[Bf-cycles] State of cycles baking

Vilem Novak pildanovak at post.cz
Wed Jul 15 19:38:41 CEST 2015



thanks to you both for your replies.

regarding memory, it is true that huge improvement has been done, thanks to 
Dalai, and I never get all my memory used now in my work.

 The problem I am facing  is crashes which report EXCEPTION ACCESS VIOLATION
in the console, 

and that's why I thought these have to do with memory use, but really not in
terms of memory overflow.

Maybe, Greg Zaal managed to create a good example in this newly reported 
bug, it might really be the same issue:


to the ones which Dalai mentioned, there's also this one recently reported 
by me:

https://developer.blender.org/T45343 (baking of displacement channel of the 
shader, aka "another normals bug")

A single bug wouldn't be a problem, but having to face with all of these in 
my daily work sometimes creates a "maze" I need to go through blindfolded 
without hitting a wall  - to get good results... :)

I really understand the limited manhours problem, and thanks for all your 
hard work.




Hi Vilem,

><i> - Not optimal memory usage. AFAIR Dalai was doing some work in there, not sure if he committed all the changes he was planning to do, would prefer to hear from him here. But optimizing something further is always possible.
All the memory optimizations were already merged back in April
(394c5318). They make a world of difference already. If this is not
enough all I can think of is for you to adjust the tiles size
according to your system and see if you get a working setup.

That said, it's really hard to handle bugs if you can't reproduce the
issue. The crashes can be happening in the pre-processing phase, in
Cycles, or in the post-processing. Each one would require a different

One optimization I would like to see implemented at some point is
multithreading in the pre-processing stage. But that will only affect
performance, not memory usage.


As for the Cycles Baking bugs those are the ones I'm aware of:

<a href='https://developer.blender.org/T43553'>https://developer.blender.org/T43553</a> (dupli groups)

<a href='https://developer.blender.org/T42015'>https://developer.blender.org/T42015</a> (boolean modifier)

<a href='https://developer.blender.org/T43388'>https://developer.blender.org/T43388</a> (concave faces when enabling AA jittering)

<a href='https://developer.blender.org/T45232'>https://developer.blender.org/T45232</a> (transparent textures)

And as you know, the only one I'm actively pursuing at the moment is
the dupligroups one. More developers are welcome to join the effort.
But I'm afraid there are no concrete plans of prioritizing this.


2015-07-15 4:31 GMT-03:00 Sergey Sharybin <<a href='http://lists.blender.org/mailman/listinfo/bf-cycles'>sergey.vfx at gmail.com</a>>:
><i> Hey,
</i>><i> Surely there is intention to improve Cycles baking, like any other are of
</i>><i> Blender. Only unfortunate thing here is limited manhours of developers..
</i>><i> To me it sounds you've got two separate thing here:
</i>><i> - Some bugs (crashes, baking not working etc). Thins things are to be
</i>><i> reported to the bug tracker. Make sure me and Dalai are CCed in there and
</i>><i> we'll look into the issues. Maybe Thomas will also be interested in looking
</i>><i> into those reports?
</i>><i> - Not optimal memory usage. AFAIR Dalai was doing some work in there, not
</i>><i> sure if he committed all the changes he was planning to do, would prefer to
</i>><i> hear from him here. But optimizing something further is always possible.
</i>><i> It any case, it will help a lot having files which demonstrates every issue
</i>><i> you've got, so we can investigate exact reasoning of failures.
</i>><i> On Tue, Jul 14, 2015 at 10:13 PM, Vilem Novak <<a href='http://lists.blender.org/mailman/listinfo/bf-cycles'>pildanovak at post.cz</a>> wrote:
</i>>><i> Hello,
</i>>><i> I wanted to kindly ask, if there is intent to improve cycles baking.
</i>>><i> I am working with it for about half a year on a project - overcoming many
</i>>><i> obstacles, building own builds with various patches from the tracker, to get
</i>>><i> the job done. I even wrote an addon to facilitate the process.
</i>>><i> I could have said - oh, it was a bad path to go, I'll do it in another
</i>>><i> software/renderer...
</i>>><i> But I also like to help with blender development, so I stick with it. I
</i>>><i> reported allready several of the bugs.
</i>>><i> Some, I didn't report because they are kind of hard to catch and almost
</i>>><i> mysterious- Baking simply often doesn't work on larger files, and crashes
</i>>><i> very often, even when baking only on CPU and memory isn't by far full..
</i>>><i> The thing is, I am a bit frustrated by seeing how great the renderer
</i>>><i> itself goes forward, with bugs getting fixed relatively quickly, while
</i>>><i> baking bugs hang there so long and lonely.
</i>>><i> I know that Dalai Felinto, who developed cycles baking, is very busy with
</i>>><i> another work on blender - multiview, and I understand that probably all
</i>>><i> cycles developers have other things to do.
</i>>><i> So, I just thougt - maybe there isn't common knowledge of the fact that
</i>>><i> cycles baking is actually broken beyond point of usability(), and if I say
</i>>><i> that, somebody just might try to look into this.
</i>>><i> I'd be also curions about some bigger finished projects with cycles
</i>>><i> baking, where something larger than the usual ""
</i>>><i> happy blending
</i>>><i> VN</i>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.blender.org/pipermail/bf-cycles/attachments/20150715/968d3cf6/attachment.htm 

More information about the Bf-cycles mailing list