Original URL: http://www.theregister.co.uk/2014/01/21/dwave_benchmark_flunked_or_wrong_test/

Boffin benchmark battle after D-Wave quantum kit crawls in test

D-Wave protests methods used to clock DW2 100 times slower than classical computers

By Richard Chirgwin

Posted in HPC, 21st January 2014 01:58 GMT

Yet another attempt to benchmark the D-Wave quantum computer – this time, in its 512 qubit configuration – has come up with disappointing results, with the company responding that it was the wrong kind of test.

The work, led by Matthias Troyer of ETH Zurich, pitted a D-Wave Two (the machine that Google bought to much fanfare last year), is two-fold: the nine scientists sought not only to run the benchmarks, but also to design benchmarks that didn't “mask or fake quantum speedup”.

D-Wave has told the BBC “the tests set by the scientists were not the kinds of problems where quantum computers offered any advantage over classical types.”

The group working with Troyer includes authors from Google's AI laboratory and the University of Southern California, something that should help fend off complaints that the test isn't fair, since both have invested in D-Wave machines.

Troyer's result mirrors last year's test on a 128-qubit machine, run by Catherine McGeoch of Amhurst College, in as much as the D-Wave machine does seem to perform faster than a classical computer – but only in some circumstances, in a subset of tests.

From the paper's abstract:

“We illustrate our discussion with data from a randomized benchmark test on a D-Wave Two device with up to 503 qubits. Comparing the performance of the device on random spin glass instances with limited precision to simulated classical and quantum annealers, we find no evidence of quantum speedup when the entire data set is considered, and obtain inconclusive results when comparing subsets of instances on an instance-by-instance basis.”

Further into the paper, the authors note: “We find that while the DW2 is sometimes up to 10 times faster in pure annealing time, there are many cases where it is 100 times slower.”

Google's Quantum AI lab discusses the benchmark here, noting that the only instance in which the D-Wave Two had a clear win was when it was pitted against a general purpose solver. When Troyer's team (with help from Nvidia) created a purpose-built solver, there was no quantum speedup.

(As it happens, Google was so impressed with the code written by a member of the ETH Zurich team, Sergei Isakov, that he is now working with the Quantum AI Lab). In a second, independent project, researcher Alex Selby was able to write a solver that matched the D-Wave.

Speaking to New Scientist, D-Wave says what matters isn't speed, but whether it has genuinely built a computer that uses quantum effects. Co-founder Alexandre Zagoskin says “new tools” are needed to work out whether the machine is a quantum computer, and that “The question about how fast this thing works is secondary”.

Another angle of D-Wave's response has been to post this counter-paper on Arxiv, in which it proposes a technique for demonstrating whether or not entanglement is taking place.

That paper notes that the currently-limited connections between qubits in the D-Wave chip could be a constraint on performance, something also put forward by Google: “the qubits in the current chip are still only sparsely connected”. The hope is that with more inter-qubit connections, the chip should run faster.

There's also the hope that by running larger datasets through the D-Wave machine, it may become easier to identify instances of quantum speedup. Google says NASA's machine, which has been tested against Kepler data, now has enough data for “400,000 problem instances”.

So far, the NASA machine has only been tested to see if it can identify already-known planets in the Kepler data (answer: yes, it can).

“We're now trying to identify a class of problems for which the current quantum hardware might outperform all known classical solvers,” the Google Quantum AI team writes. ®