[Bf-cycles] Size of tiles for hybrid system CPU+CUDA/OpenCL/Phi

[Sergey Sharybin]

Hi,

Those are actually ideas which Mai was looking into for Split kernel as far
as i know. They go as following:

- For a small tile, render multiple samples at a time.

This will keep compute device busy even for a small tile size. To my
knowledge, this is how split kernel now works. This is probably something
we need to extend to other kernels?

This approach is mainly useful for final renders. For a viewport (when we
render samples one by one) we need something different.

- Allow rendering multiple tiles with a single sample on the same device.

I think Mai also was experimenting with this, not sure what was the
conclusion here. But this is something which would help keeping viewport
performance at a max.

Hopefully Mai will drop her knowledge here.

P.S. There might be some other tricks here, but currently not sure what
that would be.


On Sun, Jul 23, 2017 at 5:03 PM, Jaros Milan <milan.jaros at vsb.cz> wrote:

> Hi,
>
> I would like to open question about size of tiles for hybrid system
> CPU+CUDA/Phi ( CUDA and OpenCL require bigger tile than CPU ).
>
> Now I use the OpenMP implementation for using bigger tile on hybrid system
> CPU + Intel Xeon Phi.
>
> Sergey has the some idea about this.
> Best regards
>
> Milan Jaroš
> Research Assistant
>
> IT4Innovations national supercomputing center
> VŠB – Technical University of Ostrava
>
> 17. listopadu 15/2172  |   708 33 Ostrava-Poruba   |   Czech Republic
>
>
>
> e-mail: milan.jaros at vsb.cz  |   web: industry.it4i.cz   |   phone:
> +420 597 329 583 <+420%20597%20329%20583>   |    map: goo.gl/maps/BpF0z
>
>
> _______________________________________________
> Bf-cycles mailing list
> Bf-cycles at blender.org
> https://lists.blender.org/mailman/listinfo/bf-cycles
>
>


-- 
With best regards, Sergey Sharybin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.blender.org/pipermail/bf-cycles/attachments/20170723/c58b87bf/attachment.htm