Tweet
In doing some tests:
- reducing these values to a point will give back viewport performance but makes render times longer...I understand that.
- increasing these values help utilize GPUs better, causing viewport performance to become slow, but faster render times...I understand that.
- When invoking DR nodes, it was stated to increase these values, but it seemed when increased beyond a point, there was no faster render times AND I observed certain GPUs usage would go to 0% (on ws) while other GPUs on nodes would still be 100%.
- I've also observed reducing the values at a certain range, would oscillate the GPU usages more but would give faster overall render time when DR over a gigabit network.
Here is what I tried using 2x980ti on ws and 1x970 on node via DR
RBS 256
RPP 64
4:06s
RBS 256 (duplicate for comparison)
RPP 64
3:48s
RBS 256 (triplicate for comparison)
RPP 64
4:04s
RBS 32
RPP 1
9:23s
I did get very good viewport/monitor performance
RBS 256
RPP 80
4:08s
RBS 320 (I added 64)
RPP 80 (I added 16)
3:54s
RBS 384 (I added 64)
RPP 96 (I added 16)
3:52s
RBS 192
RPP 8
3:32s
So you can see when adding to these values past a certain point, no increase in render times, but going back from 256/64 to 192/8 gave faster render times but more GPU oscillation.
Am I hitting a bottleneck?
Should I run some initial tests when I make a new scene to see what values work best?
EDIT:
All the above tests were run using 3.5.01 beta, I thought I would switch back to 3.4.03 for kicks and the render times were longer, so whatever improvements were made in 3.5 are working well.
11:14 AM 12/14/2016 Switched back to 3.40.03 (non-beta)
RBS 64 (these were defaults after installing)
RPP 4 (these were defaults after installing)
Plate14.max
1280x720
RT GPU noise 0.05 2 cards on ws (2 x 980 ti)
5:16s
kernals were recomplied
lc built was about same time
11:22 AM 12/14/2016
RBS 256
RPP 64
Plate14.max
1280x720
RT GPU noise 0.05 2 cards on ws (2 x 980 ti)
6:38s
11:37 AM 12/14/2016
RBS 256
RPP 64
Plate14.max
1280x720
RT GPU noise 0.05 2 cards on ws (2 x 980 ti)
6:36s
11:46 AM 12/14/2016
RBS 256
RPP 64
Plate14.max
1280x720
RT GPU noise 0.05 2 cards on ws (2 x 980 ti) + DR EVGA 970 on Sys12
7:53s
11:57 AM 12/14/2016 (duplicate)
RBS 256
RPP 64
Plate14.max
1280x720
RT GPU noise 0.05 2 cards on ws (2 x 980 ti) + DR EVGA 970 on Sys12
7:21s
- reducing these values to a point will give back viewport performance but makes render times longer...I understand that.
- increasing these values help utilize GPUs better, causing viewport performance to become slow, but faster render times...I understand that.
- When invoking DR nodes, it was stated to increase these values, but it seemed when increased beyond a point, there was no faster render times AND I observed certain GPUs usage would go to 0% (on ws) while other GPUs on nodes would still be 100%.
- I've also observed reducing the values at a certain range, would oscillate the GPU usages more but would give faster overall render time when DR over a gigabit network.
Here is what I tried using 2x980ti on ws and 1x970 on node via DR
RBS 256
RPP 64
4:06s
RBS 256 (duplicate for comparison)
RPP 64
3:48s
RBS 256 (triplicate for comparison)
RPP 64
4:04s
RBS 32
RPP 1
9:23s
I did get very good viewport/monitor performance
RBS 256
RPP 80
4:08s
RBS 320 (I added 64)
RPP 80 (I added 16)
3:54s
RBS 384 (I added 64)
RPP 96 (I added 16)
3:52s
RBS 192
RPP 8
3:32s
So you can see when adding to these values past a certain point, no increase in render times, but going back from 256/64 to 192/8 gave faster render times but more GPU oscillation.
Am I hitting a bottleneck?
Should I run some initial tests when I make a new scene to see what values work best?
EDIT:
All the above tests were run using 3.5.01 beta, I thought I would switch back to 3.4.03 for kicks and the render times were longer, so whatever improvements were made in 3.5 are working well.
11:14 AM 12/14/2016 Switched back to 3.40.03 (non-beta)
RBS 64 (these were defaults after installing)
RPP 4 (these were defaults after installing)
Plate14.max
1280x720
RT GPU noise 0.05 2 cards on ws (2 x 980 ti)
5:16s
kernals were recomplied
lc built was about same time
11:22 AM 12/14/2016
RBS 256
RPP 64
Plate14.max
1280x720
RT GPU noise 0.05 2 cards on ws (2 x 980 ti)
6:38s
11:37 AM 12/14/2016
RBS 256
RPP 64
Plate14.max
1280x720
RT GPU noise 0.05 2 cards on ws (2 x 980 ti)
6:36s
11:46 AM 12/14/2016
RBS 256
RPP 64
Plate14.max
1280x720
RT GPU noise 0.05 2 cards on ws (2 x 980 ti) + DR EVGA 970 on Sys12
7:53s
11:57 AM 12/14/2016 (duplicate)
RBS 256
RPP 64
Plate14.max
1280x720
RT GPU noise 0.05 2 cards on ws (2 x 980 ti) + DR EVGA 970 on Sys12
7:21s
