It should be included in the next round of buildbot builds (done tonight or tomorrow depending on your timezone).
Any news on Subdiv? Its still too terribly slow!
I also want to know this ,man I want so badly a subdiv animation speed up in Eevee viewport
This will be a game changer
Vertex sharpness!!! Does this finally mean we’ll be able to model with open subdiv and use vertex sharpness to control how far the subdivided mesh can move away from each vertex?
No more having to use edge crease on border tri poles (which always results in a hard pinch all the way up the mesh).
guys, we have to wait a few months …
I am sure that by blender 2.81, 2.82 at the last, all these performance problems will be focused and fired.
it’s the modelling benefits I’m interested in. Controlling subdivision using edge creases alone isn’t good enough.
Does anyone know when Blender 2.8 is going to support OpenSubDiv GPU/CPU acceleration? I refuse to use blender 2.8 until they implement this feature like 2.79 has. I use subdivision surface modifiers on my rigs because they are such low poly for better deform and animation performance.
Also, my animations wouldn’t update correctly with one of my more complex rigs in 2.8
I end up having to click the timeline position a few times before the rig position fully updates.
That means you have circular dependencies in your rig, it’s not Blender’s fault.
OSD on GPU has been postponed unfortunately.
Apparently they intend to rewrite completely the subsurf code, so that it would be a mesh option (just like it used to be in old old blender versions) instead of a modifier, to support GPU acceleration. As a modifier it’s apparently impossible to make it on GPU (although it was the case in Blender 2.79, there were limitations and requirement, such as being at the top of stack, and does not support case such as another object being shrinkwrapped on it). So this will take time to redesign and rewrite it…
About your rig issues, it’s indeed probably a cyclic dependency. Blender 2.8 depsgraph actually introduced a new kind of cyclic dep. : when a bone transform value is being driven by another transform value of the same bone ( e.g location drives scale). Maybe check that.
From how I understand the devs strategy, first they are piece by piece solving various problems related to the depsgraph and edit mode, slowing a bit all blender, especially in animations, and once this is solved (depsgraph seems to be well under way), will be able to quietly make the opensubdiv on Gpu. (probably the most important thing which takes time, I think it is to eliminate as much as possible alternations between gpu and cpu , actually, there are many functions managed by the cpu and others from the gpu, especially in edit mode and in modifiers (I hope I have not misunderstood, and above all I have described enough what happens …)
So are you saying blender 2.79 doesn’t work correctly because it handles my “circular dependencies”?
In 2.8, for example in my characters leg, the hierarchy of bones act disconnected and lag behind the foot bone I am moving with the mouse and take a 1-2 seconds to catch up. Its really strange.
None of this happens if I turn off the sub-div modifier. So my rig couldn’t have cyclic dependencies.
Something is wrong with the sub-div modifier and like I said, I refuse to use 2.8 until the acceleration is back.
no that the master its open for the blender 2.82 hopefully they will support gpu opend subdiv
Any news? It’s seems no one talking about gpu opensubdive acceleration anymore ,
Any update from the developers will be nice
Its target is blender 2.82?
It’s sad☹️ they have one month before the features implementation ends for blender 2.82 , I hope somehow they speed up the animation playback
Subdivision is basically unusable in it’s current state. I’m surprised this was allowed to go out like this. It’s a mess.
Yesterday i managed 42 million tries with multi res. I don’t how it was possible.
The issue is not display and multires, the issue is subsurfs, editing and animation of rigged characters. The whole point of OpenSubdiv was to make this faster, not slower.
removed by popular demand
As I understand it, it’s not quads, it’s poles. As in vertices with any number of edges attached that is not 4.
But frankly, Subsurf performance is utterly appalling regardless of whether you avoid those or not. And you should never, ever have to choose between good topology and performance.