As a tech company, Google has been known to branch out into different areas of interests, from self-driving cars to smart wearable technology and, more recently, to robotics. One thing it has dipped its toe in also quite recently is the field of quantum computing, having jointly purchased a fancy D-Wave 2 supercomputer. Now its A.I. Lab Team is trying to give an update on just how much that $10 million expenditure is worth.
Together with NASA, Google co-invested in a supercomputer in May last year manufactured by D-Wave. Dubbed the “D-Wave Two”, this computer boasts of having 512 “qubits” per chip and is supposedly 11,000 times faster than Intel’s fastest chip, at least on some tasks. Although at first glance it seems like a strange investment, but the supercomputer is a hardware that is perfectly suited for the type of data crunching tasks that Google needs in its business, particularly in search and automation. It may even use the supercomputer for some functionality related to Google Glass.
Given the promise of monumental speeds, it is only natural that people wonder just how fast the D-Wave 2 is. While that would indeed be an interesting thing to ask, Google says that the task of benchmarking is not so straightforward. In particular, the more practical method would be to instead compare it with traditional computer hardware that are running portfolio solvers, programs created to solve particular sets of problems.
Even then, however, the results are not set in stone and can vary depending on the problem set. When using random instances of a particular problem, the D-Wave 2 chip outpaced these solvers by as much as 35,500 times. However, this win isn’t absolute. D-Wave’s competitors are, more or less, general purpose solvers designed to work on any problem set. When pitted against solvers that have been designed to specifically compete with the D-Wave, the results were, more or less, evenly matched. There are problems for which classical solvers end up faster and there are problems wherein the D-Wave hardware can do quicker.
It is, however, too early to write this off as a failed investment. For one, the field of quantum computing is still rapidly evolving. On the hardware level at least, there are still certain obstacles that need to be overcome, in particular those related to the connectivity of qubits. There is also the fact that these tests used to benchmark D-Wave 2 and classical solvers only work on a small set of data, about a thousand. There is hope that a larger dataset that goes beyond that will prove how much the hardware will outperform these solvers.