Rather than a wholesale shift from classical to quantum computing — as when internal combustion engines replaced steam power — the future of supercomputing is likely to involve hybrid strategies, with regular digital computers augmented by other more fancy kinds: quantum computers, graphics processors like the kind that run video games, and neuromorphic machines that mimic the behaviour of the human brain.
“It’s a perfect world for D-Wave,” said Jeff Nichols, associate laboratory director of computing and computational sciences at Oak Ridge National Laboratory in Tennessee.
“I do not believe that you’ll ever replace all of traditional, classical computing with a quantum computer, nor will any of the other more exotic approaches replace classical computers,” he said. “You’re not going to carry around a quantum computer as your phone.”
Burnaby, B.C.-based D-Wave Systems’ new deal to provide quantum computing power to accelerate Oak Ridge’s supercomputers also marks a key strategy in the U.S. effort to catch up to China, which has invested heavily in its push to build computers fast enough to reach the exascale, or a quintillion calculations per second. (A quintillion is a billion billion, or 1 with 18 zeroes.)
A major difference in the two countries’ strategies has to do with the massive energy costs of running such a fast computer, which in the case of Oak Ridge’s Titan machine is nine megawatts at its peak, at a cost of $9 million.
China has tried to start with the necessary hardware, then bring the energy usage and costs down. But Nichols said Oak Ridge is taking the opposite approach with the strategic placement of quantum “accelerators” to improve the efficiency of calculation.
In future, he said they might wish to have a quantum computer on site, “tightly coupled” to their supercomputer but, for the moment, D-Wave’s service will be provided remotely, over the internet from Canada.
The prize of an exascale computer, for both China and the U.S., would be a vastly improved ability to solve some of science’s most complex problems, such as those about climate change, genetic analysis, protein folding, earthquake prediction, the performance of the electricity grid, and cosmology — problems that are too big to simply calculate by running through all the possibilities.
Many of these are what mathematicians call “optimization problems,” and these are what D-Wave’s quantum computer is best suited to solve, as they recently did, for example, in a study for Volkswagen about how to optimize traffic flow in Beijing.
The classic example of an optimization problem is of a travelling salesman who needs to visit many towns and wants to know the shortest route.
I do not believe that you will ever replace all of traditional, classical computing with a quantum computer
Classical computers, the kind made with silicon chips, would just calculate each trip and choose the shortest. But in this kind of problem, the number of possibilities soon grows impossibly large. To solve it classically, you would need to be calculating forever with a computer as big as the universe. To quantum computers, however, the math and logic of optimization problems look very different. They do not compute with the strict ones and zeroes of binary code, but rather with the strange quantum properties of superposition and entanglement.
A classical computer calculates with bits, which can be set two ways: one or zero. From this basic binary system, a computer can build up to all the complexities of modern computing.
A quantum computer, however, takes advantage of the strange properties of matter at the subatomic scale. Rather than bits, it calculates with qubits, or quantum bits, which are tiny, fragile physical systems — sometimes etched into a chip of metal cooled to near absolute zero, or a gas held in place by a magnetic field, or a sliver of artificial diamond — that can be in multiple quantum states at the same time, known as superposition. This property allows a qubit to be either one, zero, or a little bit of both at the same time, allowing for a whole new style of logic and computation.
D-Wave’s device uses a strategy known as quantum annealing to solve optimization problems not by brute calculation, but by exploiting quantum effects to find the likeliest candidates for solutions.
Using this style of computing to help a supercomputer skip unnecessary calculations helps Oak Ridge to keep its power costs down, while accelerating its performance, Nichols said.
The success of this approach is a key reason that D-Wave president Bo Ewald thinks the future of quantum computing will look different than the rapid expansion and constant improvement of classical computers since the mid-20th century.
He has a long history in top-level computation, for example at Los Alamos National Laboratory and as president of Cray Research, which once made supercomputers that filled a room, cost millions, and are now more or less matched by an off-the-shelf laptop.
“I always knew that because of Moore’s Law, things were going to get faster and shrink,” he said. (Moore’s law says the number of transistors in a computer chip doubles every two years, and it has held true for decades.) “But in quantum, you’re using more specialized materials, very cold superconducting materials in extreme vacuum, shielded from radio frequency.”
It is a more finicky hardware, he said, so he is skeptical that we will all carry quantum computers in our pockets in the future, as we do now with classical computers such as iPhones.
“So I think there’s a little more challenge to think we’ll have portable quantum computers,” he said. “I don’t think we’ll need them because I think they’ll be ubiquitous because of the Cloud.”