scaling the model decreases rendering time?

[fooprobe]

ok, this hurts my brain. it seems that if i scale down my model, i can DRAMATICALLY reduce the rendering time. anyone else seeing this? i started with a grid of 16 cubes (12" on a side), no lighting except GI skylight, white material. render time 16.5s. scaled it down by 0.001, no other changes to anything, render time 5.2s.

why would this be? is this a floating point thing? i know if the model size gets to be ridiculously too huge (x10e15 or something), the render meshes start to deteriorate, which seems to me to be related to significant digits (i.e. not enough available after the decimal). but i don't know why render time would be noticeably faster between models scaled by 10e-3. anyone?

[LeoBonilla]

Re: scaling the model decreases rendering time?

I think it has a relation, with the render mesh, try to add textures to the model, save and reipen each model with different scales.

[fooprobe]

Re: scaling the model decreases rendering time?

ok, interesting. it seems to be related to irradiance map. i used the basic v-ray wood10 material, which has reflection and therefore increased the render pass noticeably. the irradiance map pass still was distinctly faster on the scaled-down model, but the render pass appeared to be the same. 40s vs 46s

so i dropped the reflection from the material, and there's still a slight advantage on the scaled-down version, but not as dramatic. 8s vs 11s.

So then I turned off the render pass, 4.3s vs 6.8s

So then I went back to plain old white material (no image map), and no render pass:
5.3s vs 20.6s !!! white materials are slower than the wood10 image map!!!

I really don't understand this. I'll have to go take another look at how irradiance map works WRT geometry. any pointers would be very helpful!

[Alexander Petrov]

Re: scaling the model decreases rendering time?

ok, this hurts my brain. it seems that if i scale down my model, i can DRAMATICALLY reduce the rendering time. anyone else seeing this? i started with a grid of 16 cubes (12" on a side), no lighting except GI skylight, white material. render time 16.5s. scaled it down by 0.001, no other changes to anything, render time 5.2s.

why would this be? is this a floating point thing? i know if the model size gets to be ridiculously too huge (x10e15 or something), the render meshes start to deteriorate, which seems to me to be related to significant digits (i.e. not enough available after the decimal). but i don't know why render time would be noticeably faster between models scaled by 10e-3. anyone?

Нi, i have one question. Which render engines you're using? If you use IM for primary, did you reset it's cache after each test?

[fooprobe]

Re: scaling the model decreases rendering time?

IRmap primary, DMC secondary. Do you mean the check box at the bottom of the flyout for "Don't Delete" ? I kept that checked. I have two separate rhino files for the two scales.

So unchecking that box, prior to trials, yields 4.9s full scale, 3.1s at 10e-3 (wood10 no reflect, no render pass).

It's a little faster than my earlier tests, i assume because i had another app open then - it was idle. regardless, same basic difference.

[LeoBonilla]

Re: scaling the model decreases rendering time?

Render engines differ in the way they calculate light, for plain images is better to use IM/LC, and for detailed scenes (textures, and many objects) IM/DMC is better.

[fooprobe]

Re: scaling the model decreases rendering time?

Sorry for reviving this thread.

So I've finally had a chance to read through both spot3d's and F. Legrenzi's explanations of IRMap, DMC and LC engines, and can't find anything that satisfyingly explains why scaling down a model would increase (or decrease) rendering speed.

I can't figure out how to upload an image here, but interestingly the GI passes show more sample points in the full-size vs. the scaled down model, so that would seem to be the cause. Sure enough, when I zoom in, there are significant light-leaks. So at least the major issue is solved: there isn't actually an increase in render speed with the SAME render quality.

So now what i can't figure out is why there would be fewer sample points just from scaling down the model. I thought it might be the distance threshold, but modifying that alone seems to have no effect.

Any thoughts?

[dkendig]

Re: scaling the model decreases rendering time?

here's how you post images:
http://forum.asgvis.com/index.php?topic=2043.0

[fooprobe]

Re: scaling the model decreases rendering time?

thanks, sorry for being such a noob.

here they are:
http://s827.photobucket.com/albums/z...caling%20test/

[LeoBonilla]

Re: scaling the model decreases rendering time?