Making sense of the Sycamore's computing prowess - power consumption

Question

I came here after reading about the announcement regarding the Sycamore processor and Google's Quantum Supremacy claim. I am hung up on several key things and I am hoping that I could find those answers here, or at least direction to the answers.

First, I am trying to make sense of the comparison, i.e. Sycamore is a 53-qubit processor (54 but 1 is not functional). One the paper itself (doi: 10.1038/s41586-019-1666-5) figure caption 4,

... hereas an equal-fidelity classical sampling would take 10,000 years on a million cores, and verifying the fidelity would take millions of years

I assume this a million cores refers to IBM Summit at Oak Ridge, right?

My actual questions revolve around the power consumption. I assume it takes a lot of power to use that a million cores to simulate quantum state; how much power did the Sycamore consume to perform the same task? In other words, how much power did the Sycamore use along with its other auxiliary/peripheral devices for that span of 200 seconds vs. how much power theoretically Summit would use in the span of 2.5 days (IBM's counter-claim) if not 10,000 years (Google's claim)?

Second question with regard to power consumption, given power consumed as percent fraction, how much power does Sycamore consume, how much goes to the dilution refrigerator?

Thank you!

Answer 1

They say in Section X.H of the supplement that the Summit supercomputer has a power capacity of 14 megawatts. They compare that to their own setup. Their power consumption is mainly their dilution fridge, which they say is about 10 kilowatts plus about another 10 for chilled water for its supporting equipment. Their own supporting PCs and other electronics is about 3 kilowatts, they say. They give themselves a total power budget of 26 kilowatts, tops.

I would say that the comparison is so one-sided that a precise estimate seems a bit silly. Maybe you could reduce the estimate with free chilled water if the experiment were done in Alaska. Or maybe you could up the estimate by including the electricity to heat lunch for the coauthors.

---

Addendum: I guess there is a more serious answer based on the second question in the post, and related to my joke second paragraph. Aizan asks how much power Sycamore itself consumes, vs all of the other equipment around it. Indeed, most of the power consumption of the comparison Summit supercomputer in Oak Ridge is from the core activity of the computer itself, from the trillions of transistors and wires between them as they carry electrical current. Moreover, carrying current is unavoidable when a classical electronic computer changes state. Some of the power budget might go to air condition the computer and the data center to release all of the computer's heat to the outside. That is a serious extra power requirement that can be called peripheral, but it is not the main power cost.

However, all of the listed power budget for the Sycamore process is for peripheral equipment, especially but not only the dilution refrigerator. For several reasons, the power consumption of the Sycamore chip itself is negligible. One reason is that the Sycamore is only 53 qubits, which for this question is a lot like 53 bits. (How much power do you need for an old Z80 chip?) Another reason is more interesting: In a natural sense, a perfect quantum circuit always operates at zero temperature and never draws any power! Unitary quantum circuits are reversible, both in the sense of computation and in the sense of physics. In the sense of physics, that means that no heat is generated and no energy is wasted. This was indeed part of the original motivation for quantum computers in the old papers of Feynman, at a time when people could only vaguely guess that the same model might also lead to superior algorithms.