D-Wave announces general availability of first quantum computer for business | Hacker News Reader

D-Wave announces general availability of first quantum computer for business

dwavesys.com

128 pointslelf77 comments

I believe in quantum computing in the long run, I took some graduate math classes on the subject in college; I'm at least passingly familiar with the principals, so the following isn't really coming from a totally un-informed perspective:

With regards to the D-Wave systems in particular, are these actually faster than solving the problems with a classical approach? There's no benchmarks, no numbers, nothing -- just a customer list. A customer list whose engineering offices want to look hip. I mean, VW, Accenture and Save-on-Foods? I can't help but feel like if it did anything even a hair better than classical computers, the customer list would include every FAANG.

D-Wave feels a bit like a blockchain company. Is it (better|faster)? Sir, it's got them magic of (qbits|blockchain)... I don't understand your question.

> With regards to the D-Wave systems in particular, are these actually faster than solving the problems with a classical approach?

Obviously the answer to that is "No".

Google just did a "Quantum supremacy" experiment with lots of fanfare and some criticism that it really wasn't (details on that are complex). Quantum supremacy is exactly the thing you're asking for: Can we do something, anything at all, with a QC that we couldn't just do on average computers?

The important thing to realize is: Google did this with a very specific problem that was suitable for their quantum computer. It wasn't something that was practically useful. They chose a problem they could solve, nothing more. Noone has claimed to have run a quantum supremacy experiment on D-Wave.

My best guess is all their happy customers had a problem that needed to be solved, they solved it, and nobody asked if the "quantum" aspect was particularly relevant in it.

An important point about the technical meaning of "quantum supremacy": It is only the claim that a problem exists that can be solved on existing quantum computers and not on existing classical computers. The problem that Google used to demonstrate quantum supremacy has no practical uses, and no useful problem is known that can act as a witness for quantum supremacy. The practical implication: "Getting you problem solved" cannot be the motivation for buying quantum compute power from D-wave (unless you have no clue). Possible valid motations: "learning to work with quantum computers", "seeming cool", "donating money to companies that may build useful quantum computers in the future".

I thought the problem had some limited applications. Wasn't it something like modeling the quantum states of hydrogen?

That was a different test, that wasn’t trying to do something infeasible classically. The supremacy test was effectively solving a contrived combinatorics problem.

Thanks for the link, based on the reading I think I was walking whether there was a quantum advantage and it’s magnitude, although the quantum supremacy aspect is interesting too.

That's a press release, not a technical paper, so yeah, it's fairly lacking on technical details. Performance of quantum computers is rather difficult to summarize in an easily digestible measure, so press releases tend to cover the basics: a number of qubits, a degree of connectivity, etc. You might prefer a whitepaper[1] that does dig into some of the more technical performance details. (alas, it took me a couple of tries to find the paper as our webserver seems to be stumbling under load)

Our focus isn't on artificial benchmarks. Our approach is more pragmatic. We build the best hardware that we can today, and the best software stack around that we can, and we're building our business around getting better solutions into the hands of customers.

[1] https://www.dwavesys.com/sites/default/files/14-1049A-A_The_...

> Our focus isn't on artificial benchmarks. Our approach is more pragmatic. We build the best hardware that we can today, and the best software stack around that we can, and we're building our business around getting better solutions into the hands of customers.

Speaking as a sysadmin, this reads like "our numbers are so hilariously bad, that we're trying to deprecate the concept of measurement."

So, tell me. Precisely what are your results on the "artificial benchmarks"? Something that someone with one of your boxes could verify.

Do you have any case studies with partners?

There's a few out there; here's a couple that I'm familiar with.

https://www.dwavesys.com/sites/default/files/Dwave_Menten%20...

https://www.dwavesys.com/sites/default/files/Dwave_Groovenau...

There's quite a few technical reports by partners out there, as well.

https://www.dwavesys.com/resources/publications?type=externa...

Claim is that VW is using it to "calculate traffic flows" [0]. That sounds like the "traveling salesperson problem" that quantum is supposed to be good at [1].

0 https://www.dwavesys.com/sites/default/files/Dwave_Tech%20Ov...

1 https://medium.com/@michal.stechly/solving-the-traveling-sal...

Yeah, to my extremely limited understanding currently quantum algorithms are best for optimization problems in terms of obvious business value. Meaning their utilization does not need innovation, just a quantum compute platform to a known but currently intractable problem.

Old big companies with lots of logistics presumably have lots of logistics problems (optimization) they would not mind solving.

So I would expect the first customers of any relevant quantum stack to be the companies whose business depends on large scale physical, not the digital world. With problems that are already analyzed, but lack agreeable solution.

Hence FAANG are not necessarily the first clients.

The Quantum computer has to cost-effectively solve optimization problems cheaper than classical computers. Existing computers can handle large scale optimization problems especially with GPUs and/or cloud computing.

As other posters have called out, most early customers are probably just interested in getting a head start in understanding the tech and doing research now rather than too late.

Much new hyped tech is like this, a lot of people have fear of missing out in not starting too late on tech that will provide a business advantage, either execution wise or just brand wise (many people don't want to be the dinosaur, using no longer trendy tech).

They said that for 2-SAT problems, the new architecture (Pegasus) works better for hard problems, whereas for simpler problems the Chimera architecture is faster.

Surely you mean 3-SAT, because 2-SAT is trivial :-)

Save-on gave numbers: an optimization algorithm took 25 hours to run normally, now completes in 2 minutes.

I saw that, but... could they have optimized their classical code? Was it just a garbage batch script running single-threaded on a single-CPU machine backed by a MySQL SELECT * FROM query? Frequently these types of sweeping statements fall down with the slightest breeze.

If they had resources to throw at a quantum computer they have probably been throwing resources at trying to optimize for a while now and I'd assume lots of people have scratched their head trying to improve it on their own and get a promotion / applause / bonus.

It's either a problem that is hard to optimize or people lack the skills... though if they've managed to migrate to a quantum computer they probably have the skills necessary.

1500x is a impressive speedup (1500 min to 2 min)

Also - MYSQL SELECT is a I/O problem, bash scripts usually are not CPU problems... not the kind of things a quantum computer would even apply...

I cannot imagine that Save-on-foods IT has a culture of excellence. I'd love to be mistaken, but I doubt I am.

It's a very mediocre grocery store. I'm glad I don't see ads anymore because their marketing is always terrible. Their mascot is some guy named Darryl. Ugh I don't care for SOF. Their produce sucks too.

I feel like you’re asking whether the model T can be fitted or at least compared with a car with a turbo. From the Dwave machine I’ve read about, its more the algorithm must be appropriate (analogous to) in a fundamental way to the quantum system annealed for the computation.

He is saying that. If the original algorithm was badly coded and they ran the least optimised variant ignoring all updates, then the QC version may indeed have been faster, but it depends: was it tuned to a specific problem, and is there a better than model-T classical computing algorithm for the same specific problem.

I'm genetically minded to be skeptical of QC. The question isn't outrageous, even for a believer.

The follow-on question is therefore, are they going to save more time in the long run by writing poorly optimised queries for a particular quantum hardware stack (that we assume as being easier/faster to write), rather than writing highly optimised queries on classical hardware.

Writing a garbage SQL query and getting it back faster might be worth it over optimising it - especially if the questions aren't regularly repeated.

Until there is an undisputed demonstration of quantum supremacy [1], which at last check there isn't, then no, these (or any other quantum computer) is not faster than classical approaches.

The reasons companies are investing in these is R&D, or in more capitalist terms, FOMO: if quantum computing does turn out to offer an edge someday, you want to be able to make use of it fast, instead of scrambling to catch up.

[1] https://en.wikipedia.org/wiki/Quantum_supremacy

As far as I know, quantum supremacy has been unequivocally demonstrated, albeit not on a very interesting problem, and of course without solving the quantum error correction problem:

https://www.scottaaronson.com/blog/?p=4372

And noteworthy in regards to this thread: Not with a D-Wave computer.

As that article notes, IBM (which obviously has a dog in this fight) has strongly disputed Google's claims. I don't have enough expertise to judge.

Iirc IBM came up with an optimization of the classical algorithm (specifically using enormous but still (barely) practical amounts of RAM) that reduced the advantage of Google's quantum algorithm by an order of magnitude or two.

There was two issues with their argument though. First, they hadn't implemented it yet, which is rather ironic considering that Google's feat was in actually delivering an implementation. Second, Google's quantum computer is still orders of magnitude faster than IBM's optimized classical approach, not to mention cheaper.

Neither do I, but Scott Aaronson surely does, and he believes Google have unequivocally proved quantum supremacy. According to him, IBM has disputed the extent of the advantage proven by Google, but not the actual fact that they succeeded.

Quantum can be still be faster/more powerful per dollar than classical without achieving supremacy, right?

There are a lot of big problems that are tractable on classical, but if they become orders of magnitude cheaper on quantum, that's still a great win.

Usually this is termed “quantum advantage”, when it’s better than classical at some commercial metric that “matters”.

Some graduate math classes on the subject in college might make you familiar with the principles but you'd be better off doing some networking to become familiar with the principals.

Probably for a very specific problem, you can us dwave as a faster and cheaper alternative to classicals

Haven’t followed the D-Wave story for years... So, aside from quantum annealing vs “actual” quantum computing, has it been conclusively decided that D-Wave speedup is of a quantum nature?

We don't have a theoretical proof of a superpolynomial speedup on your favorite problem, no. But as a dear friend and colleague is fond of saying, "it works in theory -- but does it work in practice?"

Not your favorite problem; but one capability we've added recently is the ability to sample Ising problems with a nonzero transverse field. I'm yet unaware of a classical algorithm that can efficiently approximate this problem, though that isn't a hot research problem at the moment.

The "of a quantum nature" question has been affirmed rather conclusively. Our qubits are capable of entanglement and cotunelling (you may recall a "billion times speedup" result which got way overhyped -- the actual nugget there was a demonstration that groups of 8 qubits could cotunnel to find solutions that were "hidden" from general-purpose Ising solvers), and for example, when we simulate materials which should exhibit quantum properties, we see those properties emerge without any sort of "fine tuning" aside from compensation for known irregularities.

> but does it work in practice?

So does it?

Specificly is there a problem that businesses want to solve, that your device solves, where it is cheaper to buy your device (regardless of whether or not quantum magic is happening) than it is to buy a standard server and do the problem classically?

This should be an easy question to answer if your device actually works, pragmatically speaking.

Edit: i just RTFA'd (yeah i know im terrible), the article is claiming some practical speedups. I guess time will tell how well that pans out in practise as more customers try it on more things.

With the D-Wave flavor of quantum computers, it is like solving every issue using a traveling salesman algorithm. It can be done, but you aren't using a general purpose quantum computer to do so.

More specifically, just to elaborate on your point, it purports to solve certain “quadratic unconstrained binary optimization” (QUBO) problems [1]. It’s not a “programmable computer” in the traditional sense we think. It’s more “tune a bunch of knobs and press ‘go’.”

Quantum computers made by Rigetti, Google, and IBM are all programmable with some manner of programming language, and not coincidentally, the number of qubits is two orders of magnitude fewer.

[1] https://en.m.wikipedia.org/wiki/Quadratic_unconstrained_bina...

Yes. That is my point. I meant that it is not a general purpose quantum computer.

There is no general purpose quantum computer that has 5,000 qubits. The D-Wave machines use quantum annealing. My reference was to the probabalistic gpc algorithm for solving TSP using annealing.

It performs computations; it does so using quantum effects; therefore it is a quantum computer.

Every single thread that discusses D-Wave ends up with this same talking point about "universal" quantum computing. Yes, it's not yet another vaporware gate-model QC toy like Rigetti or Google have; it's a very specific sort of accelerator, somewhat akin to a GPU or a vector processor - not in terms of what it does, but in terms of requiring specific programming to make it useful. It's simply a vastly different architecture than gate-model QC: in many ways, it's a descendant of the analogue computers of the 60s and 70s.

The key differentiator here is that they've been able to scale their model from a couple hundred qubits to >5000, in production for anyone to use online. If you're a mathematician and you can actually model your problems this way, they have a nice SDK and it's very usable. There's every reason to assume that they'll have 10k qubits with even more connectivity in a few years, and they have this hybrid solver thing which takes in 10k+ variables already, so it's clear they believe in the problem model itself.

I just wish I could make it do more - my layman's interest has only taken me far enough to get a basic idea of what's going on for curiosity's sake. I think you need to be a mathematician in order to be able to do anything serious with it, right now.

You’re not even wrong. A digital signal processor does computations, therefore it’s a computer... right?

In general, these days, people, especially technical, usually equate “computer” to “programmable computer”. This is as evidenced by the continual confusion as to what a D-Wave machine does and how they’re inequivalent to universal gate-based architectures.

A lot of folks these days call the kind of ‘computer’ that D-Wave is an “accelerator”.

> A digital signal processor does computations, therefore it’s a computer... right?

Yes, absolutely it is. Just like how Babbage's difference engine was a computer, just like how a calculator is a computer. It's not a stored-program Von-Neumann turing complete thing, no, but that's clearly not the point - if you want one of those, wouldn't you just go to Intel et al?

The point of the D-Wave machine is to make super specialized optimization functions fast. Most of us, myself included, barely understand what those are or why they're useful. People interested in ML and neural networks, on the other hand, may find a processor designed intrinsically to work on entire graphs at a time to be somewhat useful. Maybe it's not big enough to do that yet, but ENIAC wasn't big enough to run Firefox - was it useless?

Sounds like different definitions, not not-even-wrongness. Parent commenter even mentioned it's akin to an accelerator.

D-Wave doesn't make a general purpose quantum computer. The sense of that statement is that it cannot run Shor's algorithm, though it can factor integers, for example.

Are you a bot? I'm literally responding to this exact criticism. Yes, it's not a general purpose QC. Does that somehow make it (a real product you can use) less useful than a hypothetical QC from some other company that doesn't actually exist yet?

This is like complaining about GPUs not being general purpose. Sure, that's a criticism, but it's not actually that valid.

Have we tested it with machine learning applications?

So it is more efficient at exploring the solution territory of traveling salesman type problems than classical computers?

I would safely say that it does not solve TSP faster than the best TSP code available - Concorde.

More info here, and note there is an iOS app too (Concorde TSP on the App Store) that you can play with that solves TSP to optimality: http://www.math.uwaterloo.ca/tsp/concorde.html

It has not been shown that it solves QUBO/NP hard problems any faster asymptotically.

The announcement reads like the days of the first regular computer release.

I wonder if there will be a personal quantum computer, and then a quantum mobile phone.

The negativity sounds similar to me ...

> Accenture, a leading global professional services company, is exploring quantum, quantum-inspired, and hybrid solutions to develop applications across industries. Accenture recently conducted a series of business experiments with a banking client to pilot quantum applications for currency arbitrage, credit scoring, and trading optimization, successfully mapping computationally challenging business problems to quantum formulations, enabling quantum readiness

I'd love to know more about that... Incredibly dubious on the claims

I wonder what you could do with 50 qbits. Factoring 15=3*5?

DW’s few thousands bits are not general purposes and error free.

Apparently prime factoring has been performed on d-wave hardware. However, it's not clear to me that this leads to something useful. It's not Grover/Shor's algorithm, it's something else. Based on a quick scan of this presentation, I don't see what the claim is as far as the scaling law in space or time for the general factorization of biprimes problem.

https://www.dwavesys.com/sites/default/files/32_Thurs_AM_Pur...

Ah found the "real" paper. The authors don't know how the algorithm scales either:

"Finally, we note that while our demonstrations of factoring have made use of currently available quantum annealers, there is an outstanding question regarding the asymptotic complexity for this approach." https://arxiv.org/pdf/1804.02733.pdf

and go on to talk about the "minimum spectral gap between the ground and first-excited states of the underlying time-dependent Hamiltonian" which might as well be something spoken in the engineering section of the Voyager to me.

The D-wave system is not suitable for factoring problems. AFAIK, no one has demonstrated the factoring of 15 using Shor's algorithm in a way that did not assume the answer.

https://www.dwavesys.com/sites/default/files/14-1002A_B_tr_B...

Granted, factoring isn't the best use of our hardware, but as of 4 years ago, we were hitting 80% success rate on 10-bit semiprimes without baked-in knowledge of the answer. Doesn't need Shor's: we just implement a multiplication circuit as an Ising problem, clamp the outputs, anneal and read off the inputs.

Why would anyone be interested in what D-Wave has to say?

Their previous "quantum computer" wasn't actually one, so why would I even waste time finding out if this one is?

How does this compare to the Google Sycamore quantum processor?

Would be cool if one could use this to do Hartree-Fock how google layed out [0] with the qubits to compute a bunch of binding energies for a matrix of different elements and molecules and use those to find activation coefficients on a classical computer.

[0] https://arxiv.org/abs/2004.04174

D-Wave machines are quantum annealing machines, not really a quantum computers like Sycamore.

In other words it's machine that can solve combinatorial optimization problems and nothing else.

Are these companies purchasing D-Wave computers or just renting compute time on them? The latter makes more financial sense to me

Anyone catch the price? Or is this one of those "contact sales" products...

You can’t just buy the machine like any other COTS product. You typically need a team of PhD scientists to operate it as well.

I think D-Wave has also gone the way of providing it as a cloud product with their ‘Leap’ service.

Last I heard (I work there, but not in hardware) it takes a couple of technicians a few weeks to install the machine onsite, but after that, they're extremely low maintenance. The skill required to use them "locally" is quite the same as using them on the cloud, and we regularly have undergrad co-ops get up to speed using them. No PhDs required.

That said, we're much more focused on cloud sales; under that model, big customers get dedicated resources and everybody gets access to machines just about as fast as we make them.

I’ll admit, it’s quite a feat if the machine can be self-regulated without scientist hand-holding. Is that true? Does that machine self-correct from calibration drift?

I'm not familiar enough with our calibration code to speak to that directly, but we can do calibrations remotely without sending techs out. I'm told that the impressive thing is that our dilution fridges can run for several years uninterrupted.

otherwise known as PhDaaS

You needed a bunch of Ph.Ds to operate early digital computers too. You didn’t just unbox one of those and start doing stuff.

This is probably the 1930s of quantum computing.

If I know my history,* the first programmable classical computer was made in 1941; we had the D-Wave One programmable quantum computer in 2011. So... the 50s?

* I don't, so I read a Wikipedia article or two

Also available, I think, on Amazon Bracket (and perhaps on Azure Quantum as well)

> "contact sales" products

We need a Wiki-ish Chrome plugin that pastes in actual prices on all websites that have "contact sales" pricing. If you are the first to contact sales, you can fill it in for future customers to see without needing to contact sales.

This would also help expose price discrimination, in which companies give favorable pricing to people of a certain race or nationality or industry or location.

I hate it when I contact sales and they ask "may we know more about your application?" and my next thought is always "how about you tell me your price first, and then I'll be happy to share". "may we know your name" -> "why, so you can decide your offer price based on if I'm white or asian or something?". "may we know your location" -> "why, so you can decide your price based on the average income in my area?"

This sounds just like someone calling up McClaren or Ferrari and demanding an MSRP on one of their Formula-1 cars... and screaming RACISM if they don't have an immediate answer.

Systems like this are not just sitting on a shelf with a price tag... there are tons of other considerations going on here: integration, deployment and support contracts... required infrastructure like power, network switches - even cabling can go into the tens of thousands of dollars.

No chrome plugin is going to fix this "contact sales" problem, and assuming some nefarious racist intent only goes to show that the inquiry is a waste of time for the person on the other end of the line.

> and screaming RACISM if they don't have an immediate answer

Oh I can tell you plenty of stories of suppliers giving you lower prices depending on what human language you speak with them. It happens, and people should know that it happens.

> even cabling can go into the tens of thousands of dollars

For those types of products, great, I'd like to know that cabling will be ~$10K-$50K, fixed costs will be $10K, and support contracts will be ~$1-5K/month. I just want an order of magnitude. Is this $100, $1000, or $100,000 that I'm looking at? I want an order of magnitude before a phone call. If such a plugin existed, at least I could see the distributions in pricing among past customers.

Yes, in your first example this is known as “value-based pricing.”

However, when it comes to quantum computing, they may be trying to determine if your application is even relevant to their solution.

It's something like $2k for an hour of processor time, which is a lot of submitted jobs. You can sign up for free, and I think they have credit card transactions, so no, it's a bit more direct than all that.

Soo, 5000 qbits, ey? Then why are not those D-Wave already owning all the bitcoins in the world? I mean the biggest encryption used by wallets are protected by 4096 bit length, which is just a shy below 5000. Isn't quantum suppose to fly in microseconds at cracking public key encryption for that length?

I call this snake oil.

It's quantum annealer, not a universal gate system

https://medium.com/quantum-bits/what-s-the-difference-betwee...

The number of bits in an encryption key and the number of qubits in a quantum computer trying to break it are not related in general and even if they are related, the relation is not necessarily a linear relation.

Also quantum computers do not in general offer an advantage when it comes to breaking encryption keys. They do, for example, for RSA because it relies on integer factorization and there is a faster quantum algorithm than a classical one, they do not - as far as we know - for AES for example.

Reply on news.ycombinator.com