I'm using the following code:

   radiosity
   {  recursion_limit 1
      brightness 0.5
      count 100
      #if (Rad < 2)        //pass 1
         error_bound 0.5
         pretrace_end 0.01
         save_file
      #else                //pass 2
         pretrace_start 1
         pretrace_end 1
         always_sample no
         load_file
      #end
         "unfinished.rad"
   }

The first pass took 494 seconds, and the second took 1355 seconds.  Both passes
had the exact same settings: no anti-aliasing, no focal blur, no area light, no
micronormal, and the same image dimensions.

It appears to take nearly 3 times as long to reuse the data than to compute it
in the first place!  This more than defeats the purpose of 2-pass radiosity.
Is there another setting that I've forgotten that will keep my final render to
under 5 days?

Post a reply to this message

"Cousin Ricky" <ric### [at] yahoocom> wrote:
> It appears to take nearly 3 times as long to reuse the data than to compute it
> in the first place!  This more than defeats the purpose of 2-pass radiosity.
> Is there another setting that I've forgotten that will keep my final render to
> under 5 days?

I've had problems like that before.  In my understandment, error_bound in the
second pass has to be close to the original, just slightly higher.  It gives
differences in the performance, I don't know, perhaps the data lookup method
depend on the values being close.  The values can't really be the same because
it causes some black splotches, in my experience.

I generally put error_bound in the second pass to be at most 2x the original.

Post a reply to this message

> I'm using the following code:
> 
>    radiosity
>    {  recursion_limit 1
>       brightness 0.5
>       count 100
>       #if (Rad < 2)        //pass 1
>          error_bound 0.5
>          pretrace_end 0.01
>          save_file
>       #else                //pass 2
>          pretrace_start 1
>          pretrace_end 1
>          always_sample no
>          load_file
>       #end
>          "unfinished.rad"
>    }

It's the error_bound: it has to be the same for max speed. Try that:

     radiosity
     {  recursion_limit 1
        brightness 0.5
        error_bound 0.5
        #if (Rad < 2)        //pass 1
           count 100
           pretrace_end 0.01
           save_file
        #else                //pass 2
           count 1
           pretrace_start 1
           pretrace_end 1
           always_sample no
           load_file
        #end
           "unfinished.rad"
     }

The "count 1" might give some splotches, but it's good for tests: use the 
same value as the first pass if you want more accurate results.

Also, remember to increase default max_trace_level: I found the default of 5 
can cause some artifacts with radiosity if there are reflective surfaces 
(mostly on the corners between reflective surfaces).

Regards,

--
Jaime

Post a reply to this message

"Cousin Ricky" <ric### [at] yahoocom> wrote:

>
> It appears to take nearly 3 times as long to reuse the data than to compute it
> in the first place!  This more than defeats the purpose of 2-pass radiosity.

Yes, this is an irritating situation. The entire save rad/load rad idea probably
deserves an in-depth discussion. It's interesting that, AFAIK, saved PHOTON data
*completely* overrides its settings (and loads very quickly), which is what the
concept of saving/loading is supposed to be. My own experiments bear this out.
Whereas saved rad data still uses *some* settings--and unless they are tweaked,
the 2nd pass takes just as long (or longer.)

I'm still trying to understand how some of the aforementioned rad tweaks came
about. (From time to time, I've run across them is others' scene files as
well.) They kind of seem like 'folk wisdom'. For example, using pretrace_start
1 and pretrace_end 1.

It would be quite useful to many of us if the reasons behind these tweaks could
be explained. Perhaps they have been, somewhere, but I've had no luck in
tracking them down. It sure would save each of us lots of duplicated
experimentation/testing time. (Not that experiments are a waste of time, of
course; but loading rad data seems like a black art!)

Ken W.

Post a reply to this message