Hi!
I've done some tests and i'm sharing them hoping they are useful. I can't share the blend file as is internal work from my job. I have another old machine of the same specs but i could't make it work with CR yet. I repeated the test with the gpu at the end to take into account the balance between machines that CR calculated from the previous renders.
Crowdrender version 0.13 pre-release.
Single frame render.
f12 means rendered only in the main machine.
cr means rendered with the Crowdrender button.
When rendered in CR i've made 3 renders in a row to take into account the balance calculations.
Render time in seconds.
f12 gpu 212x216 - 40.71
cr gpu 212x216 - 49.74 - 44.14 - 42.92
f12 gpu 32x32 - 1:26.06
cr gpu 32x32 - 1:24.39 - 1.20.64 - 1.17.67
f12 cpu 32x32 - 1:21.44
cr cpu 32x32 - 1:18.14 - 1:17.88 - 1:17.68
f12 gpu 212x216 - 40.81
cr gpu 212x216 - 43.68 - 42.85 - 41.63
Main machine:
CPU Intel core i5-4460, 3.20GHz 4 cores (no hyper threading)
RAM 8GB DDR3
GPU Nvidia Quadro K620, 384 CUDA cores, Driver 376.33
Windows 7
Blender 2.79 with CR add-on
Slave machine:
CPU Intel Core 2 6400, 2.13GHz 2 cores (no hyper threading)
RAM 2GB DDR3, 8GB swap on a disk HDD
GPU GForce 7300 GT (no CUDA compatible, not used for rendering)
Manjaro Linux 7.1.3 (last version rolling release)
Blender 2.79 with CR in the terminal as instructed in this post: https://www.crowd-render.com/forum-1/general-questions/crowdrender-headless-comand-on-slave-computers
The slave machine is quite old and for now it's not enough to reduce render times in comparison with just the main machine gpu. Maybe if i get the second one working i'll get lower render times :-)
Really happy to help!
Here are more tests:
Crowdrender version 0.13 release (updated from pre-release)
Single frame render.
f12 means rendered only in the main machine.
cr means rendered with the Crowdrender button.
When rendered in CR i've made several renders in a row to take into account the balance calculations.
Render time in seconds.
From the previous test I updated Blender to 2.79a and CR on every machine.
With 0.13 final the second slave machine connects! Still have to start blender in terminal, no luck connecting to blender UI with the add-on in the Manjaro Linux machines. So the setup is:
Main machine:
CPU Intel core i5-4460, 3.20GHz 4 cores (no hyper threading)
RAM 8GB DDR3
GPU Nvidia Quadro K620, 384 CUDA cores, Driver 376.33
Windows 7
Blender 2.79a (normally started with the GUI) with CR add-on
And Blender 2.79 daily for some tests
Slave machine (x2):
CPU Intel Core 2 6400, 2.13GHz 2 cores (no hyper threading)
RAM 2GB DDR3, 8GB swap on a disk HDD
GPU GForce 7300 GT and GPU GForce 7300 GTX (one on each machine) (no CUDA compatible, not used for rendering)
Manjaro Linux 7.1.4 (last version rolling release)
Blender 2.79a with CR in the terminal as instructed in this post: https://www.crowd-render.com/forum-1/general-questions/crowdrender-headless-comand-on-slave-computers
Internal scene from my job (with denoiser):
F12 - 3:06:06
CR - 4:57:74 - 5:01:29 - 4:45:75 - 4:53:03 - 4:01:54 - 3:34:86 - 3:19:93
-> Blender 2.79 daily-build with GPU+CPU (Only on the main machine, Slaves still with 2.79a):
F12 - 2:17:95
BMW test scene, with 100 samples, didn't touch anything else (note that there's some CR balance calculated from the start from the previous renders... Is there?):
F12 - 5:26:32
CR - 4:55:02 - 4:40:62 - 4:08:47 - 4:02:05
-> Blender 2.79 daily-build with GPU+CPU (Only on the main machine, Slaves still with 2.79a):
F12 - 2:55:28
CR - 3:15:49 - 4:56:33 - 6:23:74 - 9:07:68
Some concluisons:
My main computer benefits on most scenes from using the GPU+CPU feature. When using the daily build with this feature CR doesn't manage to improve render times and after each render unbalances the node setup increasing render times further.
It did improve render times for the BMW scene in stable blender 2.79a, good for CR!, but it's still faster for me to use Blender daily with the GPU+CPU feature without CR.
It seems that CR has some trouble before finally calculating a good balance between my nodes, it' would be great to set the balance manually for some cases like mine. Knowing my machines I would start setting a 80/10/10 work distribution.
When i get some spare time I will test the main machine with my personal laptop wich has a close performance to the main one at my job.
Hi, thanks for testing! Please do try again with another machine, its important we hear how other users go with testing, I love all the detail you gave, hope you test again soon:)