[polynurb]

I am building a natural landscape atm, with some instanced trees and some non instanced ones (all penetrate the ground plane slightly) and i see some big impact on benchmark.. as low as 5!

i believe to remember not to intersect instances.. but was the problem
intersecting instance with instance or instance with normal geometry?

or is that issue obsolete for 1.7 and my trouble comes from somewhere else?

thnx,

p.

[Bubbaloo]

How many polygons are the trees? If you have a lot of very high poly instances, your benchmark will fall. Do you run out of RAM if you try to render the scene without instancing? If you have the RAM, it's best to not use instancing.

[polynurb]

trees are about 150k, have to use 32bit backup system atm..

i moved some of the non instanced trees away from the instanced ones (leavs intersected) and benchmark is now at ~20 .. im looking further into this...

[Bubbaloo]

Are the leaves clip mapped?

[Fernando Tella]

I think intersecting a lot of clipped planes is what mainly causes the slowdown. I did some experiments about it a while ago.

http://www.maxwellrender.com/forum/view ... ht=clipped

[polynurb]

thanks for the hints guys...

the leaves are not clipmapped but weightmapped with a ghost layer.

my nodes are still 64 bit, tried turning instances off benchmark is ALOT better now.. somwhere around 75 ram hitting 2.5gb... guess i will go this way..

[Mattia Sullini]

I think intersecting a lot of clipped planes is what mainly causes the slowdown. I did some experiments about it a while ago.

http://www.maxwellrender.com/forum/view ... ht=clipped

I just realized it a couple days ago and then completely agree. I had 5 trees (i geometry and 4 instances of it) intersecting and got a benchmark of 4.0, then shifted them and got a bencmark of 35.

[Fernando Tella]

the leaves are not clipmapped but weightmapped with a ghost layer.

I used that method too for the tests.

Good to know the other route worked nicely too.