Multithreaded object update in trunk

For those who don’t know yet, it’s quite old news, since the feature is in blender trunk for about a week allready.
I wodered, how does it come not many applaud to this internal change in blender. I almost never post here anything like this, but this deserves it’s own thread.

For me, it’s definitely the feature number one of the last year!

For anybody serious with blender and animation with it, this is a huge and important update.

Thanks go to Sergey Sharbin!

P.S. It will be interesting to post some FPS benchmarks here, compare with your own files.

Hi pildanovak, what exactly does this new feature does?
I still dont come around…

It enables to update objects on multiple cores, thus makes a very essential part of blender multi-threaded.
This means blender will now use more processor cores during e.g. animation playback, when you have more objects with modifiers in your scene. It means big speedup of blender generally. So, nothing “visible”, no new buttons e.t.c.:slight_smile: but simply better speed and use of your hardware. If you do animation regularly, you should notice the difference.

Which is fantastic as most character models have several or even many object chunks. Like hair, beard, eyebrow mesh objects - Clothing and equipment objects.

Hadn’t realized this wasn’t already there. Yet another quiet surge forward into 2014.

there was a lot of comments about this that it needs that… but u know most blender uses are like We want cycles yeeeeej or we want new UI by andrew price… yeeeeej… So few post from people who acctualy know stuff where gone in 100000 post about UI.

On the last few days builds from builbot.
BTW On these with the new (and very interesting) cycles volumetric shaders.

The default cube
Enter edit mode
Enter sculpting mode

or just sculpt
enter edit mode
back to sculpting

On Mac builds.

A confirmation could help before I report it.

This is not yet the fabled depsgraph update though, is it?

It’s half of it, technically.

There were two gsocs.
I believe Joshua Leung(Aligorith) did the one you’re thinking off, which is trying to restructure the architecture of the depsgraph.
This one is is by Sergey(Naz-gul), and is focussed on allowing Blender to use multiple cores for dependancy graph related things.

This looks like a bug, please report it in the tracker, thanks.

@michalis: I am using OSX 10.6. I followed your directions and also experienced a crash.


hello, this file shows the case where this works nicely: the playback is x3.5 faster on a 4core computer here.

also, if you e.g. duplicate these objects, you can also see the influence of viewport drawing which is not threaded. E.g. look away from the objects, and the playback will be even faster with the new version.

I am using OSX 10.6. I followed your directions and also experienced a crash.

Well, we may facing some other nasty issue:
I fixed it anyway.
Deleted the User/Library/ApplicationSupport/Blender/2.69 file, made a fresh blender setup, and now it works.

In preferences/file/ load UI is checked by default.
Loading an older file coming from the official blender 2.69 will mess the application again.
hmmm really annoying.
Better keep LoadUI unchecked, or should I propose to get rid of it?
As it is now, you never know. Maybe some warning?

So, I didn’t report it yet.
Better discuss it first.

Only slightly more than 2x here, also quadcore. A substantial improvement still, but anyway. My specs in signature.

But this could improve also the particles generation??
Could it cut the particles “preparation” time when you click “render”?

I’m doing an archviz exterior and everytime I render, I’ve to wait minutes before Cycles starting because blender is generating the particles instances and it uses only a CPU core. This “multithreaded object update” can help for this??

Are you sure your i5 is quadcore? a quadcore should usually show like 8 cores in windows task manager, because of hyperthreading…
also, it can be speed of drawing on the card which does it perform not so effective, try e.g. bounding box draw mode, that shows how much time the actual calculations take…

@pildanovak I think only i7 and xeon come with hyper-threading

no i5 goes in 2 core with 4 threads and 4 cores and 8 threads

only i7 and xeon, with hyper threading, correct.

does this also affect the boolean modifiers? I think they are currently only single threaded or is that a task that has to be addressed with the way how the CARVE library is being used?