I was wondering if there is something that will allow a machine of more than
one cpu to render an image faster. As I couldn't find one, I've written
something in python that takes a render, divides it into X vertical slices,
say 20, then runs as many render processes as there are CPUs available to
it. On completion of all slices, it joins them all up into the product.

Some points on it:

 o It leaves the decision of which CPU to give a render to, upto
   the kernel which up until now has done the job perfectly ok.

 o If the process is stopped, the next time it runs it will go through
   all generated slices, make sure they are valid, png format, correct
   dimentions, and finally md5sum (I'm not sure if this is a good idea
   as images with any kind of a random aspect may pose an issue,
   although I havnt had an issue yet.)

 o Once I've done a little more fine tuning, and if there is any
   intrest, I'll post the benchmark, but I think it's safe to assume
   that >1 cpus rendering will be much faster than one, the overhead of
   joining multiple slices and file verifications are relatively small.

Nima

Post a reply to this message

Are you aware that the soon-to-come POV-Ray 3.7 will support SMP
all by itself?
  The greatest advantage of it is that rendering in multiple threads
will not consume more memory than rendering in a single thread. (Of
course other advantages include the fact that since it supports
multiple threads "natively" it's handier because you don't need to
hassle with external scripts).

  In your case if the scene takes 1 gigabyte of memory, rendering with
two processes will require 2 gigabytes of memory.

-- 
                                                          - Warp

Post a reply to this message

Warp <war### [at] tagpovrayorg> wrote:
> Are you aware that the soon-to-come POV-Ray 3.7 will support SMP
> all by itself?
I am now! :)

Well that would most definitely be much better than anything I could do as
it would not have none of the overheads of external software, including
python itself, as you mentioned, as well as the fact that joining 8 slices
to create one image, even though looks identical (as far as I can tell),
the file size is about 25% larger than a single render would create.

Thanks for enlightening me Warp

Nima

>   The greatest advantage of it is that rendering in multiple threads
> will not consume more memory than rendering in a single thread. (Of
> course other advantages include the fact that since it supports
> multiple threads "natively" it's handier because you don't need to
> hassle with external scripts).
>
>   In your case if the scene takes 1 gigabyte of memory, rendering with
> two processes will require 2 gigabytes of memory.
>
> --
>                                                           - Warp

Post a reply to this message

Just another question, when rendering only a portion of an image, the +C
flag doesn't always work so nicely. Does anyone have any comments on what
they think of the +C flag for continuing a render?

Nima

Warp <war### [at] tagpovrayorg> wrote:
> Are you aware that the soon-to-come POV-Ray 3.7 will support SMP
> all by itself?
>   The greatest advantage of it is that rendering in multiple threads
> will not consume more memory than rendering in a single thread. (Of
> course other advantages include the fact that since it supports
> multiple threads "natively" it's handier because you don't need to
> hassle with external scripts).
>
>   In your case if the scene takes 1 gigabyte of memory, rendering with
> two processes will require 2 gigabytes of memory.
>
> --
>                                                           - Warp

Post a reply to this message

Hi Warp,

Thats good news!

I guess it was high time for that, as 2005 is the year of the "Double 
CPU's".
Do you know more Details if its "only" for SMP-Systems or also about 
"Network-Rendering" ?

--Theo


"Warp" <war### [at] tagpovrayorg> schrieb im Newsbeitrag 
news:42832fb4@news.povray.org...
>  Are you aware that the soon-to-come POV-Ray 3.7 will support SMP
> all by itself?
>  The greatest advantage of it is that rendering in multiple threads
> will not consume more memory than rendering in a single thread. (Of
> course other advantages include the fact that since it supports
> multiple threads "natively" it's handier because you don't need to
> hassle with external scripts).
>
>  In your case if the scene takes 1 gigabyte of memory, rendering with
> two processes will require 2 gigabytes of memory.
>
> -- 
>                                                          - Warp

Post a reply to this message

* * * Theo Gottwald <yne### [at] it-beraterorg> wrote:
> Do you know more Details if its "only" for SMP-Systems or also about 
> "Network-Rendering" ?

  AFAIK it's SMP only currently (memory must be shared between the
processors). I assume that the current modifications will make it
easier to implement networked rendering, though.

-- 
                                                          - Warp

Post a reply to this message

Warp <war### [at] tagpovrayorg> wrote:
> Are you aware that the soon-to-come POV-Ray 3.7 will support SMP
> all by itself?

soon-to-come ? on linux too ? hu !!?
2006 or 2007 ?
;-p

Post a reply to this message