Blender 2.8 Viewport Performance


(zeauro) #91

Sculpt mode needs to be responsive.
Light cache of a probe takes some time to be computed. An update would start at each brushstroke.
EEVEE quality can be pushed far enough to expect this kind of attempt to be unusable.

Dyntopo works with EEVEE. But in order to be dynamic, dyntopo ignores materials. So, you can only sculpt grey objects in EEVEE.

But we can expect a usable workflow where we could make several sculpt strokes with workbench solid display and quickly switch to EEVEE, wait 1 or 2 second and check EEVEE result ; and then go back to solid display to continue sculpting session.

Currently, you have to go back to object mode to see updates of geometry+material.
But I doubt that would stay as is.


(Pitiwazou) #92

Working with sss is really great, I do it on zb and it helps a lot IMO.

So, even if we cannot use all the eevee features, I hope we will have some kind of SSS and AO even if it’s not as good as eevee.


(Thornydre) #93

I don’t really know if it fits in this thread, but someone is developing a vertex gradient for proportional editing, and this is great, and very useful, hope this will be soon in Blender :


(Jason van Gumster) split this topic #94

A post was split to a new topic: World Backgroud


(Hadriscus) #95

Oh yes. If that ends up in master I will sell my horse. I’ve been wanting that for so long !


(Thinking Polygons) #96

The best way to feel the performance improvements is by running blender on old and unsupported hardware… hahaha with opengl 2.0 :ok_hand: :wink:


(so3Datel) #97

I opened my scene with animation and a lot modifiers in 2.79. Then press play animation, and I have 30 fps.
I opened my scene in 2.8, and play animation, and I have 2-3 fps. Then I turning off all modifiers visibility in viewport and fps up to 26. Then I apply all modifiers, I selected all objects and converted they to mesh (Alt+C > Convert to mesh). Then My fps increase up to 35 fps.
In 2.8 modifiers very much reduced performance.
sorry for my English


(Hadriscus) #98

What type were they ? Curves ?


(so3Datel) #99

Sorry, I mean that, I apply all modifiers through Alt+C > Convert to mesh


(so3Datel) #100

Same scene. Played animation. Mesh without modifiers.
2.79 Wireframe mode - 30fps
2.8 Wireframe mode (xray =0, wireframes =1) - 17fps
2.8 Solid + Wireframes - 17fps
2.8 Solid - 30fps
and
2.8 Modifiers + Wireframe - 1.5 fps
2.79 Modifiers + Wireframe - 30fps

I can summarize. At the moment 2.8 slower if you use modifiers and wireframe


(Hadriscus) #101

Well that is not a good sign… the benchmark is a very good idea !
On the other hand, since the Spring team is -supposedly- using it in production right now, they probably noticed this slowdown on their own, so I imagine that’s being worked on.


(cekuhnen) #102

this makes to a degree sense - more visual effects added = more to process => slower performance


(Hadriscus) #103

He means slower than 2.7x - although you’re right, adding overlays will inevitably bog the thing down a bit, I imagine.


(staughost) #104

Have tested alpha build. It still has massive performance issues with 100k meshes and above. Is there any news about performance fixes?

For some reason I have a feeling that lot’s of artists don’t see problem with performance because they use 2.8 for viewing not modeling (at least not for High poly HSM modeling).


(Ace Dragon) #105

Until now, the reason why there’s not much modeling and scene assembly in 2.8 is because it’s just been too unstable with too many bugs to do so.

Now we are seeing a shift towards smaller tasks, stability, and bugfixing. The calls to improve performance for important tasks will increase very soon.


(• I don't speak English "by default", so... ) #106

Most of the viewport performance issues in 2.8 are coming from the overlays, and that may not be an easy fix… Other than that, I was able to achieve a lot more smoother strokes in 2.8 when sculpting on hi rez meshes, than 2.79.
So :sunglasses:


(Piotr Adamowicz) #107

I’ll just quote this tweet by Emil Persson, senior graphics engineer at Epic:

Rules of optimization:

  1. Design for performance from day 1
  2. Profile often
  3. Be vigilant on performance regressions
  4. Understand the data
  5. Understand the HW
  6. Help the compiler
  7. Verify your assumptions
  8. Performance is everyone’s responsibility

(Dantus) #108

Have the coders made obvious mistakes from your point of view which contradict this list?


(Piotr Adamowicz) #109

I don’t know. I’m no expert. Nor do I know the code in question. I just like the tweet :slightly_smiling_face:

He even wrote a short article where he expands on this.


(Dantus) #110

There is of course some truth to it, but even calling them rules is too far fetched for my taste. That’s the kind of bullshit I often have to waste time with, because I have to explain to customers or project leads that just because it is catchy does not mean you should apply it as written and it does often not even make sense for your use case. Most people who need to care about this sort of stuff already know it! For the ones who don’t know it yet, it is way too oversimplified. And everyone who does not need to know it is mislead.
Except for a little entertainment with a catchy thing, there is very little value to it.