[Bf-committers] GPU computing

Istvan burbank iman1100 at gmail.com
Tue Nov 24 22:26:07 CET 2009


I know little to nothing about this, but the rendering engine as a
whole was thrown out there, and deemed not the best area. But what if
it was more specific? What if only the ray tracing (AO, refl, transp
etc) were the only part done by the gpu? That might be a smaller task,
and more feasible a project. Maybe ray tracing isn't the best part,
but the concept could be any 'part' of the rendering engine.

Also, i have heard that Bullet physics will have OpenCL integrated
into it, and it would be cool to have particles, *smoke* and fluids
working on the GPU as well as the CPU if that is possible. These are
all examples of 'small' equations that GPUs are better at doing,
however using the GPU in addition to the CPU would be even better so
that the GPU is treated just like another thread in that it isn't
dependent on CPU vs. GPU speed. For all of these if a library was
built that could be used by these simulations it might be easier to
test/integrate since taking on the project full on has already been
deemed to be too much work.

Again, I all of what I just said may be useless and wrong, sorry if
that is the case, but I guess what I'm suggesting is be more modular
and then as that succeeds add more pieces to those that use this
library.

~This will be awesome, I can't wait to see what comes of it, Istvan.

PS. I'm not a real 'coder' (working on it - still in high school) but
I would love to help in what ever I can. If this attempt fails, please
excuse me.

On Tue, Nov 24, 2009 at 3:05 PM, Shaul Kedem <shaul.kedem at gmail.com> wrote:
> Hi,
>  I've actually wrote a test to check opencl with post processing
> (using nodes to add gamma, for example) and here are my findings:
>  - Brecht gave me the idea to work on the post pro area, saying that
> if we can move a full post pro flow to the GPU we might see
> performance gain
>  - Moving data from memory to the GPU takes time. It can (and in my
> case did) take more time than doing the calculation in the CPU to
> begin with. This can be lowered with shared memory usage and having
> more to do on the CPU, but that's the general idea.
>  - Writing opencl code mean finding parallel algorithms where there
> are none is use in the code right now
>  - When I asked ton, he said that the first thing is to have work
> units in the post pro engine, so multi-CPU work can be done
>
>  Anyway, good luck, opencl looks like its going places :-)
>
> Shaul
>
>
> On Tue, Nov 24, 2009 at 1:28 PM, Marko Radojcic <sambucuself at gmail.com> wrote:
>> Hi! I was working on such an idea as well, some time ago, I think that
>> Blender would most benefit from GPU raytracing and fluid
>> simulations...
>>
>> I recommend OpenCL, it is a framework that integrates CPU, ATI GPU and
>> NVIDIA GPU functionality.
>> _______________________________________________
>> Bf-committers mailing list
>> Bf-committers at blender.org
>> http://lists.blender.org/mailman/listinfo/bf-committers
>>
> _______________________________________________
> Bf-committers mailing list
> Bf-committers at blender.org
> http://lists.blender.org/mailman/listinfo/bf-committers
>



-- 
Knowledge is a measure of how many answers you have, intelligence is a
measure of how many questions you have.
"Elegance has the disadvantage that hard work is needed to achieve it
and a good education to appreciate it." - E. W. Dijkstra
"The price of humility doubles each time you have to buy it back" ~
rick dickulous
"Intelligence is not enough. Intelligence plus character, that is the
goal of true education." ~ Martin Luther King, Jr.
"If everybody does it, it must be stupid" ~Ton Roosendaal


More information about the Bf-committers mailing list