I don't believe this has anything to do with processing on the GPU vs CPU unless I'm missing something. That particular property is supposedly just related to which scheduler the UI and RenderThreads use. It changes those to run on SCHED_FIFO instead of SCHED_OTHER.
Supposedly this property can make the UI more smooth (removes jitter), but Google says to only use this property if the RT scheduler is capacity-aware:
However, because the RT load balancer was not capacity-aware, there was a 30% reduction in application startup performance because the UI thread responsible for starting up the app would be moved from a 2.1Ghz gold Kryo core to a 1.5GHz silver Kryo core. With a capacity-aware RT load balancer, we see equivalent performance in bulk operations and a 10-15% reduction in 95th and 99th percentile frame times in many of our UI benchmarks.
The end result may be that the UI is a bit smoother (microseconds vs milliseconds), but that anything not running in the foreground may be unable to use the 1.5GHz cores in this big.LITTLE AML architecture on my T95z plus and may take much longer to complete. I suspect this issue would be limited specifically to systems that have cores running at different speeds such as the 4 1.5GHz cores and the 4 1GHz cores.
Thoughts?
Comment