The bucket size is sometimes an undervalued factor in render times. The more 'seems' need to be calculated the slower it gets, but with multiple cores or DR you have a better utilization of each core, so it's a matter of finding a sweet-spot.
The problem I have with DR (as I type I'm rendering with 12 cores/buckets and I'm sure people use a lot more) is that sometimes a single core on one of the machines is stuck rendering the last bucket and all the other cores are idle, waiting. So my question is:
Would it be possible to have an adaptive render region size?
The render would start with a large render region and as it would approach a number of render buckets left to compute, equal to the number of cores/processes used, it would subdivide the remaining buckets, to minimize the effect of idle cores waiting for the last render region to be calculated.
It's not a huge problem, but could make the render quicker by a couple of percent. I think making the render region allocation "more intelligent" would be especially beneficial in cases where one has a lot of processes running (using DR).
Below is a simplified case of an adaptive region allocation with 4 cores.
The problem I have with DR (as I type I'm rendering with 12 cores/buckets and I'm sure people use a lot more) is that sometimes a single core on one of the machines is stuck rendering the last bucket and all the other cores are idle, waiting. So my question is:
Would it be possible to have an adaptive render region size?
The render would start with a large render region and as it would approach a number of render buckets left to compute, equal to the number of cores/processes used, it would subdivide the remaining buckets, to minimize the effect of idle cores waiting for the last render region to be calculated.
It's not a huge problem, but could make the render quicker by a couple of percent. I think making the render region allocation "more intelligent" would be especially beneficial in cases where one has a lot of processes running (using DR).
Below is a simplified case of an adaptive region allocation with 4 cores.
Comment