[Bf-cycles] Amd cpu - ati saphire radeon gpu

[David Fenner]

I am already using openCL on GPU+CPU   with an nvidia card and an intel
cpu. The thing is that it is a fast processor but not an extremely fast
video card, so with opencl they both seem to render at the same speed,
which is actually faster than using only the video card with cuda. The
problem is, however, that on systems with a faster GPU, the tile that
finishes first (probably gpu tile)  needs to wait for the other tile (cpu)
to finish before jumping into a new tile, so in systems with a difference
in gpu/cpu speed this bottlenecks everything. Is this an opencl bug or a
cycles problem? seems like the obvious thing is that even the first tile to
finish should jump to the next one right away, but it is not the case with
opencl gpu+cpu. I know that this could be considered a bug, but since
opencl isn't even officialy supported I'm just commenting on that opencl is
already useful, just bottlenecked by this.


2014-02-12 13:25 GMT-03:00 Nazim Mer <nazim.mer at gmail.com>:

> Hi,
> according that threads participants, AMD 14.1 drivers have a working
> implementation...
>
> http://blenderartists.org/forum/showthread.php?254521-A-good-news-for-AMD-ATI-Graphic-cards-owners/page61
>
> With that in mind, any developer thats adding features to cycles should
> now consider using conventions and code suitable enough to work on  both
> CUDA and Opencl Capable devices.....
>
> Bare in mind that INTEL,NVIDIA and AMD use opencl.....
>
> Getting things a step closer to multi-platformness should be an objective
> for blender cycles.....
>
> As well focus on opencl bug reports(commencing February 1st 2014)
>
> And finally optimizations for the opencl implementation of cycles...
>
> Peace out
>
> PS can cycles developers review the following quote, and possible reflect
> this on the cycles code/kernel.....
>
>
>
>
>
>
>
> *So, in the end, for practical purposes pretty much only the basic
> features already enabled by default will work. These are the features I
> really haven't been able to enable in any way: __LAMP_MIS__ --> doubles the
> OpenCL kernel build time, PC hangs or Blender crashes when the render
> starts __CAMERA_MOTION__ --> errors in console __HAIR__ --> needs
> #__MULTI_CLOSURE to not end up in console errors, but makes Blender crash
> when compiling the OpenCL kernel otherwise *
>
> Peace
>
>
> On 3 January 2014 14:31, <emmi.spammi at t-online.de> wrote:
>
>> Hi,
>>
>> I'm also following the developement/fixing of the ATI issues. It seems
>> to work. I can render on my AMD Pitcairn.
>> However it is much slower than my CPU (which was not the original
>> intention ;)).
>>
>> I would be interested in any news of performance improvements.
>>
>> Best regards
>>
>> Emanuel
>>
>> PS: CPU: Intel i5-4570 (3,2GHz); GPU: AMD Radeon HD7850 1GB
>>
>>
>>
>> Am 03.01.2014 14:24, schrieb Gottfried Hofmann:
>> > Check this threa, especially the last few pages:
>> >
>> >
>> http://blenderartists.org/forum/showthread.php?254521-A-good-news-for-AMD-ATI-Graphic-cards-owners
>> >
>> > Seems like there is hope for AMD users...
>> >
>> > On 01/03/2014 02:12 PM, Geo A wrote:
>> >> Are there any news regarding cycles being compatible with AMD cpus or
>> >> ATI gpus sooner or later?
>> >
>>
>> _______________________________________________
>> Bf-cycles mailing list
>> Bf-cycles at blender.org
>> http://lists.blender.org/mailman/listinfo/bf-cycles
>>
>
>
> _______________________________________________
> Bf-cycles mailing list
> Bf-cycles at blender.org
> http://lists.blender.org/mailman/listinfo/bf-cycles
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.blender.org/pipermail/bf-cycles/attachments/20140212/39c89331/attachment.htm