) i had to go back in and delete all the extra leaves from each grouping.

Once i did that i rendered and compared. here's the results.
10 plate trees...1:39 min
10 mesh trees...1:07 min
mh

[MoonDoggie]

Good stuff mikeh, thanks for the updates and the tests. -Colin

[werticus]

My hero! , just saved me a few hours render time

good work team! :P

What happens if you turn off filtering (on the opacity map) in the plates (opacity) version and turn the blur to .01? I'd be interested seeing your results, because others have reported faster (albeit not that much faster) results when doing this.

[Aldaryn]

Funny, but it even helps to have a "crisp" opacity map without greys. I don't know what's the big difference, since Vray has to sample the map anyway, but sorta works.

Best regards,
A.

[Jonferimonic]

Damn if only id known this a few weeks back. Ive just retested a really slow old scene of mine with a few trees in it, turned off filtering in the opacity map and dropped the blur value and a low res test has halved in rendering time. The leaves are much sharper too. Thanks for the tip.

[ene.xis]

if ur turning the bitmap filtering off it doesn t matter wich value u set for blur, as its not being filtered anyway....right?

[Aldaryn]

Yup. (Or at least it's shouldn't matter... )

Best regards,
A.

[ene.xis]

thought so

[mikeh]

more samples. here is 100 trees of the two i have been using. one group is mesh the other are plate. The mesh group has been disassociated frpm treestorm (a very tricky thing to do without crashing max and corrupting files). exact same camera angles and environment.

mesh render time is 4:12 minutes
plates render time is 5:15 minutes.

It just confirms my dumping opacity plates and using mesh ...contrary to what other people are doing here on the forum. mesh, as always, renders faster than plates unless you are using one tree...and that never occurs on my projects.
mh

http://xs.to/xs.php?h=xs126&d=08142&...00-tres857.jpg


http://xs.to/xs.php?h=xs226&d=08142&...00-tres521.jpg