>> Thanks for these additional comments Alain. However, before I read them I
>> already ran a case at 20 samples and 20 iterations, with the output
>> below (and
>> also rotating the cube a bit so the light isn't shining in the
>> camera's eyes).
>>
>> http://yfrog.com/2mcode20int20samplesp
>>
>> It took a little over 4 hours to render on my desktop (ran it
>> overnight). I can
>> really see how those doing much, much more complicated scenes run into
>> ridiculous rendering times.
>>
>> Alain<aze### [at] qwertyorg> wrote:

>>>> I figured out my issues, which were numerous. Here is my current
>>>> result:
>>>>
>>>> http://yfrog.com/fvcode5int5samplesp
>>>>
>>>> Took 22 minutes to render on my computer. I'm steping up the
>>>> interval and
>>>> sampling rate to make a better image though. This program is
>>>> mind-blowingly fun
>>>> :)
>>>>
>>>> Thanks for all the help. No more questions in this thread :)
>>>>
>>>>
>>>
>>> Don't increase the intervals value as it will kill your render times.
>>> Only increase samples.
>>>
>>> samples 100 will render faster than intervals 10 samples 5 for twice the
>>> total samples.
>>>
>>> While you can use samples 10, 100, only the "10" will be used ans the
>>> "100" will be totaly ignored.
>>>
>>> The defaults value from the documentation are for smpling method 1. The
>>> default sampling method is method 3. It's defaults are:
>>> intervals 1 // NEVER change that!
>>> samples 10 // MUST be 3 or larger.
>>>
>>>
>>> Alain
>>
>>
>>
>>
> I notted an anomaly in your scene:
> The spots are set into 6 sets of 9, for a total of 56. The problem is
> that you have 84 cylinders, and looking further, you have LOTS of
> duplicates like: <0,0,10> <-0,0,10> <0,-0,10> and <-0,-0,10> with many
> apearing in 2 or even 3 colours sets. (manualy changing very small
> values to zeros helped a lot to spot those)
> Those duplicates increase your render time and cause coincident surfaces
> artefacts each time that you have more than 1 colour at the same place.
>
>
>
> Alain
After removing 18 duplicates (6 for each colours, only for the 
orthogonals ones and not checking between colours sets), I got an 
impressive boost in my render time. Down to 66 elements down from the 
original 84. Sould be further reduced to 54.
Still have 3 for <0,0,10> and <0,0,-10> that have all 3 colours. 
Resolving those removes 4 more elements-> down to 62.

It should be noted that intersection and difference are inherently poor 
at bounding: Any ray that encounter those need to be tested against all 
components. This also apply to the shadow rays. So the less components 
you have, the faster the shadow rays can be evaluated.

I'm thinking of another aproach:
Have only a plain sphere and place a collection of tight spot_light in 
it's center. Add double_illuminate to make the spots visible on the 
surface. Add no_shadow so that the beams can continue outside of the sphere.
You use the points that you have presently as the point_at for the spots.
Now, any duplicate will be obvious due to the increased lighting.



Alain

Post a reply to this message