Q-Day - let's hope we're preparing for it

[Tristan]

Good post from James O'Malley on a potentially world changing future event, Q-day. The day quantum computing allows people to break current encryption standards. Interesting parallels with the millenium bug and upcoming 2038 problem, which shows we can fix this stuff ahead of time with focus.

Worth a read: https://takes.jamesomalley.co.uk/p/this ... ried-about

[yoss]

Sabine recently posted this video https://www.youtube.com/watch?v=ALwrwbnWLjo , which was as usual a pretty interesting take I thought.

[dyqik]

I'll be dead before it happens.

Quantum computers currently need >20 kW of electricity to run 53 bits. That number cannot reduce substantially, due to thermodynamics - quantum computers need to be at low- or sub-Kelvin temperatures to operate. Room temperature BCS superconductivity might give a way round it, but that'll be along just after cold fusion (and would be massive boon for high temperature fusion, although high-Tc superconductors can work there.

[dyqik]

It's interesting that this scaremongering starts just as a bunch of quantum computing startups are trying to get another round of VC funding. Much like the AI scaremongering, this is a key marketing strategy for their companies.

[shpalman]

Our internet is being sh.t today so I'll spare you a long explanation of how Shor factoring is more difficult than people make out and only gets worse as you try to go to bigger numbers.

There's no reason to expect a Moore's Law style of exponential progress, only digital computing managed that. Quantum computing just gets exponentially harder to scale up to more qubits and/or longer sequences of operations. Errors accumulate and qubits pick up noise. It's analog rather than digital computing and it's suggested that you might need thousands of physical qubits all working together to behave as one robust logical qubit.

Room temperature superconductivity is unlikely to help because we don't just need to maintain the IBM transmons at millikelvin temperatures for the superconductivity but also to avoid noise. There are qubits which use semiconductors rather than superconductors which also need to be kept cold.

I do note, though, that none of the well known algorithms make use of entanglement. They all just use something a banal effect called "phase kickback" as if it were magic though.

ETA I think this can go in Nerd Lab.

[IvanV]

Useful quantum computing sounds like one of those still-30-years-away kind of things. If it does come down a bit, that might make it sound like we have plenty of time to migrate to using the many options for cryptography that is resistant to quantum computer decryption. But, as that article notes, probably people you don't want to read your messages are using a harvest-now-decrypt-later approach, in the hope that they will be able to. So you would like your messages using the potentially breakable methods to be no longer sensitive by the time someone can decrypt them, if that time ever arrives.

So, it might "never" happen. But the sensible thing is to put in place stronger cryptographic standards right now.

I think there has already been one upgrading, in that conventional computing became able to break certain codes that people once thought were strong enough. For example, RSA with somewhat shorter primes than is now normal.

[lpm]

Quantum computers currently need >20 kW of electricity to run 53 bits.

? I don't understand this point. A common or garden car recharger at the local services delivers 200 kW. A nuclear reactor can deliver 1,000,000 kW. Why is clearly electricity use a problem?

[shpalman]

Quantum Computers: What are they?

The tl;dr is that for his physics calculations he can do a lot more, to higher precision, with a digital computer now that he'll likely to ever be able to do with a quantum computer.

The Shor algorithm part starts 20 minutes in.

He also makes the point that while the demonstration that 15 = 3 × 5 is the classic test problem for the Shor's factorization algorithm, it's by no means trivial to get a quantum computer to actually multiply 3 × 5.

A Scientific Reports from 2021 which shows that 21 = 3 × 7 apparently using some entanglement for some reason

Maybe a quantum computer will eventually demonstrate that 42 = 6 × 9.

[monkey]

Quantum computers currently need >20 kW of electricity to run 53 bits.

? I don't understand this point. A common or garden car recharger at the local services delivers 200 kW. A nuclear reactor can deliver 1,000,000 kW. Why is clearly electricity use a problem?

I thought dyqik's point was one about efficiency, not ability to deliver power. Right now they're not very efficient, as far as I know.

But as most of the energy use in quantum computing is making stuff very cold, it shouldn't scale the same way that power usage of traditional computing does. Doubling the number of qbits shouldn't mean a doubling of the power requirement. ETA: I don't know how it would scale, mind.

The computers having to be cold means turning them on and off again is expensive too, making your IT department's job much harder.

[dyqik]

Quantum computers currently need >20 kW of electricity to run 53 bits.

? I don't understand this point. A common or garden car recharger at the local services delivers 200 kW. A nuclear reactor can deliver 1,000,000 kW. Why is clearly electricity use a problem?

Because you can do absolutely f.ck all to break encryption with 53 bits. You need 4096 bits to even store the newer web encryption keys (currently https uses 2048 bit keys, and 4096 bit keys are starting to come online), and more to crack it, and hours of processing set up and readout time, at several MWs of electricity input, or about $2000-$4000 an hour in electricity. Your quantum computer will look like a CERN detector system, and cost 100s of millions to build. All so you can read what I'm typing on this webpage.

It's significantly cheaper to send someone around to bash me over the head and read my computer screen.

[dyqik]

Quantum computers currently need >20 kW of electricity to run 53 bits.

? I don't understand this point. A common or garden car recharger at the local services delivers 200 kW. A nuclear reactor can deliver 1,000,000 kW. Why is clearly electricity use a problem?

I thought dyqik's point was one about efficiency, not ability to deliver power. Right now they're not very efficient, as far as I know.

But as most of the energy use in quantum computing is making stuff very cold, it shouldn't scale the same way that power usage of traditional computing does. Doubling the number of qbits shouldn't mean a doubling of the power requirement.

It probably does, because the cooling power required is dominated by the thermal parasitics and power dissipation of the readout of the qbits.

I use the same readout devices on my superconducting receivers, and the same cryogenic systems.

[dyqik]

Oh yeah, and you better hope that a cosmic ray doesn't hit your quantum computer, or your result will be wrong.

[lpm]

Quantum computers currently need >20 kW of electricity to run 53 bits.

? I don't understand this point. A common or garden car recharger at the local services delivers 200 kW. A nuclear reactor can deliver 1,000,000 kW. Why is clearly electricity use a problem?

Because you can do absolutely f.ck all to break encryption with 53 bits. You need 4096 bits to even store the newer web encryption keys (currently https uses 2048 bit keys, and 4096 bit keys are starting to come online), and more to crack it, and hours of processing set up and readout time, at several MWs of electricity input, or about $2000-$4000 an hour in electricity. Your quantum computer will look like a CERN detector system, and cost 100s of millions to build. All so you can read what I'm typing on this webpage.

It's significantly cheaper to send someone around to bash me over the head and read my computer screen.

That sounds remarkably cheap. An unnoticed fraction of the CIA budget.

[monkey]

? I don't understand this point. A common or garden car recharger at the local services delivers 200 kW. A nuclear reactor can deliver 1,000,000 kW. Why is clearly electricity use a problem?

I thought dyqik's point was one about efficiency, not ability to deliver power. Right now they're not very efficient, as far as I know.

But as most of the energy use in quantum computing is making stuff very cold, it shouldn't scale the same way that power usage of traditional computing does. Doubling the number of qbits shouldn't mean a doubling of the power requirement.

It probably does, because the cooling power required is dominated by the thermal parasitics and power dissipation of the readout of the qbits.

I use the same readout devices on my superconducting receivers, and the same cryogenic systems.

I claim no expertise, but that's not what I've read before (what I've read might be wrong, of course).

And if you missed it, I added an edit to my post saying I don't know how it would scale, cos scaling differently doesn't necessarily mean scaling to the point of being better at a useful level.

[IvanV]

Quantum computers currently need >20 kW of electricity to run 53 bits.

? I don't understand this point. A common or garden car recharger at the local services delivers 200 kW. A nuclear reactor can deliver 1,000,000 kW. Why is clearly electricity use a problem?

Common or garden? A 20kW device is not something you can normally run at home, at least not in Britain, unless you have obtained some special 3-phase connection to run a ground-source heat pump or something like that. If you have a number of these devices, then even in a commercial property it can can soon add up to something you'd need a special industrial-sized connection to the grid for. As an indication of size, taking into account peak evening demand, 1MW is probably enough for about 800 average dwellings, at present. Demand has tended to fall recently, because appliances are increasingly efficient. But if EVs and heatpumps become common, it will grow again substantially.

If you had a modern larger house with a 100A fuse, then the 20kW is an 83A draw. So you could just about get away with that, provided you were careful over what else you did in the house at the same time. Clearly normal domestic cabling would not suffice to plug it in. In practice, I think a 14kW car charger would be the maximum recommended for a house with a 100A fuse. Many older - like mine - or smaller houses have only a 60A fuse, and we can only run a 7kW EV charger in that case. It is typically the actual cable connection which is the impediment to increasing the fuse rating. Sometimes it turns out you can get away with changing it for an 80A fuse.

The service stations which have a bank of ultra-fast chargers have typically self-selected as being in locations that are, by chance, already close to high voltage cables. This is because they typically require a new connection and their own substation to get something like a 1MW connection. Fortunately such cables often run close by to motorways and trunk roads, so quite a few service stations suitably located for the traffic are also suitably located for the cables, but far from always. If you are in some random location and want a 1MW connection, you are likely not to enjoy the response from your local distribution company, as it costs the 10km run, or whatever, to the nearest suitable high voltage cable, likely across much intervening private land. That can be an especially expensive problem in urban locations.

[lpm]

Are some people here thinking this problem is everyone having a quantum computer in their spare bedroom?

Or thinking the problem is GCHQ won't be able to use it because our country is too incompetent to run a grid connection to Cheltenham?

[El Pollo Diablo]

Quantum computing vs fusion power - which will win? Only one way to find out!

[shpalman]

Oh yeah, and you better hope that a cosmic ray doesn't hit your quantum computer, or your result will be wrong.

It will be "wrong" anyway, because if the information is put into the basis which depends on the phase difference between 0 and 1 states via a QFT (the Shor algorithm prepares a state which looks like the QFT of something and then requires the inverse QFT to put it back into 0 and 1 states which can be read out) then the precision of your output depends on how precisely you can induce those phase shifts and not have them drift, but you get more precision with more qubits because in the QFT state the phase of each qubit is double the phase of the previous one (but then you don't know how many complete 2π rotations there have been).

fourier-basis.gif (269.46 KiB) Viewed 456 times

computational-basis.gif (265.37 KiB) Viewed 456 times

[dyqik]

Are some people here thinking this problem is everyone having a quantum computer in their spare bedroom?

Or thinking the problem is GCHQ won't be able to use it because our country is too incompetent to run a grid connection to Cheltenham?

I don't think GCHQ is interested in cracking the https key for my current session on this website.

ETA: if it was actually something important, I'd be using a much much bigger key. The cost of using one is not much.

[TopBadger]

Quantum computing vs fusion power - which will win? Only one way to find out!

My thoughts exactly...

[lpm]

Are some people here thinking this problem is everyone having a quantum computer in their spare bedroom?

Or thinking the problem is GCHQ won't be able to use it because our country is too incompetent to run a grid connection to Cheltenham?

I don't think GCHQ is interested in cracking the https key for my current session on this website.

ETA: if it was actually something important, I'd be using a much much bigger key. The cost of using one is not much.

You didn't read the article, did you. Becoming a habit of yours.

[dyqik]

Are some people here thinking this problem is everyone having a quantum computer in their spare bedroom?

Or thinking the problem is GCHQ won't be able to use it because our country is too incompetent to run a grid connection to Cheltenham?

I don't think GCHQ is interested in cracking the https key for my current session on this website.

ETA: if it was actually something important, I'd be using a much much bigger key. The cost of using one is not much.

You didn't read the article, did you. Becoming a habit of yours.

Why read illinformed hype? I'm not wading through pages of popups and italics to see that someone isn't familiar with basic cryogenic engineering.

[shpalman]

That's unfair, he's completely unfamiliar with all the physics and mathematics of quantum computing, not just the cryogenics:

The technical explanation of exactly why encryption can be broken by quantum computing gets complicated quickly.

I can’t claim to understand the physics myself, but...

There's just about zero actual content, just people insisting that Q Day will happen and it will be bad, as if the digital computing power available to perform encryption won't be way ahead of anything a quantum computer can help decrypt for as far as we can see. It even says in the article that the iPhone already got upgraded to an encryption which is much less vulnerable.

Powerful quantum computers will probably end up needing the same sort of infrastructure as datacenters or digital supercomputers currently do, and in fact will need very powerful digital computers to run them and process their outputs. (Currently they also require a load of expensive electronics and a lot of time consuming tuning and that's for just a few qubits let alone the thousands if not millions you'll need to do anything you couldn't just simulate on a laptop). You lose a whole load of quantum advantage (which you don't even have yet) if it takes you longer to figure out how best to transpile your quantum circuit into actual qubit operations than it does to execute those operations.

[dyqik]

That's unfair, he's completely unfamiliar with all the physics and mathematics of quantum computing, not just the cryogenics:

The technical explanation of exactly why encryption can be broken by quantum computing gets complicated quickly.

I can’t claim to understand the physics myself, but...

There's just about zero actual content, just people insisting that Q Day will happen and it will be bad, as if the digital computing power available to perform encryption won't be way ahead of anything a quantum computer can help decrypt for as far as we can see. It even says in the article that the iPhone already got upgraded to an encryption which is much less vulnerable.

Powerful quantum computers will probably end up needing the same sort of infrastructure as datacenters or digital supercomputers currently do, and in fact will need very powerful digital computers to run them and process their outputs. (Currently they also require a load of expensive electronics and a lot of time consuming tuning and that's for just a few qubits let alone the thousands if not millions you'll need to do anything you couldn't just simulate on a laptop). You lose a whole load of quantum advantage (which you don't even have yet) if it takes you longer to figure out how best to transpile your quantum circuit into actual qubit operations than it does to execute those operations.

I'd expect more like an experimental nuclear fusion reactor or particle accelerator facility than a data center or supercomputing facility. Big datacenters and supercomputers are lots of repetition of the same things, while an accelerator or fusion experiment are lots of infrastructure around a central core.

And if we have the technology to make this not true, we also already have the technology for quantum encryption on large scales, which can't be broken by a quantum computer.

[Tristan]

That's unfair, he's completely unfamiliar with all the physics and mathematics of quantum computing, not just the cryogenics:

The technical explanation of exactly why encryption can be broken by quantum computing gets complicated quickly.

I can’t claim to understand the physics myself, but...

There's just about zero actual content, just people insisting that Q Day will happen and it will be bad, as if the digital computing power available to perform encryption won't be way ahead of anything a quantum computer can help decrypt for as far as we can see. It even says in the article that the iPhone already got upgraded to an encryption which is much less vulnerable.

I feel like people are missing the point of the article or simply haven’t read it.

He’s not saying it WILL happen, he’s saying it might and that we shouldn’t be complacent given potential impact if it does.

We know there are things that we can do to mitigate it, as the iMessage example shows, but that’s based on a closed system. We should be looking at how global open standards can be updated too.

Another global pandemic was a hypothetical risk only till 2020. 2 or 3 years ago very few would have predicted how far generative AI would get to in such a short space of time.