I’d be very interested to compare the power efficiency of this implementation versus the OS-native version of same over a 12-hour benchmark.