Blender 2.8 Viewport Performance


(zeauro) #211

It does not look like a major speed up when developer himself write in the commit that he don’t perceive the improvement and that change is valid for a point size of 30 that we don"t really know to what it corresponds.

It is hard to make improvement on something that was not tested a lot, yet. It is easy to have modifier preview doing weird thing by changing way to draw background from Theme to Viewport. or Color Type from Material to Single.
I obtained this.

But it looks like Flat shading is a bit faster than default Studio.
We can let them stabilize viewport features before asking for speed.


(noki paike) #212

What is this picture? I do not understand :slight_smile:


(Ace Dragon) #213

Good news for those concerned about editmode performance.

Brecht has opened a task which identifies the cause of the regressions and requests that they be fixed before the official release. Believe it or not, much of it is not related to the actual drawing code.
https://developer.blender.org/T57936


(Ace Dragon) #214

Clement just committed a big optimization for the workbench mode (when specular lighting is enabled).
https://lists.blender.org/pipermail/bf-blender-cvs/2018-November/116443.html

In his scene, the time needed to draw everything dropped from 1.52 milliseconds to 0.24 milliseconds.


(noki paike) #215

wowwwow
:heart_eyes::mage::dragon::troll:


(pitibonom) #217

hmmm interresting ^^

how to disable specular lighting ?
flat shading seems not as ‘flat’ as in 2.79


(ambi) #218

I wonder how long it takes for this technology to hit the mainstream:
https://devblogs.nvidia.com/introduction-turing-mesh-shaders/
http://reedbeta.com/blog/mesh-shader-possibilities/


(Hadriscus) #219

Wow, just tried using dyntopo in wireframe mode… it slows down to a crawl. :frowning: It’s alright though, I just wanted to see the subdivision size more clearly before starting to work, I’m not actually using wireframe mode to sculpt.


(staughost) #220

Yep still waiting for wireframe drawing optimizations to kick in. Or effectively modeling with meshes that have more than 100k polygons.


(Hadriscus) #221

I knew about modeling performance, but that’s edit mode performing poorly, isn’t it ? Always been the culprit as far as I know ? Sculpting however… I guess a mesh changing topology 30 times a second must be hard to make smooth.


(staughost) #222

Yes edit mode. But in 2.79 was much much better. Sculpting was fine (without wires tho).


(Most Hated) #223

For some reason, today, after I updated now my performance while trying to sculpt has become nearly unusable. It was fine before I updated and nothing changed on my PC between then and now. Anyone else having a similar issue?


(staughost) #224

It seems slower all over he place. In edit mode with high poly meshes its noticeably slower. Eevee starts firing up more slow than before.


(Piotr Adamowicz) #225

Edit mode performance has always been bad, yes, but not even remotely close to how awful it is now. With subsurfs in particular, it’s over 10 times slower.


(Ace Dragon) #226

It looks like 2.8’s drawing engine will become multithreaded soon, as Clement is doing commits to prepare for it.
https://lists.blender.org/pipermail/bf-blender-cvs/2019-January/118812.html
https://lists.blender.org/pipermail/bf-blender-cvs/2019-January/118814.html

Usually if something gets multithreaded, it means big performance boosts, here’s hoping :slight_smile:


(aermartin) #227

Awesome! Now, fingers crossed for multi-threaded shader compile in the near future for eevee.


(6eye) #228

Sounds great


(Alor) #229

Was 2.79’s multi threaded?


(Dzmitry) #230

Considering how ancient 2.79 view port and technology behind it is… no


(Ace Dragon) #231

Blender 2.79’s viewport code has its origins all the way back in the 1990’s. Most consumer PC’s back then only had a single thread (so multithreading would’ve made no sense).