auto bucket size to ensure combined cpu cores render at the same speed as one GPU bucket
I notice that there’s a bit of a problem with hybrid rendering. If you have a lot of low speed cores, like with some xeons for example, then you’re inevitably going to waiting a very long time for the CPU buckets to complete once the GPU finishes it’s final bucket.
I think a good solution to this would be automatic variable bucket size, where the CPU bucket size is automatically calculated so that all cores combined equal the rendering speed of one GPU bucket. This way it won't matter if the large gpu bucket or the combination of small cpu buckets finish first, because they will both render at the same speed.
Perhaps also, once VRAM is depleted and Cycles needs to go out of core, at this point, sufficient cores should be removed from rendering so they can ensure data can be passed to the GPU fast enough to keep it rendering at full speed, at the moment the cpu bottlenecks the GPU when going out of core.
5 Comments
Hi 3D Illusions, Do you want to talk about thi's V-Ray feathure ? "Dynamic bucket splitting – When enabled, V-Ray automatically reduces the size of the render region as the rendering nears completion in order to use all available CPU cores." docs.chaosgroup.com/display/VRAY4MAX/System
* @jbbaron * I think my suggestion is better, as it ensures that it doesnt matter if the last bucket is multiple CPU cores or one GPU bucket, because both will render at the same speed.
Please do not linkbomb. Add text to describe the feature and how you would like to see it integrated in Blender.
* @Jason van Gumster * why?
* @3d illusions * Because this website is here for making well thought-out, actionable proposals that other users can discuss and developers can implement. Just posting a link doesn't give any context, explain why the feature would be useful, or describe how a user would interact with it.