The world’s only commercially-available quantum computer has faced much controversy about whether it is actually faster or better than a conventional computer. A new independent speed test helps to answer that question.
In short: the D-Wave quantum computer is thousands of times faster than other commercial computers at the very specific problem it was designed to solve. The computer is about average on other types of problems, and, importantly, it is still not clear whether the speed advantage will scale up as the computer gets bigger. That would be necessary to fulfil one of the big promises of quantum computing: making otherwise-intractable problems solvable.
Catherine McGeoch, a computer-science professor and algorithm-speed tester at Amherst College in Massachusetts, was asked by D-Wave, a quantum-computer company based near Vancouver, Canada, to put the company’s latest quantum computer through its paces. This was not an easy thing to do. The D-Wave device operates differently from other computers, not just because it uses quantum bits that exploit fuzzy quantum behaviour to speed up calculations, but also because it doesn’t use logic gates to perform operations. Instead it does something called ‘annealing’, where an answer is arrived at by looking for the lowest energy state of the bits in the computer chip. “It’s like comparing apples and oranges, or apples and fish,” says McGeoch.
The D-Wave Two computer, which has 512 quantum bits, is designed to tackle classification-type problems that are useful in machine learning and image recognition. Essentially it is good at determining the best ways to sort things into different categories, such as X-ray scans that contain an image of a bomb and ones that don’t.
McGeoch compared a 439-qubit version of D-Wave to a commercial product from IBM designed to solve the same sorts of problems. The IBM product is designed to deliver a confident answer to a given problem after 30 minutes. McGeoch found that D-Wave did just as well at finding the right answers, but in a half-second run time. That’s 3,600 times faster. “It was really amazing,” she says.
On other sorts of problems, the D-Wave computer was slowed down by having to ‘translate’ the question for their quantum chip, using a conventional front-end computer. In these cases it was about the same speed as conventional computers; overall McGeoch gave it an “above average” grade.
This doesn’t necessarily mean that D-Wave is the fastest approach even for the problems it is designed to tackle. Another test, performed by a different group, previously showed that a quantum annealer like D-Wave could be beaten in speed tests by a non-quantum, conventional annealer (see this previous blog post). “That didn’t surprise me. They’re comparing two highly specialized solvers in ideal laboratory conditions,” says McGeoch.
It is unclear whether D-Wave’s speed advantages will stick as the computer gets more qubits: it’s possible that it will keep producing correct answers to larger and larger problems in half a second, while conventional computers take longer and longer to crack them. “Right now it’s hard to say. I have a bunch of data sitting around for me to try and answer that this summer,” says McGeoch.
McGeoch will present her peer-reviewed paper at the International Conference on Computing Machinery in Ischia, Italy, on 15 May.