I’m editing a model with 34 objects in it, and currently a mesh with about 10-20 materials and 600k faces.
When I select 500k of those faces, just rotating around the viewport tanks my framerate to about 1fps.
Editing in general is about a 1s wait on anything I do. I’m using local view and the rest of the faces hidden.
I’m on a 1080ti, i7-6700k, 32gb ram, using today’s build of 2.80 (as of posting).
EDIT: Hmm… switching from UV/Editing to Modeling workspace and then back again seems to have solved the framerate issue, but editing is still dog slow. Just a circle select takes 0.5 seconds.
EDIT: Operations are still painfully slow. Unwrapping 21k polygons took more than 5 seconds. Rotating the result is again, at most 1fps in the UV editor.
I don’t know the internals to confirm that however, from what I’ve gathered on other threads, this is related to it. I wondered about that too since my sculpt objects don’t have any subdivision surface modifiers.
Here you imply that it is a difficult problem to solve, you do not make it clear that the problem will be solved in 2.80 and that it will be rather in the future, Instead a clear “we will fix/improve the perfomance it before 2.80 stable”… Although you link to the same task, it is not clear in this one that the problem will be solved in 2.80. If not you’d “like” to try to fix it.
That’s why you should link to the post rather than giving your interpretation without the nuance or accuracy.
There will be more optimizations for 2.80, they are a priority and the aim is to at least fix any performance regressions. But if they will address the specific performance problem X or reach framerate Y with Z amount of polygons is impossible to say without defining the problem clearly.
Transforming many vertices is likely to improve a lot still for 2.80, UV unwrapping 21k polygons not but navigating in the UV editor will likely be improved.
I just found a worse case scenario than editing a single mesh with a few hundred thousand polygons in it:
Editing a mesh with a few hundred thousand polygons in it that has 10+ linked duplicates.
We’re not talking about 1fps then, but rather 5-10 SECONDS between each frame, making it 100% unfeasible to work with. And I just selected a few dozen edges to try.
Here, 3D Studio Max shines, as it handles instances like they’re free (and I’m not even using RailClone or some other optimization plugin here). Editing edges in the same file in Max runs at a smooth 60fps.
Again, since there’s a subdivision discussion around this: I’m not using it. This is imported mesh data.
@brecht is this also an optimization issue? 10 seconds between each frame vs. 60fps is quite a gulf to cross…
So 2.8 is now out and I it’s amazing for low-poly models… but high-poly… I know which program I (sadly) still prefer:
(Note, this isn’t the worst case scenario file I tested with that had 5-10 seconds between each frame. I don’t have time to test that yet, but from the above results, I don’t expect anything to have happened since February this year.)
EDIT: Note that the recording software I used only captured 25fps, but you can see at least in the second video that I have 150+fps in Max compared to single digits in Blender.
It’d be really interesting to hear a developers perspective.
Being one of many who are checking that high priority ticket on a weekly basis for months, I just die to know what the consensus is, and what the plans are.
Be it edit mode, undo, weight paint, hair edit performance, all these things are causing so much headache for many people, I’m sure the community could sponsor 1-2 developers dedicated to this. Count me in at least.
I do realize that having access to development plans and schedules is a perk for Blender folks that does not come cheap, but the developers have to forgive us, we’re all just human after all.
Toss my hat into the ring as to where we are at in terms of solving the performance issues, especially with Mesh Batch Cache is being worked on and with the BF now being able to have over 10 developers on the payroll. That is not to mention that the pre-release technical debt is becoming less of an issue.
In my opinion, the issue is serious enough that this could make or break Blender in the next few versions, thereby undoing all of the renewed interest around the industry when 2.8 was being developed. We must ensure that the next two or three releases will make strides in being able to handle the huge scenes that are rather commonplace these days.
The question to ask is this, does Ton still want Blender to bring true, professional-grade 3D to the masses? Would he be fine with letting Autodesk more or less own the field permanently?
Not to hog this thread, but my impressions are that Ton is very focused on movie making, and not at all focused on arch-vis or prod-viz (but that’s the reason that both Max and Maya still exists, they are aimed at two distinct industries… it’s a huge task for Blender to try to out-compete both of them).
But I agree with everything you said. Blender had a bad reputation due to its UI. Now it risks getting a bad reputation due to its performance. Hope this won’t ruin the momentum!
I’m left to wonder how the artists behind Spring even managed to animate the characters and model the scenes, surely there was a time where the artists wished things were snappier and more responsive.
You need high levels of performance for movie making too, artists should just say no to working on the Open Movies if they decide to not have it as a priority. In the meantime, we at least have Mano-wii working on improving performance with selection, and we’ll hopefully soon see Clement’s batch cache work.
