QUANTUM computers are a grand idea. By harnessing the famous strangeness of quantum mechanics, they should be able to perform some (though not all) calculations far faster than any ordinary computer. But building one has proven tricky. The idea was first floated in the 1970s. Four decades later quantum computers are still small, fragile devices confined to the laboratory bench—with one exception. In 2011, to a great fanfare, a Canadian firm called D-Wave announced a commercially available quantum computer, the $10m D-Wave One. Deals with Google, NASA and Lockheed Martin, a weapons firm, followed.
Admittedly, D-Wave’s device is a very specialised sort of computer, restricted to a single area of mathematics called discrete optimisation. But it was big news, and many scientists were rather sceptical. In the past couple of years the firm has published enough papers about its device to convince academics that it has indeed built a quantum-mechanical machine. Now the question is whether it is any faster than the competition.
One set of benchmarks, published in May, suggested that it was: they found that D-Wave’s machine was up to 3,600 times faster than its non-quantum rivals. But it was not quite a fair fight, for the classical machines were running off-the-shelf programs. With a bit of clever tweaking, they could have been made much faster. When a group led by Matthias Troyer, a computational physicist at the Swiss Federal Institute of Technology, tried such souped-up algorithms, they found that D-Wave’s lead vanished. D-Wave, in turn, pointed out that the researchers had been using an old version of its system.
Now Dr Troyer and his colleagues have re-run the analysis, this time using the most up-to-date D-Wave system available. In a paper posted to arXiv, an online repository, they report that, on average, there is still no sign that D-Wave’s machine offers a performance boost. Averages may disguise many things, and there are cases in which the D-Wave device is up to ten times faster than an ordinary chip. But there are also cases in which it is around a hundred times slower. And, tellingly, while both the classical and quantum machines get slower as the problems they are asked to solve become more complicated, they seem to slow down at roughly the same speed.
Shortly after the paper was published, Google’s Quantum Artificial Intelligence Lab, which has been playing around with one of D-Wave’s machines, published some thoughts of its own. It admitted that non-classical machines seem to be able to match the quantum device’s pace. But it had some suggestions for greasing the quantum wheels, such as improving D-Wave’s chips by allowing more of their quantum-mechanical bits to connect to each other—admittedly, a far-from-trivial task. And it speculated that an unambiguous advantage for the quantum machine might be lurking somewhere in the roughly 400,000 benchmarks it has already performed. (On the other hand, one might presume that any really significant speedup would have been spotted already.)
Troels Ronnow, one of Dr Troyer’s colleagues and the new paper’s lead author, has tried to be as fair as possible. The first half of the paper consists of a lengthy discussion about how best to compare quantum computers with classical ones, and lists several pitfalls that the rest of the paper carefully avoids. Echoing the Google team, he speculates that tweaks to D-Wave’s hardware—such as improving its error-correction—might bring benefits. And he concedes that there may be other problems, besides his synthetic benchmark, that the machine does better at. But it might also be that D-Wave’s device is simply no faster than a well-tuned non-quantum machine
Correction: We originally described Dr Troyer as a computer scientist. He is, in fact, a computational physicist. Our apologies.
Powered by Facebook Comments