Re: OpenCL/C++ hybrid rendering
Posted: Thu Feb 15, 2018 1:48 pm
niiice indeed 
& here's a thought that is lately occupying my mind...
Could it be possible to manage units/devices so that they render per frame (for example as Farms work)?
For now it's possible to do it manually as one-by-one (1scene : 1blender app) rendering - at the same time:
1.frame/CPU1 + 2.frame/GPU1 + 3.frame/GPU2 + ...
One could even go further splitting rendering work (frames) for CPU by core (but haven't tried this yet
).
The advantage is mostly prominent with animations of advanced (more complex) scenes:
+ more secure - if something goes wrong, not everything is wrong (unless the whole system collapses)
+ better time/resource prediction
+ less computing time per frame (compiling only for one unit/frame)
+ still enough computing resources available (at least one core) to continue working on other areas (post, composite, tonne mapping, checking, debugging, communications...)
+ after getting used to it, workflow/thought flow gets faster into the zone (less interrupted)
Have already done this, but manually & only with Cycles. On a single machine - otherwise i would need to have 3.
As time was of the essence, this workflow made it possible for me to deliver on time & some extra.
& here's a thought that is lately occupying my mind...
Could it be possible to manage units/devices so that they render per frame (for example as Farms work)?
For now it's possible to do it manually as one-by-one (1scene : 1blender app) rendering - at the same time:
1.frame/CPU1 + 2.frame/GPU1 + 3.frame/GPU2 + ...
One could even go further splitting rendering work (frames) for CPU by core (but haven't tried this yet
). The advantage is mostly prominent with animations of advanced (more complex) scenes:
+ more secure - if something goes wrong, not everything is wrong (unless the whole system collapses)
+ better time/resource prediction
+ less computing time per frame (compiling only for one unit/frame)
+ still enough computing resources available (at least one core) to continue working on other areas (post, composite, tonne mapping, checking, debugging, communications...)
+ after getting used to it, workflow/thought flow gets faster into the zone (less interrupted)
Have already done this, but manually & only with Cycles. On a single machine - otherwise i would need to have 3.
As time was of the essence, this workflow made it possible for me to deliver on time & some extra.