2.93 LTS suggestion; No new modeling/rigging tools, make editmode/OSD fast instead

As the title says, I believe that if the BF was to focus on performance for edit-mode and OpenSubDiv for 2.93 (instead of adding new features to those areas), the net benefit for users over the next year. especially for LTS users, will be far higher than if new tools and tool options were created instead.

We have enough tools for now, but we can’t use them to the fullest if we can’t work with a huge number of polygons. We also can’t take full advantage of rigging tools if it is not possible to have subsurf active during animation. Also think of the people working on your new open movie, do you think they would benefit greatly if performance was a bit better and they did not have to resort to workarounds (like cutting objects into little pieces)? The bottlenecks are known, now help us have access to a Blender that can plow through scene creation at 60 frames a second.

Please get this done, as it is one of the few major sticking points left in the whole application.

Thanks.

19 Likes
I would like to agree with a text above. In my field (characters design) i always face performance problems from mid/end of the project where scene gets a bit more heavier.
Blender (just my opinion) has already very strong toolset to create things, very stable, plus great addons.
But weak subdiv( slow to edit and in animation) and undo performance( object-edit mode- object) problems makes it very hard to get job done in the industry.
Thanks!
My Video3|video
13 Likes

Slow editing is also problem with large imported CAD models which have to be modified regularily for various reasons.
Better editing performance would immensly help Product Viz, Automotive Viz and similar disciplines.

12 Likes

I absolutely second that, but my intuition is the BF lacks of developers in these areas and/or these are huge tasks that involve to re-design (and possibly break) a whole working system, that would take ages to do. It’s more high-end than interface development or grease pencil, many developers simply aren’t cut out for that.

One of Blender 2.8 targets (1 year and a half ago!) was new open subdiv implementation, and the benefits never made it through. Currently Blender 2.79 subsurf performances are better. While I regularly donate important amounts to the BF (donated more than 10k over several years), and let’s admit it they make fantastic work on other features, I feel baffled.

9 Likes

@dfelinto @ideasman42 @sergey
Can you please join the discussion ? The community is asking for better performance for edit mode and subdiv, and there’s no dev actively working on that side of blender.

3 Likes

Yes, we have stopped using Blender at work partly for that reason. Having a large CAD model with lots of instances makes Blender crawl while other software* doesn’t bat an eyelid.

*I say “other software” now, because this is nothing new and I guess it’s only a matter of time until the videos in that thread gets erased… then again, developers never read feedback anyway, so maybe it’ll stay. :wink:

3 Likes

Hi,

There is indeed no active developer working in OpenSubdiv performance for edit mode in Blender. There are other strategical projects for this year that are priorities at the moment for the core development team.

The community is as always encouraged to pick up on any of the other old targets, as it is already happening for the USD importer.


More details:

For the previous year the development team tried to prioritize 10 different projects to be tackled mostly by 10 individual developers. For those unfamiliar with the list, you can read about them here.

The result was a bit mixed. In some cases it went well and everything was achieved. In some cases (like the edit mesh performance) the main issue was tackled (getting 2.7x level of performance) although some todos were left unfinished (e.g., OSD and GPU transfer overhead).

It was a unique year in many ways. But besides that, the way the projects were structured didn’t turn out to be the most efficient.

For 2021 instead of focusing on multiple projects the team decided to try projects runned by teams and an even stronger user collaboration. You can learn more about this approach here. Whether or not this will prove to be more effective, only time will tell. But the geometry nodes is a good start already in this direction.

In this initial quarter of 2021, library overrides, asset browser and geometry nodes are being prioritized. All of them original targets from 2020. They are also specific projects that can be developed in close collaboration with the Blender Studio, with a clear focus in professional production quality tools.

Given the current size of the Blender core team I believe that this will allow the development team to tackle more of those long standing issues. And to pick them up based on priority, the need in productions the risks of the projects, …

14 Likes

Key word “Long standing issues”. Outside of Everything Nodes which is important to have for a modern workflow (not to mention the fact it will replace the dusty old particle system), Blender is now at the point where it has nearly everything most users could wish for as far as features go. As a result, I believe it is time to switch gears toward tackling weak points like performance, polish, and of course the current projects in resolving shortfalls in workflow and in scene management (such as the concept of reusing content).

6 Likes

@dfelinto
The BF certainly has not to justify for every decisions taken, you have made great work so far, no one can argue that and I feel profoundly thankful for what Blender has become over time.

This said, the BF should be more concerned with the missed goals of Blender 2.8. Many users who regularly donate, and those especially who donated for the “Code Quest” can feel fooled/used. The “Code Quest” was promising many shiny features, including open subdiv performances, that never made it years after the release of Blender 2.8. There were roadmaps with schedules, showing than OSD performances were supposed to be tackled for Blender 2.9, 2.91, it never happened and won’t happen anytime soon according to you.
Hearing that “there is indeed no active developer working in OpenSubdiv performance or edit mode”, that developers have other work to do, and that the community can do it instead if they’re not happy, is disrespectful toward users who donated for the promised features. Not saying that to you particularly of course, you are not responsible. It’s a general remark.

8 Likes

Add pose mode, I’m working on a feature film and a short film either of them not high end, with the classic stuff one environment 2-3 characters that are fairly simple but elaborated and there is no way to get the scene running above 12 fps, the only way is to get one character alone and get rid of the environment.
Even moving keyframes in the timeline and dope sheet become impossible, not even workable with the graph editor.
The key thing I want in blender that is missing for me is performance in pose mode, to be able to pose and animate Realtime is the single most important missing thing.

9 Likes

Weight painting or the node editor in general as well. Especially with geo nodes it’s concerning that even selecting/moving a node becomes laggy in larger scenes.

It’s good to see the quarterly plans though, helps to set expectations and try planning ahead for productions.

I’ve been saying the same since 2.8.
(I mean basically " No new modeling/rigging tools, make editmode/OSD fast instead")

Hi, I agree these projects are important.

As Blender grows, the overhead of running the project grows too, we need to be able to scale better, that’s easier said than done.

The issue I run into these days is it’s now reasonably difficult to get focused time on single projects (even a week of focused time feels like a luxury).
This could be an excuse for failing on any project though, of course I have a responsibility to finish projects, even so - to complete some of these tasks I would have had to take time away from other areas and focus which (for the most part didn’t happen).

There isn’t one thing I can point to as the reason, code-review takes up quite a lot of time, there are also weeks we focus on fixes, issues for the release, investigating high priority bugs in modules, going over and reviewing issues in modules I’m involved with, investigating when/why a test broke… etc.

Recently another post here suggested we stop everything and focus on reviewing code.

I don’t think stopping everything (at least for extended periods of time) is a very good solution though as it’s likely to be disruptive - holding up other areas of development that are also important.

Personally, I dislike that Blender as a project tends to over-promise & under deliver. We should be more realistic about our plans.

Having said this, it’s important to communicate our goals, being too wary of failing in public might limit our goals too much.
That we don’t always succeed in every goal doesn’t mean we’re abject failures either.


As Dalai mentioned - having teams form to work on projects is something we’re trying out,
at some point we could look into having a team focus on modeling performance, as it remains an important project that shouldn’t be overlooked.

23 Likes

I donate mysefl and don’t plan to stop (unless I loose my Job :D) since obviously the more money, the more faster those issues would be adress (I’m sure they will eventually). And this is just an awesome project.

I wonder if some sort of “bounty” system could be implemented to hire dev who know Blender well and tackle those issues.

Financing goal could be implemented that would approximate fund needed to resolve an issue (edit & undo performance) or implement a feature (light linking, shadow catcher).

Then if a dev is interested somewhere, he would get paid once most of the code is commit.

I’m sure very talented/clever mind would be willing to have a “side job” for a few months knowing they can get paid and help BF.

So more “code-quest” could be made once a year or two to help resolve those long standing and important issues that I read over and over on Devtalk and Twitter…

1 Like

This comes up periodically. I made an rcs proposal a while ago, some valid points have been brought up for and against it, but I don’t think this will ever happen.

I’d be all for it though, anything that gets me away from subscriptions.

I would not see it as a replacement of the current Devfund since BF need stable income, but more of a periodic/if necessary “code-quest” to resolve specific issue/feature.

Thx for the RCS link, indeed this is more easier said than done.

I had also proposed something similar but since I think about it…I think this kind of financing would work better as a periodic thing to help resolve specific and long standing issue as to ensure people doesn’t leave monthly donation.

In any case, I don’t want to derail this topic to a crowdfounding campaign discussion…

Just trying to propose/find solution here :slight_smile:

edit : @dan2 I read through your RCS, there is a lot of very good ideas I think and valid point from both side of view. But mostly I agree with you that this could work…

As you said… A financing system could be implemented inside RCS to crowdfund a proposal… And many others things.

1 Like

A post was merged into an existing topic: A Feature Quest Proposal

About Instance editing performance, nothing to say


System Information: GTX 1650 SUPER, 16Gb ram, 6 core 3.5Ghz. Not the good one, but the scene is not so heavy…

And it’s only 3 instances and 1 master.

3 Likes

I don’t understand exactly what are these modeling improvements to stop doing? If this module is almost abandoned for five years, there is no project on modeling. It receives updates from time to time but many times external.

As I say there is no project (it doesn’t look like it to me), there is a list of things that would be cool to do but there is no clear project with resources based on the experience of modeling users.

if you look at the modeling workboard and you find discussions from two years ago about changing a default parameter in a command… This is to have abandoned and aimless blender modeling.

1 Like

The commit made last week…
https://developer.blender.org/rB196dfc01a3e99c3bef0e44acf599bca50ae0300e

Was technically a fix, but it was a sign of hope for those of us waiting to see if and when editmode performance becomes a higher priority. Let us not stop there though, let us go beyond fixes and bring in the long awaited optimizations that we know can be done, but have not yet been done.

1 Like