Alain nous apporta ses lumieres en ce 2008/01/28 15:06:
> nemesis nous apporta ses lumieres en ce 2008/01/27 20:20:
>> so, any free time in your hands?  want to put that quad core to stress 
>> with
>> povray 3.7beta?  If you're, try the little scene below.  With +AM2 
>> +A0.04 +J1.0
>> 1024X768 on a P4 2.66, it just brought my PC to a crawl for the whole 
>> weekend!
>> It all began with a challenge in a brazilian forum, when some guy 
>> claimed this
>> scene:
>> http://200.213.105.41/greysilica/exemplos/cjcopojateado.png
>>
>> took him 48 hours to render with Blender internal renderer.
>>
>> Surely something was wrong, or so I thought.  Well, in the region 
>> between the
>> bottom of second row cups and the top of first row cups, render time 
>> was about
>> 11 minutes a line!  It is all about refractions, normals and fresnel
>> reflections.  40 hours total on a P4 2.66 with povray 3.6.1...
>>
>> I'm sure it's possible to optimize it with some no_image, no_shadow 
>> trickery
>> with doubles and no_reflection in the main cups, but I didn't thought 
>> of it
>> beforehand!
>>
> Just rendered here with 3.7.
> LOT faster with +am2 +a0.3 at 1360*1024 AMD Athlon TBird 1400MHz.
> Added adc_bailout 0.013.
> Render time: 1:48:15.
> Only thing, your glasses look much darker.
> 
New render with +a0.1. Render time of 8:54:55 with nothing else changed. 
Imperceptible difference.

So, what's killing you is the aa with the micro normals.
Increasing the resolution will have the benefit of reducing the aa work, 
increasing the chance that adjacent pixels difference to be under the aa treshold.

-- 
Alain
-------------------------------------------------
A clear conscience is usually the sign of a bad memory.

Post a reply to this message