Topic: Experimental build performance

Hi.
I just tried to porting ManualUse example into experimental branch, and bullet too as dependency.
I think it's quiet good as it is an early build. Here is my performance result :

Experimental (800x600)
* vsync on:
  - fps: 60
  - frame time: 16.6667ms
  - cpu usage: 13.3% - 14.6%

* vsync off:
  - fps: 561 - 572
  - frame time: 1.7482ms - 1.7825ms
  - cpu usage: 39.5% - 41.5%

Trunk (original 800x400)
* vsync on:
  - cpu usage: 17.6% - 22.2%
    * unstable, first start with 92.3% - 97.5%, 10 secs later decline gradually until quiet stable, sometimes can rise again up to 30%+.

* vsync off:
  - cpu usage: 95.0% - 97.1%

Trunk (modified - draw every frame 800x600)
* vsync on:
  - fps: 60
  - frame time: 16.6667ms
  - cpu usage: 19.3% - 20.3%

* vsync off:
  - fps: 508 - 531
  - frame time: 1.8832ms - 1.9685ms
  - cpu usage: 63.5% - 64.2%

My laptop specs:
- Intel Core i3-2330M @ 2.2Ghz
- Intel HD Graphics 3000
- Ubuntu 14.04 LTS x86_64
- Mesa 10.3 (Intel Open Source)

The ManualUse experimental test and bullet libs are just quick porting, I do not optimize or check for memory leaking, or etc that may be missing and cause an error. Hence this performance result may be wrong.

CPU usage unstable at the original version, I don't know why, or it's just me.

The changes at modified trunk version:
- resolution 800x600
- fps counter
- comment "don't wait too much" block
- draw every update (comment "if (steps > 0)" line)

I turn on and off vsync using "vblank_mode=0"

Looking forward experimental branch to be main branch and stable big_smile
Good luck devs smile

Thanks

Last edited by jurgel (2014-12-09 15:48:26)