The optimized implementation might use less power, though. With a slower algorithm, the CPU isn't the bottleneck, but it might still have more work to do.