Hi,

I am new to try with Povray application. I am trying to test performance of
various compiler like (gcc, clang) on CPU with
scenes/advanced/benchmark/benchmark.pov (as suggested here-
http://www.povray.org/download/benchmark.php)

What I observed is the rendering is finishing in ~1.5 seconds only.
So it's difficult to compare the results with various compiler for such short
duration.

Is there any large data set which runs for considerable time? or how to change
or what parameter to change on .ini file to run it with considerable time.


Thanks in advance for your help,
Deepak

Post a reply to this message

hi,

"Deepak" <dee### [at] iiitbnet> wrote:
> I am new to try with Povray application.

enjoy.

> I am trying to test performance of
> various compiler like (gcc, clang) on CPU with
> scenes/advanced/benchmark/benchmark.pov (as suggested here-
> http://www.povray.org/download/benchmark.php)
>
> What I observed is the rendering is finishing in ~1.5 seconds only.
> So it's difficult to compare the results with various compiler for such short
> duration.
> Is there any large data set which runs for considerable time?

every POV-Ray install comes with a large number of example scenes[*], some of
which will take a fair amount of time to render; animations too are available.

also, if you "nose around", some of the newsgroups here often contain SDL
code/scenes provided by other users.

[*] typically in %INSTALLDIR%/scenes.

> or what parameter to change on .ini file to run it with considerable time.
> Thanks in advance for your help,
> Deepak


regards, jr.

Post a reply to this message

"jr" <cre### [at] gmailcom> wrote:
> hi,
>
> "Deepak" <dee### [at] iiitbnet> wrote:
> > I am new to try with Povray application.
>
> enjoy.
>
> > I am trying to test performance of
> > various compiler like (gcc, clang) on CPU with
> > scenes/advanced/benchmark/benchmark.pov (as suggested here-
> > http://www.povray.org/download/benchmark.php)
> >
> > What I observed is the rendering is finishing in ~1.5 seconds only.
> > So it's difficult to compare the results with various compiler for such short
> > duration.
> > Is there any large data set which runs for considerable time?
>
> every POV-Ray install comes with a large number of example scenes[*], some of
> which will take a fair amount of time to render; animations too are available.
>

animations are an interesting choice.  i have noticed that an average of the
time-stamps yields a different number from that reported by the program.
perhaps it is system overhead, or perhaps it is just my machines, or maybe the
numbers converge with a large enough sample.
  to do animation add at the bottom of povbench.ini

; begin
Output_To_File=On ;so you will have files with time-stamps

Initial_Frame=1
Final_Frame=100
Initial_Clock=0
Final_Clock=1

Pause_when_Done=off

Cyclic_Animation=off  ;;insures it reaches 1.0
;end

i believe without testing that later parameters overwrite earlier ones.  if not
so you will have to comment out those earlier variables.

> also, if you "nose around", some of the newsgroups here often contain SDL
> code/scenes provided by other users.
>
> [*] typically in %INSTALLDIR%/scenes.
>
> > or what parameter to change on .ini file to run it with considerable time.
> > Thanks in advance for your help,
> > Deepak

changing the ini file might be a better choice for you as opposed to animation.
first to mind is the resolution; the ini specifies a small picture, increasing
it to ten times height and width should increase rendering time a hundred fold.
something like this, again at the end of ini to overwrite original values

;begin
Width=5120
Height=5120
;end

you can also increase antialiasing (these are the default values)

;begin
Antialias_Depth=3        ;increasing takes longer
Antialias_Threshold=0.3  ;decreasing takes longer
Sampling_Method=1        ;no idea what the effect on time is
;end

however, povbench.pov might not be the best choice for you.  it has a fast parse
time and a slow render time.  parsing uses only one core; therefore slow,
many-core machines are faster than fewer-core, fast clock machines when you
compare similar cores*hertz.  i am not familiar with the standard scenes that
come with povray to advise you which scene files have long or short parse versus
render.

but now the only thing that prompted me to post; what kind of monster machine
are you running that is 100x faster than my best rendering engine (drool)?

Post a reply to this message

Le 19-06-28 à 06:13, Deepak a écrit :
> Hi,
> 
> I am new to try with Povray application. I am trying to test performance of
> various compiler like (gcc, clang) on CPU with
> scenes/advanced/benchmark/benchmark.pov (as suggested here-
> http://www.povray.org/download/benchmark.php)
> 
> What I observed is the rendering is finishing in ~1.5 seconds only.
> So it's difficult to compare the results with various compiler for such short
> duration.
> 
> Is there any large data set which runs for considerable time? or how to change
> or what parameter to change on .ini file to run it with considerable time.
> 
> 
> Thanks in advance for your help,
> Deepak
> 
> 

Some things that you can do that will increase the rendering time :
(original values from the ini file)
For the INI file :
Increase the resolution. Something like 4000 by 4000 (512 by 512)
Lower the antialias threshold to 0.01 (0.3) and Sampling_method=2 (1)
Set High_Reproducibility=on (off)

Within the scene itself :
Turn radiosity ON (this will have a large effect):
#declare use_radiosity = true; (false)
In the global_settings block:
Set max_trace_level to 255 (12)
Increase the radiosity count to 800 (80)
Set nearest_count to 20 (5)
Reduce error_bound to 0.2, or even 0.1 (0.5)
Change low_error_factor to 0.2 (0.5)
Increase recursion_limit to 3 (1)
Increase the photons quality:
Set photons{spacing 0.001} (0.007) [this will multiply the number of 
photons shot by 49]

Change the area_light:
light_source {
    <-0.7, 0.83, 0.24>*150000
    color rgb <3.43,2.87,1.95>
    #if (use_area_light=true)
       area_light 4000*x 4000*y  129,129 (4,4)
   adaptive 3 //added Will start sampling at 9 by 9
       jitter
       orient
       circular
    #end

    media_attenuation on
    media_interaction on

}

In the definition of M_Watx4, add dispersion :
    interior {
       ior 1.31
dispersion 1.01 //added
dispersion_samples 20 //added
       fade_distance 0.8
       fade_power 1001.0
       fade_color <0.02, 0.20, 0.06>
    }
Do the same for Mat_Glass :
    interior {
       ior 1.5
dispersion 1.02 //added
dispersion_samples 20 //added
       fade_distance 0.12
       fade_power 1001
       fade_color <0.6, 0.5, 0.7>
    }

Change the media used for the clouds:
samples 120 (7,7)

Post a reply to this message

Le 19-06-28 à 10:01, green a écrit :
> "jr" <cre### [at] gmailcom> wrote:
>> hi,
>>
>> "Deepak" <dee### [at] iiitbnet> wrote:
>>> I am new to try with Povray application.
>>
>> enjoy.
>>
>>> I am trying to test performance of
>>> various compiler like (gcc, clang) on CPU with
>>> scenes/advanced/benchmark/benchmark.pov (as suggested here-
>>> http://www.povray.org/download/benchmark.php)
>>>
>>> What I observed is the rendering is finishing in ~1.5 seconds only.
>>> So it's difficult to compare the results with various compiler for such short
>>> duration.
>>> Is there any large data set which runs for considerable time?
>>
>> every POV-Ray install comes with a large number of example scenes[*], some of
>> which will take a fair amount of time to render; animations too are available.
>>
> 
> animations are an interesting choice.  i have noticed that an average of the
> time-stamps yields a different number from that reported by the program.
> perhaps it is system overhead, or perhaps it is just my machines, or maybe the
> numbers converge with a large enough sample.
>    to do animation add at the bottom of povbench.ini
> 
> ; begin
> Output_To_File=On ;so you will have files with time-stamps
> 
> Initial_Frame=1
> Final_Frame=100
> Initial_Clock=0
> Final_Clock=1
> 
> Pause_when_Done=off
> 
> Cyclic_Animation=off  ;;insures it reaches 1.0
> ;end
> 
> i believe without testing that later parameters overwrite earlier ones.  if not
> so you will have to comment out those earlier variables.

If any parameter appears more than once, the last will always prevail.

> 
>> also, if you "nose around", some of the newsgroups here often contain SDL
>> code/scenes provided by other users.
>>
>> [*] typically in %INSTALLDIR%/scenes.
>>
>>> or what parameter to change on .ini file to run it with considerable time.
>>> Thanks in advance for your help,
>>> Deepak
> 
> changing the ini file might be a better choice for you as opposed to animation.
> first to mind is the resolution; the ini specifies a small picture, increasing
> it to ten times height and width should increase rendering time a hundred fold.
> something like this, again at the end of ini to overwrite original values
> 
> ;begin
> Width=5120
> Height=5120
> ;end
> 
> you can also increase antialiasing (these are the default values)
> 
> ;begin
> Antialias_Depth=3        ;increasing takes longer
> Antialias_Threshold=0.3  ;decreasing takes longer
> Sampling_Method=1        ;no idea what the effect on time is
> ;end
> 

Sampling_method have a minimal effect on the render time. Method 2 is 
slightly slower at low depth, but gives better results. The time 
increase get less important with larger depth due to it's adaptive and 
recursive nature.

Another thing that will take longer is to turn on radiosity. Only demand 
changing a single variable in the scene itself.

Post a reply to this message

On 6/28/2019 11:37 AM, Alain wrote:
> Some things that you can do that will increase the rendering time :

Subsurface sampling will also increase rendering time (a lot!!)

I encountered this this past week and was nearly heartbroken. :(


Michael

Post a reply to this message

Alain <kua### [at] videotronca> wrote:
> Le 19-06-28 à 06:13, Deepak a écrit :
> > Hi,
> >
> > I am new to try with Povray application. I am trying to test performance of
> > various compiler like (gcc, clang) on CPU with
> > scenes/advanced/benchmark/benchmark.pov (as suggested here-
> > http://www.povray.org/download/benchmark.php)
> >
> > What I observed is the rendering is finishing in ~1.5 seconds only.
> > So it's difficult to compare the results with various compiler for such short
> > duration.
> >
> > Is there any large data set which runs for considerable time? or how to change
> > or what parameter to change on .ini file to run it with considerable time.
> >
> >
> > Thanks in advance for your help,
> > Deepak
> >
> >
>
> Some things that you can do that will increase the rendering time :
> (original values from the ini file)
> For the INI file :
> Increase the resolution. Something like 4000 by 4000 (512 by 512)
> Lower the antialias threshold to 0.01 (0.3) and Sampling_method=2 (1)
> Set High_Reproducibility=on (off)
>
> Within the scene itself :
> Turn radiosity ON (this will have a large effect):
> #declare use_radiosity = true; (false)
> In the global_settings block:
> Set max_trace_level to 255 (12)
> Increase the radiosity count to 800 (80)
> Set nearest_count to 20 (5)
> Reduce error_bound to 0.2, or even 0.1 (0.5)
> Change low_error_factor to 0.2 (0.5)
> Increase recursion_limit to 3 (1)
> Increase the photons quality:
> Set photons{spacing 0.001} (0.007) [this will multiply the number of
> photons shot by 49]
>
> Change the area_light:
> light_source {
>     <-0.7, 0.83, 0.24>*150000
>     color rgb <3.43,2.87,1.95>
>     #if (use_area_light=true)
>        area_light 4000*x 4000*y  129,129 (4,4)
>    adaptive 3 //added Will start sampling at 9 by 9
>        jitter
>        orient
>        circular
>     #end
>
>     media_attenuation on
>     media_interaction on
>
> }
>
> In the definition of M_Watx4, add dispersion :
>     interior {
>        ior 1.31
> dispersion 1.01 //added
> dispersion_samples 20 //added
>        fade_distance 0.8
>        fade_power 1001.0
>        fade_color <0.02, 0.20, 0.06>
>     }
> Do the same for Mat_Glass :
>     interior {
>        ior 1.5
> dispersion 1.02 //added
> dispersion_samples 20 //added
>        fade_distance 0.12
>        fade_power 1001
>        fade_color <0.6, 0.5, 0.7>
>     }
>
> Change the media used for the clouds:
> samples 120 (7,7)


Hi Alain,

Thanks for your suggestion. I tried changing benchmark.pov and benchmark.ini
files and observed that render time is increased too much. The application is
running for hours. Is it expected?
Also let me know if any specific option to run/render in parallel on multicore
system.

I will Try your another suggestion too and update.

Thanks,
Deepak

Post a reply to this message

"Deepak" <dee### [at] iiitbnet> wrote:

> Thanks for your suggestion. I tried changing benchmark.pov and benchmark.ini
> files and observed that render time is increased too much. The application is
> running for hours. Is it expected?

Yes.  Alain was showing you things to do that will increase render time.
You can easily write scenes that would take a week to render.
Search the forums, and you will find posts where people took months or even a
year to let POV-Ray run its course.

So, now that you have a "too fast" example, and a "too slow" example, you can
start to fine tune the code to produce a scene that renders within a desired
time frame.

> Also let me know if any specific option to run/render in parallel on multicore
> system.

POV-Ray has to use a single core to parse (read and "assemble" your scene in
memory) and then should automatically use all available cores to render your
scene.

http://wiki.povray.org/content/Documentation:Tutorial_Section_5.1

"As of version 3.7, SMP systems are fully supported; by default POVWIN will use
all available processing cores when rendering."


http://wiki.povray.org/content/Reference:Tracing_Options


Symmetric MultiProcessing
Central to the many feature enhancements offered with version 3.7, POV-Ray now
supports Symmetric MultiProcessing or SMP. The command line option
Work_Threads=n or the +WTn switch allows you to specify the number of work
threads to be used while rendering a scene. On Windows systems, the default is
the number of detected cores. On Linux/Unix and OSX based systems, the default
is first based on the number of detected cores, otherwise, the number of
configured cores. If detection is not possible the default is set to 4. In all
cases the maximum value is 512.

Post a reply to this message

Le 19-07-02 à 04:48, Deepak a écrit :
> Alain <kua### [at] videotronca> wrote:
>> Le 19-06-28 à 06:13, Deepak a écrit :
>>> Hi,
>>>
>>> I am new to try with Povray application. I am trying to test performance of
>>> various compiler like (gcc, clang) on CPU with
>>> scenes/advanced/benchmark/benchmark.pov (as suggested here-
>>> http://www.povray.org/download/benchmark.php)
>>>
>>> What I observed is the rendering is finishing in ~1.5 seconds only.
>>> So it's difficult to compare the results with various compiler for such short
>>> duration.
>>>
>>> Is there any large data set which runs for considerable time? or how to change
>>> or what parameter to change on .ini file to run it with considerable time.
>>>
>>>
>>> Thanks in advance for your help,
>>> Deepak
>>>
>>>
>>
>> Some things that you can do that will increase the rendering time :
>> (original values from the ini file)
>> For the INI file :
>> Increase the resolution. Something like 4000 by 4000 (512 by 512)
>> Lower the antialias threshold to 0.01 (0.3) and Sampling_method=2 (1)
>> Set High_Reproducibility=on (off)
>>
>> Within the scene itself :
>> Turn radiosity ON (this will have a large effect):
>> #declare use_radiosity = true; (false)
>> In the global_settings block:
>> Set max_trace_level to 255 (12)
>> Increase the radiosity count to 800 (80)
>> Set nearest_count to 20 (5)
>> Reduce error_bound to 0.2, or even 0.1 (0.5)
>> Change low_error_factor to 0.2 (0.5)
>> Increase recursion_limit to 3 (1)
>> Increase the photons quality:
>> Set photons{spacing 0.001} (0.007) [this will multiply the number of
>> photons shot by 49]
>>
>> Change the area_light:
>> light_source {
>>      <-0.7, 0.83, 0.24>*150000
>>      color rgb <3.43,2.87,1.95>
>>      #if (use_area_light=true)
>>         area_light 4000*x 4000*y  129,129 (4,4)
>>     adaptive 3 //added Will start sampling at 9 by 9
>>         jitter
>>         orient
>>         circular
>>      #end
>>
>>      media_attenuation on
>>      media_interaction on
>>
>> }
>>
>> In the definition of M_Watx4, add dispersion :
>>      interior {
>>         ior 1.31
>> dispersion 1.01 //added
>> dispersion_samples 20 //added, default = 7
>>         fade_distance 0.8
>>         fade_power 1001.0
>>         fade_color <0.02, 0.20, 0.06>
>>      }
>> Do the same for Mat_Glass :
>>      interior {
>>         ior 1.5
>> dispersion 1.02 //added
>> dispersion_samples 20 //added
>>         fade_distance 0.12
>>         fade_power 1001
>>         fade_color <0.6, 0.5, 0.7>
>>      }
>>
>> Change the media used for the clouds:
>> samples 120 (7,7)
> 
> 
> Hi Alain,
> 
> Thanks for your suggestion. I tried changing benchmark.pov and benchmark.ini
> files and observed that render time is increased too much. The application is
> running for hours. Is it expected?

Absolutely.

ALL of my suggestions are intended at making the render time go up.
Some, like radiosity and area_light, affect the rendering speed for 
every pixel. Anti aliasing make higher contrast areas take longer. 
Other, like adding dispersion, only affect the areas where you have 
transparent objects.

Increasing the resolution mean that there are now WAY more pixels to render.

> Also let me know if any specific option to run/render in parallel on multicore
> system.

POV-Ray version 3.7 automatically use all available cores.

> 
> I will Try your another suggestion too and update.
> 
> Thanks,
> Deepak
> 
> 

To see the impact of «adaptive», try adaptive 0 and compare with adaptive 3.

You can safely reduce the radiosity count to something more reasonable, 
like something in the 60 to 100 range.

I don't think that you ever reach the max_trace_level.

An anti aliasing threshold of 0.01 is uselessly low. Something like 0.05 
to 0.1 is more reasonable.


Alain

Post a reply to this message

On 7/2/2019 7:11 PM, Bald Eagle wrote:
> In all
> cases the maximum value is 512.

Yeah, it sucks so much when you run into this limit!!

(Joking.)



Michael

Post a reply to this message