[Bf-committers] The Daily Dweebs: Performance comparison of w/o Motion Blur on Haswells

Jaros Milan milan.jaros at vsb.cz
Wed Nov 15 12:10:14 CET 2017


Hi,

I would like to ask on integrate of Embree into Blender. That means it will be create a patch from Stefan's code? When do you plan create a branch (blender.org) for this feature? Because I don’t know, where can I send the fixes:)

I created emulator (alpha version) for SSE intrisics instructions. I tested on CPU with Embree - maybe it could be use for GPU:)

Thanks.
Milan

-----Original Message-----
From: Bf-committers [mailto:bf-committers-bounces at blender.org] On Behalf Of Brecht Van Lommel
Sent: Wednesday, November 15, 2017 11:29 AM
To: bf-blender developers <bf-committers at blender.org>
Subject: Re: [Bf-committers] The Daily Dweebs: Performance comparison of w/o Motion Blur on Haswells

We want to integrate Embree and get improved motion blur performance that way. So if you're interested in optimizing motion blur I think the logical thing would be to work on the Embree code and its integration into Blender, and figure out what we can improve there.

From this benchmark there's two things that would be useful to investigate I think. One is why rendering with MB is faster than without MB, that's really odd. And the other is why there's still frames where MB is 2x slower overall, even though the extreme motion blur seems to be only in a small part of the frame. Maybe there's something in the Embree algorithms that's still weak, or we're not providing the data properly.


On Wed, Nov 15, 2017 at 9:42 AM, Jaros Milan <milan.jaros at vsb.cz> wrote:

> Hi,
>
> I added two rendered pictures (153th frame) -  http://blender.it4i.cz/ 
> rendering/the-daily-dweebs. You can see what is difference between 
> blender's motion blur and embree's motion blur. But I think the 
> article https://embree.github.io/papers/2017-HPG-msmblur.pdf from 
> embree looks very nice.
>
> I don’t know If somebody works on improvement of the motion blur in 
> Blender Cycles? I could look at on this feature.
>
> Thanks.
>
> Best regards
> Milan
>
> -----Original Message-----
> From: Bf-committers [mailto:bf-committers-bounces at blender.org] On 
> Behalf Of Brecht Van Lommel
> Sent: Tuesday, November 14, 2017 7:46 PM
> To: bf-blender developers <bf-committers at blender.org>
> Subject: Re: [Bf-committers] The Daily Dweebs: Performance comparison 
> of w/o Motion Blur on Haswells
>
> That render time increase at the end is odd. The fact that it happens 
> with Embree but not BVH4 should give a clue, but I'm not familiar with 
> the implementation. It's also odd that Embree MB is faster.
>
> Maybe all the particles are separate objects? In the native Cycles BVH 
> without motion blur, those would have their transform applied.
>
> On Tue, Nov 14, 2017 at 12:22 PM, Jaros Milan <milan.jaros at vsb.cz> wrote:
>
> > Hi,
> >
> > I tried to make some comparison of w/o Motion Blur (BVH4 vs EMBREE-BVH8):
> >
> > http://blender.it4i.cz/rendering/the-daily-dweebs/
> >
> > Do you know why the rendering time increases from 907/908 frame?
> > (particles, ....)
> >
> > Thanks.
> >
> > Milan Jaroš
> >
> > _______________________________________________
> > Bf-committers mailing list
> > Bf-committers at blender.org
> > https://lists.blender.org/mailman/listinfo/bf-committers
> >
> _______________________________________________
> Bf-committers mailing list
> Bf-committers at blender.org
> https://lists.blender.org/mailman/listinfo/bf-committers
> _______________________________________________
> Bf-committers mailing list
> Bf-committers at blender.org
> https://lists.blender.org/mailman/listinfo/bf-committers
>
_______________________________________________
Bf-committers mailing list
Bf-committers at blender.org
https://lists.blender.org/mailman/listinfo/bf-committers


More information about the Bf-committers mailing list