Hello,

I hope I've posted this in the right group. My name is Scott Holland.
I am currently a computer science graduate student at Appalachian
State University in Boone, NC.  I would like to use POVRay as a part
of my thesis project.

In particular I would like to apply something called GPGPU to POVRay.
GPGPU stands for General Purpose processing on a GPU. More information
on GPGPU can be found here:  http://www.gpgpu.org 

In short, I would like to see if it is feasible to use the latest
graphics cards to accellerate POVRay. My goal would be to modify the
SMP code of POVRay to identify each GPU in the system as just another
CPU to split the workload with. In the past, ray tracers have been
successfully written that use the GPU. However, they have been what I
would term bare bones ray tracers. They have not supported things like
refraction or radiosity. My thesis would test if it is possible to do
the same with a full-featured ray tracer.

In order to do this, I would need the source code for version 3.7.
Since 3.6 does not support multiprocessor systems. What would be
required of me to get the 3.7 source code? Thank you.

Scott Holland
sah### [at] csappstateedu

Post a reply to this message

Oops. Let me apologize for the name and e-mail used in the Author and
e-mail fields of my previous post. I use this same client for posting
to the general Usenet as well, and I'm not as keen to have my real
information in those fields when I'm out there. I'm sorry if I
offended anyone.

Scott Holland

On Fri, 04 Nov 2005 02:06:04 -0500, A Large Negro <non### [at] yobidness>
wrote:

>Hello,
>
>I hope I've posted this in the right group. My name is Scott Holland.
>I am currently a computer science graduate student at Appalachian
>State University in Boone, NC.  I would like to use POVRay as a part
>of my thesis project.
>
>In particular I would like to apply something called GPGPU to POVRay.
>GPGPU stands for General Purpose processing on a GPU. More information
>on GPGPU can be found here:  http://www.gpgpu.org 
>
>In short, I would like to see if it is feasible to use the latest
>graphics cards to accellerate POVRay. My goal would be to modify the
>SMP code of POVRay to identify each GPU in the system as just another
>CPU to split the workload with. In the past, ray tracers have been
>successfully written that use the GPU. However, they have been what I
>would term bare bones ray tracers. They have not supported things like
>refraction or radiosity. My thesis would test if it is possible to do
>the same with a full-featured ray tracer.
>
>In order to do this, I would need the source code for version 3.7.
>Since 3.6 does not support multiprocessor systems. What would be
>required of me to get the 3.7 source code? Thank you.
>
>Scott Holland
>sah### [at] csappstateedu

Post a reply to this message

A Large Negro wrote:

> In order to do this, I would need the source code for version 3.7.
> Since 3.6 does not support multiprocessor systems. What would be
> required of me to get the 3.7 source code? Thank you.
> 
> Scott Holland
> sah### [at] csappstateedu

In short, until 3.7 is out of the beta phase and the source is cleaned 
up, it's a no-go. Your only choice is to wait or find a way to adapt 3.6 
to work with multiple processors.

At least, this is the stance the POV-Team has taken for a number of 
reasons and I don't know that they're willing to make an exception.
-- 
~Mike

Things! Billions of them!

Post a reply to this message

Mike Raiford <mra### [at] hotmailcom> wrote:
> or find a way to adapt 3.6 to work with multiple processors.

  There are only two options for this:

  Run multiple copies of povray (like all multiprocessing patches for 3.6 do).
Disadvantage: Rendering will consume n times as much memory for n processes
than it does for 1 process (meaning that if 1 process requires 1 GB of
memory, 12 processes will require 12 GB).

  Replicate what has been done for 3.7. Disadvantage: An incredible amount
of work and completely useless (because it is being done anyways).

-- 
                                                          - Warp

Post a reply to this message

A Large Negro wrote:
> 
> In short, I would like to see if it is feasible to use the latest
> graphics cards to accellerate POVRay. My goal would be to modify the
> SMP code of POVRay to identify each GPU in the system as just another
> CPU to split the workload with. In the past, ray tracers have been
> successfully written that use the GPU. However, they have been what I
> would term bare bones ray tracers. They have not supported things like
> refraction or radiosity. My thesis would test if it is possible to do
> the same with a full-featured ray tracer.

It is unlikely that something like this can be successful.  Current 
graphics cards only have very specialized capabilities.  Therefore they 
are usually programmed in special languages.  Typical tasks for this are 
relatively simple and uniform algorithms like fluid simulation or image 
processing.  The main tracing function of POV-Ray and what's called by 
it deal with a huge bandwidth of algorithms and data structures.  The 
vast majority of the POV-Ray source code can get involved in this 
process somewhere.  You would either need to rewrite all this or have a 
fully featured C++-Compiler available that generates code for your GPU. 
  I'd say if it was possible to run the POV-Ray tracing function on a 
GPU you could run pretty much anything on it.

The only sensible way of making use of GPUs in POV-Ray would be 
secondary processings.  For this POV-Ray 3.6 and 3.7 would not make that 
much of a difference.

Christoph

-- 
POV-Ray tutorials, include files, Landscape of the week:
http://www.imagico.de/ (Last updated 31 Oct. 2005)
MegaPOV with mechanics simulation: http://megapov.inetart.net/

Post a reply to this message

A Large Negro wrote:
> Hello,

> In particular I would like to apply something called GPGPU to POVRay.
> GPGPU stands for General Purpose processing on a GPU. More information
> on GPGPU can be found here:  http://www.gpgpu.org 

Find me a GPU with double precision floating point, and you might make a 
start.  Currently, they're only just now (or maybe in the last year or 
so) supporting single-precision through the entire pipeline.  Given that 
most people are still running cards that don't even have that, I doubt 
NVidia or ATI will soon be releasing double-precision cards.

Anyway, the paralelization of GPUs is different from what is needed for 
POV-Ray.  True, GPUs have parralel pipelines - that are usually doing 
the same thing.  Even if they are working on different pixels, they're 
usually at least running the same program.  You wouldn't call different 
shaders for different pixels.  For POV-Ray, this might be needed, due to 
the interdependance of most scene elements (thanks to refraction, 
reflection, etc).  The only way to do this under the current GPU 
paradigm would be to write a shader that contains the *whole* of 
POV-Ray, and let it loose.

...Chambers

Post a reply to this message