Considering DFVS again, there are (were?) two potential strategies (Aesop: Tortoise v Hare):
To compute quickly and halt we need a higher frequency clock but consume the same number of active cycles.
So the work-rate product, af, unchanged, so no power difference ?
No. Running the same number of work cycles at a lower frequency requires a lower voltage and hence we save energy according to V^2.
But: current geometries only have a narrow operating voltage range
Too low -> too much leakage.
Too high -> tunnelling and wear and heat.
So today, we operate at around one volt always and the trade off is just between high and low leakage technology - a static fab-time decision.
There are further (unexaminable) prospects on the table, such as dynamic body bias ...
35: (C) 2012-18, DJ Greaves, University of Cambridge, Computer Laboratory. |