Page MenuHome

2.8 test
Closed, ArchivedPublic


A simple test done on one machine so it’s an apple for apples test which makes hardware specs pointless, a basic cube was rendered.
2.79b – cycles = less than 1 second not enough time to see Gpu usage
2,8 – cycles = 50% to 65% Gpu 9min 44sec Eevee = 7.39 seconds Gpu 100%



Event Timeline

Philipp Oeser (lichtwerk) triaged this task as Needs Information from User priority.

Dont think hardware specs are pointless here.

Could you provide information as requested here

  • Operating system:
  • Graphics card including driver version:
  • Blender version used (hash from splash screen)

Win 10 Home
GeForce 840M 418.81

Quick note on this machine, it is about 5 years old and Blender has run as smooth as can be on it over the years, absolutely no issues whatsoever.
I have just started testing 2.8 so will let you know what I find, there is one thing I would like to tag on this as it may not be a bug so do not want to create a new action.
The laptop that I am using for the test was originally Win 8, then 8.1 and now 10, not once have I installed Python but when I tried to install 2.8 on my tower I get a no Python installed error and have to close the program.
Should Python not be included in the download as it is for previous versions, I did not get the error on the laptop, only on the tower.

Jacques Lucke (JacquesLucke) raised the priority of this task from Needs Information from User to Needs Triage by Developer.Mar 7 2019, 3:48 PM
Sebastian Parborg (zeddb) triaged this task as Needs Information from User priority.

Is this still an issue?

I’ve just downloaded and done a fresh 2.8 install and the same test recorded these results.

Cycles CPU 1m 07sec, Cycles GPU 1m 06sec which is strange as this is a sad CPU so the GPU should destroy it, Eevee less than 1 second.

Fantastic improvement, excellent work and thanks to all those involved.

Not completely, I’m still getting an 18 second longer cycles render time with 2.8 when tested side by side against 2.79b.
It has improved dramatically though.

When comparing render times, please attach a .blend file created in 2.79b. Then render that same .blend file in 2.79b and 2.80, and tell us the render time for CPU and GPU.

There have been quite a few changes to defaults. Also be sure to that Cycles is not set to render CPU + GPU in 2.80, which does not exist in 2.79b.

Craig (Cx2) added a comment.EditedJun 12 2019, 10:46 AM

I used a basic cube without any materials or lighting effects to simplify the test, I checked that Blender is using the GPU during the tests, should we not be seeing better GPU times?

5 renders were done and these are the average times.

2.79b Cycles CPU 12.4sec Cycles GPU 12.3sec
2.8 Cycles CPU 10.4sec Cycles GPU 9.9sec

Brecht Van Lommel (brecht) claimed this task.

Those render times are better in 2.80, so I see no bug then.

With a laptop GPU like the 840M, the render times are not always significantly faster.

Sorry but I think I should have done a better job; Blender is not fully using the GPU in both versions and that’s why the CPU and GPU times are so close.
The reason I said it was is because I looked at the Nvidia icon in the taskbar tray that says it is but when I look at the performance monitor it only indicates 3% GPU usage on both 2.79 and 2.8.
The reason I have not picked up on this before is because I do not use this machine for rendering, I use my tower which has a GTX card.

The GPU usage in the performance monitor does not count the kind of GPU compute usage that Cycles does.