D-Wave's 2,000-Qubit Quantum Annealing Computer Now 1,000x Faster Than Previous Generation (tomshardware.com)

An anonymous reader quotes a report from Tom's Hardware: D-Wave, a Canadian company developing the first commercial "quantum computer," announced its next-generation quantum annealing computer with 2,000 qubits, which is twice as many as its previous generation had. One highly exciting aspect of quantum computers of all types is that beyond the seemingly Moore's Law-like increase in number of qubits every two years, their performance increases much more than just 2x, unlike with regular microprocessors. This is because qubits can hold a value of 0, 1, or a superposition of the two, making quantum systems able to deal with much more complex information. If D-Wave's 2,000-qubit computer is now 1,000 faster than the previous 1,000-qubit generation (D-Wave 2X), that would mean that, for the things Google tested last year, it should now be 100 billion times faster than a single-core CPU. The new generation also comes with control features, which allows users to modify how D-Wave's quantum system works to better optimize their solutions. These control features include the following capabilities: The ability to tune the rate of annealing of individual qubits to enhance application performance; The ability to sample the state of the quantum computer during the quantum annealing process to power hybrid quantum-classical machine learning algorithms that were not previously possible; The ability to combine quantum processing with classical processing to improve the quality of both optimization and sampling results returned from the system. D-Wave's CEO, Vern Brownell, also said that D-Wave's quantum computers could also be used for machine learning task in ways that wouldn't be possible on classical computers. The company is also training the first generation of programmers to develop applications for D-Wave quantum systems. Last year, Google said that D-Wave's 1,000 qubit computer proved to be 100 million times faster than a classical computer with a single core: "We found that for problem instances involving nearly 1,000 binary variables, quantum annealing significantly outperforms its classical counterpart, simulated annealing. It is more than 10^8 times faster than simulated annealing running on a single core," said Hartmut Neven, Google's Director of Engineering.

Comments

ELI5

[OverlordQ]

What can you actually do with this?

Re:ELI5

[gweihir]

Nothing you cannot do much better and much faster on a much cheaper conventional computer.

Re:ELI5

[kaiser423]

Annealing is an optimization algorithm, mainly. It can be applied to other things, but generally it is really good at optimizing complex problems with lots of variables. Used extensively in simulation packages for pretty much everything, and other problems without easy closed form solutions. Good for the traveling salesman problem also.

Re:ELI5

[networkBoy]

Can you provide links?
Seriously, I (and I suspect many others) have a decent idea of the *concept* of quantum computers, but understanding actual application is... elusive.

Re:ELI5

[ShanghaiBill]

What can you actually do with this?

Simulated annealing, which is used to solve a huge number of optimization problems. It is very compute intensive, and is often run on GPUs. If the claimed speed up is real, that would be a really big deal.

Re:ELI5

[legRoom]

I (and I suspect many others) have a decent idea of the *concept* of quantum computers, but understanding actual application is... elusive.

Just FYI, D-Wave is not a general-purpose quantum computer. It's a specialized device for solving one very specific class of problems; gaining insight into it probably won't help you understand the full capabilities of quantum computers.

I can't explain quantum computers to you in general, because I don't understand them either. I do know one very important application though: using Shor's factorization algorithm to break RSA encryption. You'll hear about it when real quantum computers reach commercial maturity, because a bunch of Slashdot articles will appear about how everyone is in a panic to rush and replace RSA with something else. :-)

Re:ELI5

[JoshuaZ]

Someone downvoted you, possibly due to a lack of sourcing. So in case anyone is in doubt, they should look at this blog by Scott Aaronson http://www.scottaaronson.com/blog/?p=2555 and the discussion. Aaronson is one of the top quantum computing experts on the planet. The comments there are also very relevant. Alex Shelby notes that the algorithms that D-Wave has used to compare on conventional (classical) computers are substantially less efficient than the best classical algorithms. We are going to eventually have actual quantum computers, and when we do they will be awesome. Right now, it isn't clear that D-Wave's system can be reasonably called a quantum computer, and is even more clear that they aren't useful at all.

Re:ELI5

[ClickOnThis]

Can you provide links?

Here you go.

TL;DR? The basic idea behind a simulated annealing algorithm is that it searches for successively better solutions, but occasionally accepts a "worse" one, so as to reduce the possibility of getting stuck in a local minimum when there is a better minimum nearby (sort of like jumping out of a caldera at the top of a mountain, so that you can reach a a better minimum closer to ground level.) As time goes on, the probability of accepting a worse solution is reduced, according to an "annealing schedule" until finally only better solutions are accepted.

Seriously, I (and I suspect many others) have a decent idea of the *concept* of quantum computers, but understanding actual application is... elusive.

Simulated annealing is not an exclusively quantum-based algorithm. It works quite well on classical computers. But it is a method that would perform very well on a quantum computer.

Re:ELI5

[Rockoon]

On the contrary simulated annealing fell out of common usage due to other stochastic search methods being better at solving many problems types.

For instance the Extended Compact Genetic Algorithm converges much faster, and dont let its name fool you its not a genetic algorithm as the name Compact Genetic Algorithm is derived not from the technique, but instead the name is derived from the space it searches which is exactly equivalent to a simple genetic algorithm with a crossover probability of 0.5. The Compact Genetic Algorithms is instead an estimation of distribution algorithm, and the Extended version detects and leverages the dependencies between different elements of the solution vector in a theoretically optimal (information theory) way, which gives it an advantage over algorithms that don't (which includes Simulated Annealing, which is why it fell out of favor.)

Annealing is still used for problem sets where there isnt a lot of dependencies within the solution vector.

Some of the d-wave haters have moved onto the argument that the system isnt faster than a conventional one when the conventional one runs a "better" algorithm .. see the big paragraph above. "Better" means searches a different solution space and therefore cannot solve all the same problems.

Re:ELI5

[JoshuaZ]

If a cheap conventional computer works better at the same tasks then it

First these fucks said d-wave wasnt doing any quantum stuff. Then these fucks said it was slower than conventional hardware. Now these fucks say its still slower than conventional hardware if you use a different algorithms that wont solve the same set of problems...

This is not accurate. The first statement is that it wasn't clear that the D-Wave system was engaging in any quantum computation. That's still not clear. Part of the issue here is that it simply isn't completely clear what one means by quantum computation in this context. For example, your laptop's transistors use quantum mechanics in a critical fashion, but they aren't doing quantum computations. The question has always been twofold a) is non-trivial entanglement going on and b) is that entanglement being used to do processing that cannot be easily simulated on a classical system. Those are both strongly connected to questions of efficiency. Right now, the answer to a seems to be yes (although it took forever for the evidence to actually come out).

Your second two sentences are even more wrong. The fact is that it is slower than cheap conventional hardward if one *uses the best known classical algorithms*. That's being used to solve the same problems, as would be clear, if you read the link I gave.

Your insistence that one must use the "the same algorithms" to benchmark is also incredibly wrong in this context, since one cannot use the same algorithms on both at a fundamental level. D-Wave's system uses a variant of an annealing algorithm and cannot run classical algorithms in any meaningful way. In that context, the classical computers are in this sense essentially emulating an annealing process. If you insist that one must use the same algorithms rather than look actual time for solving problems, then the systems are simply incomparable. Actually looking at cost and time to solve problems makes more sense.

As someone else noted.. Google, NASA, etc must be complete idiots for not bowing to the clearly rational flying goalpost these fucks swing around.

Let's recall for a moment that the primary "fuck" you are talking about is Scott Aaronson who is one of the world's most respected quantum computing experts. He's responsible for many major results including the algebraization barrier http://www.scottaaronson.com/papers/alg.pdf and the first substantially non-trivial lower bounds on the basic collision problem http://www.scottaaronson.com/papers/collision.pdf among other work.

But let's for a moment think about what is going on with Google and NASA and consider other explanations that are relevant here. First, both Google and NASA both have major interests in basic research, and there's a valid basic research interest in what D-Wave is pursuing. (I personally consider it unlikely to go anywhere that useful compared to gate-based quantum computing research but that's a judgment call.) Moreover, large corporations and governments like fads: it doesn't take much for some mid-level manager to decide that quantum computing is a shining new thing and realize that the easiest way to jump on the bandwagon is to buy a D-Wave machine.

Re:ELI5

[gweihir]

Aaaaand, fail. You do not compare the same algorithms if the computing devices have fundamentally different characteristics. For example, you do not compare a single CPU computer and a large cluster using the same algorithm. You compare them using algorithms that deliver the same results, but one gets a classical algorithm and the other gets a parallelized one. That is actually benchmarking 101.

Because if you insist in the same algorithm, you will find that one device has to simulate the other in order to be even able ti run that algorithm. That is not a relevant comparison in any way. May as well compare the speed of an airplane and a bicycle and require the bicycle to fly for that. Stupid.

Re:ELI5

[ChrisMaple]

Gee, you use lots of foul language. You must be right.

Re:ELI5

[Pseudonym]

You want to know what this D-Wave technology represents? A start.

Yes and no.

Yes, D-Wave technology represents the first practical step in one possible direction for building practical quantum computers.

No, in that it is an insane hype machine. How many pieces have you seen which claims that D-Wave is working on solving NP-complete problems efficiently, something which pretty much everyone believes quantum computers are incapable of doing? If D-Wave can't live up to the hype, it may ruin quantum computing for everyone, and that would be a bad thing.

My mother (who has spent a lot of her life on committees of various kinds) taught me from a young age that the problem isn't often what is done, but the way it's done. D-Wave embodies this principle.

Re:ELI5

[JoshuaZ]

I know you are joking, but it is worth noting that a) most of the D-Wave skeptics are concerned about articles that portray D-Wave as practical right now or b) Are concerned because they see it as a lot of funding going to what they see as an unlikely to succeed research avenue when it could go to gate-based quantum computer research which is seen by most in the field as more promising.

Re:ELI5

[MachineShedFred]

Why do I give a shit what algorithm is used, if both systems give me the same answer, but the cheaper conventional system does it faster? Is the label "quantum annealing computer" supposed to give me a gold star or something?

Re:ELI5

[gweihir]

And that is _exactly_ the right question. What matters is what the thing can do in comparison to existing technology. And there the D-Wave fails utterly.

On the down side...

[Black Parrot]

One cat dies for each bit that is settled in the solution.

Re:On the down side...

[Killall -9 Bash]

Every other time you collapse a super-position, Science kills a cat.

pesky qubits

[JustNiz]

Goddammit, I just found a qubit that needed annealing but I think its slipped down the back of the sofa.

Commercial v. Government Actors

[SeattleLawGuy]

You'll hear about it when real quantum computers reach commercial maturity, because a bunch of Slashdot articles will appear about how everyone is in a panic to rush and replace RSA with something else. :-)

"commercial maturity" being the key word here, because we should assume that significant portions of major classified intelligence budgets are being thrown at the problem by the US and China, maybe also by a few other players (India? Israel? The UK? Russia?). Like how it's widely believed that differential cryptanalysis was known to the NSA well before it became known to the world, only today encryption is much more prevalent and much more important to anyone doing signals analysis.

Re:Uh, useless except for breaking RSA

[WaffleMonster]

And ECC. Probably not with this generation, but it's entirely possible three letter organizations are close to having a machine that can at least break smaller key sizes. Which in and of itself isn't especially worrying, except that the more sophisticated and rogue state-sponsored criminals won't be lagging very far behind.

There is no evidence in existence to suggest code breaking quantum computers are even feasible.

There's no good reason why our web security infrastructure shouldn't immediately begin upgrades to support multiple, extensible and arbitrary methods of key exchange, including but not limited to stateful solutions using out of band preshared keys.

TLS provides agility for key exchange and adding cipher suites is a routine affair - there are literally hundreds of them. Everyone will at least have to rekey and at the most update their TLS stacks. Both have been done before and isn't such a big deal in the grand scheme of things.

By far the biggest problem takes the form of retroactive decrypting of previously captured encrypted data including data protected by forward secure algorithms.