About Wireframe viewport view. In Master if Optimal Dispay is checked in Subdivision Surf modifier, then if you modify Wireframe slider under “Viewport Overlays” → “Geometry” there are no visualization changes in wireframe of the mesh. Is this something intentional in Master?. In Subdivision_work branch that slider still influences wireframe visualization when Optimal Display box is checked.
Edit:
I see you guys talked in previous posts about Wireframe view. I’m not sure if about this I have mentioned.
Old 2.7* scenes are opened in Blender 2.8+ with that Wireframe Slider=0.5 and not =1 as it is by default in 2.8+
Wow! I am testing now, and it’s a pretty major speed improvement!
I’m running Windows 10 64bit, with an old 980 TI OCed.
Trying subDiv Level 3 on a mesh:
Toggle into Edit mode:
2.93 - 5 seconds
3.0 - 1 or < 1 second
Soft Selection Movement:
2.93 - Doesn’t even work/never even starts to move. Keeps lagging out for a long while.
3.0 - maybe 5-10 FPS, but can somewhat interactively move the mesh/verts!!
Edit: Also, running my cloth simulation/modifier stack in 3.0 vs 2.93.
3.0 is definitely more “interactive” with the viewport updates, where 2.93 stalls out again/i can’t see what’s going on and gotta just try and stop the playback/wait for it to update.
While its not performing as well as I expected it too in every circumstance overall I am very happy with what we’ve gotten.
Obviously the pose performance drop has been fixed with the latest build and I posed Rocket Girl to celebrate. I think this expresses pretty well how I feel right now.
In what circumstances isn’t it working for you?
Practically all my tests went great here.
Even the Rain rig, which is rather performance consuming -45 fps without subdiv here- leads to same 45 fps with subdiv level 2, if we remove the Data Transfer modifier of the face, which is a performance killer. I think it’s pretty awesome!
I can confirm the crash your experiencing with an RTX3060 (not that I think the graphic card matters in this case).
Insofar as performance I find Rain V2 performing “ok”, im not hitting your numbers but she at least stays above 24fps. I’ve tried looking for the data transfer modifier that should be on her face but couldn’t find it.
I will take another look later this evening because if there are specific things on rigs that slow them down I’d love to know what they are so I can either tweak the rigs or avoid those things in my own rigs.
Another rig whose performance I found disappointing is the commercial Jim rig. => And I’ve actually just fixed that. Chalk it up to user error. There were some cases where the subdiv modifier was not the latest in the stack. I don’t know how I’ve missed that. Ive recently redownloaded it because there were some issues with library overrides so perhaps something changed with that version or i’ve just been stupid. Both are possible
Jims rig performs swimmingly now. Its not as fast with subdiv as it is without (obviouisly) but its more then fast enough now.
I hope I didn’t come of as too negative. I am still very happy with what we got.
Maybe we should ask Clément and Sergey if the behavior in Master is due to any reason. Also ask if old 2.7 files should be opened with Wireframe slider=1 in 2.8+
1 - Add subdivision surface modifier to cube
2 - Enable “on cage” in the modifier
3 - Turn X-ray mode on
4 - Go into edit mode
5 - Try to select a vertex where its shown
Result: no selection
Expected result: vertex under cursor is selected
What happens in X-ray mode is that it takes the position of the vertices on the non subdivided mesh. If you guess where they are and click you can actually make selections. Surprisingly if you turn X-ray mode of it works as expected.
The sculpt data buffer was not filled, should be fixed now. Custom normals is still not supported though.
No, it is unlikely to go in 3.0 after all (tomorrow is the current deadline). Code review took a while, and I took even longer to address the concerns, I still have a little bit to do. Who knows, maybe it can still go in.
It would be good, but on the other hand, this might mean 3.1 introducing both the new evaluator and OpenSubDiv becoming a mesh component (as opposed to a modifier). The modifier won’t go anywhere, but it won’t be near as fast. It could also mean additional optimizations if it can be done.
Thanks for the update!
Indeed, would have been great to have it part of Blender 3.0, but it’s already very fortunate to have a working patch, for impatient beta testers we are. Some of us have been waiting eagerly for years for this moment It will feel like a breath a fresh air for animators, when it will be officially part of an animated production movie pipeline. Good luck for the final polish and fixes for inclusion into master.
I second this, it is a little disappointing in my opinion, but I understand why the devs. really want to make sure it is ready for prime time with no regressions (as noted in the tracker entry). Skipping 3.0 and getting it into 3.1 early in the development cycle will help greatly with that (not only that, it will ensure that serious improvements are still coming for LTS users).
I am waiting this from 2.8 ) not sure why developers do not thought about regression then(2.79 was very fast).
I thought maybe they will fix it in 2.81 ,or 2.82
I am Glad Kévin started to make Subdiv great again, but for me very sad it was delayed again.
