[Bf-cycles] Branched PT + Correlated Multi-Jitter & MIS on Background increase rendertimes tremendously!

[Jeffrey]

With CUDA, I've found the two methods are now within a second or two of 
each other on the test scene. CMJ is averaging at 43 seconds, Sobol is 
averaging around 42. I went back to the official distribution of 2.73 
and CMJ went up to 1:08 on average. Sobol is 41. This is rendering using 
the GUI, not CLI. I was not able to test OpenCL GPU because the kernel 
failed to compile.

Fedora 20 x64 with proprietary Nvidia drivers 346.35
Nvidia GTX 560 Ti 2GB

On 02/27/2015 01:42 AM, Sergey Sharybin wrote:
> Here's a really quick patch which seems to make CMJ really close to 
> Sobol on my computer now: http://www.pasteall.org/56937/diff
>
> Can't commit it to master because it's quite dangerous change (using 
> fast sqrt might backfire in some scenes), but if someone can compile 
> blender could confirm performance improvements that'd be cool :)
>
> On Fri, Feb 27, 2015 at 11:49 AM, Sergey Sharybin 
> <sergey.vfx at gmail.com <mailto:sergey.vfx at gmail.com>> wrote:
>
>     It's not getting disabled, but it doesn't use some optimizations
>     either.And seems we can use some CUDA intrinsics in few places to
>     hopefully speed things up. Will give it a try.
>
>     On Fri, Feb 27, 2015 at 8:55 AM, Zauber Paracelsus
>     <zauber at gridmail.org <mailto:zauber at gridmail.org>> wrote:
>
>         Doesn't Correlated Multi-Jitter Sampling get disabled when
>         rendering in
>         GPU mode?
>         _______________________________________________
>         Bf-cycles mailing list
>         Bf-cycles at blender.org <mailto:Bf-cycles at blender.org>
>         http://lists.blender.org/mailman/listinfo/bf-cycles
>
>
>
>
>     -- 
>     With best regards, Sergey Sharybin
>
>
>
>
> -- 
> With best regards, Sergey Sharybin
>
>
> _______________________________________________
> Bf-cycles mailing list
> Bf-cycles at blender.org
> http://lists.blender.org/mailman/listinfo/bf-cycles

-- 
Jeffrey "Italic_" Hoover

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.blender.org/pipermail/bf-cycles/attachments/20150227/3228c82b/attachment.htm