The conventional view is that society has an insatiable appetite for energy. But that's not quite right; we have an insatiable appetite for the things that energy does for us. Nowhere is this more apparent than in the growth of the internet and data centers. Driven by growing demand for computation, multi-MegaWatt data centers, which barely existed a decade ago, today account for 2-5% of US energy consumption, according to the EPA (Koomey).
Since this is a first blog entry in this series, I'll start by introducing myself, Winston Saunders. My job in the Data Center Group at Intel involves keeping our data center energy efficiency initiatives on track. One reason I love my job is because the scope is immense; optimizing compute efficiency requires saving energy on time scales from nanoseconds to years. To put that in perspective, that spans almost seventeen orders-of-magnitude, about the same as the ratio of one second to the age of the Universe. Wow!!
So why does Data Center Efficiency Matter? To keep pace with the growth of the internet and data demand, compute needs to keep growing - and growing exponentially. What cannot grow exponentially is energy consumption. Already at 2-5% of US production, server and data center energy consumption cannot grow much more without incurring huge costs of additional generation capacity, not to mention risking regulatory scrutiny, carbon taxes, higher energy costs, etc. It’s in everyone’s interest to keep a lid on it. We must "Turn the Tide." The question is, how?
It's just a matter of some algebra to figure out the problems we need to solve. Normally we think of efficiency as the ratio of computations to energy. Since computations are the real deliverable, let's rewrite this as:
Computation = Efficiency * Energy
It's the same equation but an entirely different philosophy. If computation must continue to grow exponentially, and energy is constrained, our only lever is efficiency! Stated another way, to continue to deliver exponential growth of computation into the next decades, we MUST, from this perspective, drive evolutionary and revolutionary changes in our approach to Energy Efficiency.
Of course, the transistor scaling inherent in Moore's Law will continue to play a huge role here, but more and more, what we do architecturally within and beyond the CPU, and all the way up to the data center itself, will come into play.
What do you think? Can efficiency really help us to "turn the tide?"
In coming blogs I’ll talk about some perspectives on why I think it may be possible.