Hi-Res Long Render Times

[jujubee]

I desperately need some advice. My scene is 2890x1870 for 11x17 print. My mat sample subs are turned down low so they shouldn't be acting up. I was getting splotchy looking disco effects on my ceiling so I turned up the hsph up to 60. I hunted around for raytrace mats and didnt come across any. I also adjusted all the light diffuse subdivs up to 150. Now my render times are incredibly painful to watch. Any ideas/suggestions as to how to go about rendering my scene and reducing render times?

Here are the settings:

[percydaman]

Is it really necessary that you use max irradiance rate of 0? To me its just too slow, your not getting ANY undersampling if I recall correctly. Try -3, -1.

[czoog]

Your QMC noise threshold seems very low (which can slow things down), try using .002 or .003 and increasing the samples to 8 or 16.

[percydaman]

oh ya I missed that. .005 is the default and thats even kinda low. What I do is use one setting for the irradiance map, then set it back to .001 for the render when using adaptive qmc for AA.

[flipside]

turn of the fixed filter for your lightcache. I noticed it can have a drastic impact on rendertimes if you don't have the right settings. In almost all of my test I get better and faster results with only the pre filter.

[R2J2]

I'm shore I remember correct, Vlado said any values below .002 won't reduce the noise anymore, you have to up the global subdivs or localy for a light.

also if you limit the raybounces to 1 or 2 that will help with render times, obviously things like glass cups ect will have dark areas.

[meanadam]

150! You are crazy man.
Sounds like a candidate for Adaptive Subdivision.

Did you try one with "overlapping" sample lookup? I never use Density on a multi-hour rendering (or animation).

Bucket size could be smaller? Do others use 128? I don't think I ever go over 64 anymore. And with your crazy light samples - that could be taking a long time for your computer to render.

[Morbid Angel]

the reason for 128 is because he has a very large render and smaller buckets will take longer time. Juju, percy is right about -3/0 irmap samples, usually -3-1 do the job.
can you show some renders? if not, watch out for the glossy mats.
you may want to try setting your qmc sampler to 0.002 and global subdivs to 2 or 4.
if you have dr awailable to you try that.
you can try setting light map to 1500 subdivs with 0.02 screen sapce and pre filter of 30, filter none.
Also check to use light cache for glossy rays, that might speed it up
thats all i can suggest for now.

[jacksc02]

Hey Jujube
how come your secondary ray bias is not at 0??
Do you have co planar faces?

[samuel_bubat]

I recently had splotch problems, too. I don´t know if this is the same
problem. I got splotches around small Splices on walls only when rendering in high quality (max rate higher than -2). The only way I got rid of them was to turn down the "Clr Treshold" to 0.1. I see yours is set to 0.3.

[flipside]

I'm shore I remember correct, Vlado said any values below .002 won't reduce the noise anymore, you have to up the global subdivs or localy for a light.

also if you limit the raybounces to 1 or 2 that will help with render times, obviously things like glass cups ect will have dark areas.

I does reduce the noise, but the visible result doesn't justify the increase in rendertime.

[jujubee]

Thanks everyone for all your suggestions. It just shows how much I don't know I also didn't realize 0 wouldn't allow for undersampling. I'll have to take all your tips and see if we can bump the times up some.

can you show some renders? if not, watch out for the glossy mats.

Heheh - lots of glossy mats here and not much I can do about it. And 8 hours later, one night's sleep, and still 1/4 the way through first prepass so nothing to show except an aborted render attempt. THanks for the other tips - will implement them.

150! You are crazy man.
Sounds like a candidate for Adaptive Subdivision

I noticed it started to develop Maxwell grain after a long while.

how come your secondary ray bias is not at 0?

I am the master of all things coplanar...

[flipside]

-3/0 is also undersampling. 0/0 that is no undersampling. Every pass uses the previous one to compute, so when you reach the last pass, a lot of info is already gathered and it will not be the same as no undersampling at all.

[percydaman]

yikes your right. Not sure what I was thinking on that one....

[meanadam]

128 is alot for the computer to bite each time. Especially if you have a lot of glossy and shadow.

Smaller buckets that render faster could be faster in the end. It would depend on the machine speed. But i am sure there is an optimized size. Would be nice to graph this as well (man, I am on a graphing kick this month). Like I said - I usually use 64 or 48.