pathtracing vs pmc, any suggestions?

Mon Sep 12, 2011 8:21 am

Hello everyone,

This is not another thread trying to flame any of these techniques, i just want to note that it's been happening in every render. pmc is said to converge faster than PT, but i never found that...

my latest sample is this scene, the noise is far more in pmc than PT, both images rendered @16000 samples, each took the same time of 4.5Hours, also both scaled down from 1440 to 1024.

My question is, does everyone else have this "problem"? I've also used portals in the scene (yes, correctly set up) and 2.51. Does anyone have something to suggest in the settings of pmc to make it more efficient?

Thank you, and all i have to say to the team, it's an amazing release, thank you guys!

Mon Sep 12, 2011 9:03 am

vagos21 wrote: pmc is said to converge faster than PT, but i never found that...

AFAIK this is true for scenes with lots of reflected & refracted caustics not for every scene. PT is still very much usable in the scenes like yours above, that's why it was kept around.

necko77 · Mon Sep 12, 2011 9:34 am

in my test i didnt find PMC faster then PT . . . PMC probably needs optimization cos radiance wrote somewhere on forum that PMC should be 3-4 time faster then PT...

I didnt find portals usufull as i think they should be . . .

Only thing that will make more efficiant PT, PMC, DL is is CUDA cores

Thats only thing that matters...
no matter what radiance and octane team
do with their development i know that i will improve Octane, adding more CUDA cores

We buy octane at very fair price, now you must spent your money on CUDA cores and thats the place where all magic happens

Investing in more CUDA cores will imporve your rendering speed in 2x, 3x, or 10 x
I wish that can be done with software but i think thats not going to happend

Mon Sep 12, 2011 12:34 pm

Why did you not find portals useful. In my tests it has improved the speed 2x to 3x.

necko77 · Mon Sep 12, 2011 1:52 pm

this is why...i never get that speed up or im not doing something right . . .
but if i change this 460 with 590 card, i will get 3x speedup

quick test,

Hey, can you show me example of that . . .can you show me that 2-3x speed up plz

without portal :

with portal : (even slower

)

blender file with normals

necko77 · Mon Sep 12, 2011 2:14 pm

after 500 semples

with portals

without portals

alpha shadows enabled !

same time as with portals but model is without portals (cos its faster and in same time it would render more samples)

If i take a closer look at these images, i can say that image with portals wins cos its cleaner
so probably,tahts our speed up, scence is optimized more and interiors gets cleaner faster !

Mon Sep 12, 2011 2:15 pm

Make sure you have alphashadows enabled in the render kernel. This will disable the light portal even if you the portal material enabled. See if this helps. The kernel should be pathtracing or pmc if it's not.

Mon Sep 12, 2011 2:22 pm

In my tests, there is no remarkable speed difference with and without portals.
I found useful input into my glass windows diffuse material with transparency and blackbody emitter.

Mon Sep 12, 2011 2:31 pm

The more polygons you have for the portals the slower the engine becomes. So if have multiple windows in an even row than use a single polygon to cover all windows.

Tue Sep 13, 2011 9:06 pm

Hi Vagos!!

How did you get make a image with a little bit of noise, with only 16000 samples?

I´ve made this image rendered with 64000 samples, spending about 14 hours of render and my image stills with noise, specially in mirror.

I rendered with PMC Kernell.

Is there one trick to do this? I´ve tried what radiance posted to render 4x bigger and decrease image.. but honestly, I don´t see too many differences.

My image is below