Home Gotopia Articles Quantum computin...

Quantum computing in the cloud

Many firms attempt to make quantum hardware available to thousands of developers, a tool that scientists only began to conceive three decades ago. Quantum computing has brought us closer to the quantum computing speed and capacity required to revolutionize the world. But that isn't enough. There are still a lot of issues to be answered, such as how quantum computers function and how they will impact our world.

Share on:
linkedin facebook

Read further

Quantum computing today is seen as a promising technology for addressing various challenges and quantum computers are already accessible through major cloud-computing providers, coexisting with classical computing hardware. However, it cannot operate in isolation. Embark on a journey into the integration of quantum computing with classical high-performance computing (HPC). In a new GOTO Unscripted talk, Søren Gammelmark & Stig E. Rasmussen spoke to James Lewis about the significance of solving complex computational problems efficiently by quantum computing and its potential benefits for humanity, especially in areas like drug design, logistics, and finance. They dive deep into the heart of computational conundrums, showcasing quantum computing's promise and revealing Kvantify's ingenious fusion of quantum and classical computing.

What is quantum computing tackling?

James Lewis: Hello and welcome to GOTO Unscripted. This time it's me, James Lewis being the interviewer. I'm joined today by two people who are talking tomorrow in GOTO who's about high-performance compute and quantum computing. So, Stig and Søren Gammelmark, would it be possible for you to introduce yourselves to our audience?

Stig Elkjær Rasmussen: Yeah. Sure. My name is Stig Elkjær Rasmussen and I'm a PhD in quantum computers and currently working a lot with quantum computers, trying to make them do some interesting stuff.

Søren Gammelmark: Yes. My is name Søren, I have a PhD in quantum physics as well. I have been working with professional software development for 10 years about and then I'm now working as a software act ticket at Kvantify and try to bring quantum computers to real customers.

James Lewis: So, welcome. Good morning, good afternoon, good evening, whatever time zone you're in. And I guess we'll get straight on with it. So, I worked for a company called Thoughtworks many years ago, well, 5 years ago, 2018. We had quantum computing featured on our thing called the Radar. It's a publication. We publish every six months , but we haven't had anything about quantum computing since. It was sort of a super exciting thing that we were really excited about starting to sort of I guess take off back then. So, would you wanna just go back to, like, 2018, and was that around that sort of time, and talking a bit about where we are now with quantum computing?

Stig Elkjær Rasmussen: So, in 2018 was about the time I started my PhD in quantum computing. And back then it was just some sort of fluffy area where it was fun because quantum computing is fun, but it wasn't anything serious. But then doing my PhD, hardware providers started to actually make these quantum computers bigger and bigger, and suddenly they became big enough to actually do something on it to calculate small molecules, and I also never seen before. Then suddenly small companies started to pop up actually doing quantum computing stuff, and in the last couple of years now you can access quantum computers using Cloud software.

James Lewis: You'll be talking about this tomorrow. What sort of problems are you talking about solving?

Stig Elkjær Rasmussen: The biggest problem for quantum computing is probably chemistry, and that is drug design, drug discovery, and the making of batteries because chemistry is inherently quantum mechanical, so it fits perfectly in a quantum computer. But other topic could be finance, which is also quite suitable for quantum computers, or logistics.

James Lewis: Okay. So, and these are all things that you are actively involved in looking at at the moment, is it?

Stig Elkjær Rasmussen: Yes.

Søren Gammelmark: We see some applications are more likely to be possible in the near term. For example, chemistry applications within finance is probably a bit further off but we do have activities in all these areas where we just employ more traditional high-performance computing techniques.

James Lewis: What does that mean? So, high-performance computing, the Cloud, quantum computers, like, unpack this for me because I'm kind of intrigued as to how this works. Do you have your own quantum computer and then you, like, plug it into the Cloud somehow? I mean, how does this all this work?

Stig Elkjær Rasmussen: No. So, we're a software company, so we don't do any hardware. We just access everything through the Cloud and the quantum computers are also part of the Cloud and they probably will be for many, many years. And then we just access the hardware through the Cloud. Right now we're using the classical hardware, the normal high-performance computing. Then once the quantum computers are ready, we know how to use them because we're also doing that. Then we are ready to plug that in once they become good enough.

James Lewis: So, you're kind of simulating quantum computers and not simulating them, but you are using. When you say HPC, do you mean, like, big rack graphics, Clouds, and this kind of thing? What's the...

Recommended talk: Quantum & High-Performance Computing in The Cloud • Søren Gammelmark & Stig E. Rasmussen • GOTO 2023

Søren Gammelmark: Whatever compute resource is most reasonable. So, there are a couple of different ways to assess risk in finance, for example, different ways to optimize your exposure to risk. You can use quantum computers to solve some of these risk models, but you can also just maybe spin up a thousand computers and do heavy calculations for a short while and get the result. So, either use either graphics cards or just a lot of CPUs, depending on the algorithms you want to run, one or the other might be more beneficial. It actually very much in line with, well, right now maybe a problem is beneficial to run on the CPU, and in some years it might be good to do on a quantum computer, or maybe a new algorithm pops up and we need to drill in AGP. We are agnostic to watch that and the products we deliver are going to be agnostics, to note that.

James Lewis: Right. And it's a product company you work for. It's not bespoke solutions you are building or...?

Stig Elkjær Rasmussen: It is bespoke solution usually.

James Lewis: Yeah. So, obviously, it's not a generic, "Hey, I can solve all your chemistry problems."

Stig Elkjær Rasmussen: No, not at all.

Søren Gammelmark: No. But we want to be able to solve a range of chemistry problems at some point and sell it as a generic solution, right, not for a specific company.

Classical vs quantum computing

James Lewis: Oh, super cool. I mean, I know a little bit about HPC, not a huge amount, I have to admit. I know a little bit about quantum mechanics, not a huge amount. I have a very old undergraduate physics degree, so that was going way back. I didn't do well in advanced quantum mechanics. I will point I'll have to admit that. But I know some little bits in here and there, you know. So, for things like HPC, I mean typically you've got a number of different types of problem, right? Some which will be optimized for, as you say, like CPU or some that are gonna be massively paralyzable that you can run on big rates of graphics cards. How do they sort of then translate into the world of quantum computing? I mean, is quantum computing suited to one type of calculation to another. Does that make sense?

Stig Elkjær Rasmussen: You can't just take your problem on a classical computer and put it on a quantum computer. That's very difficult. It's not all problems that are suitable for a quantum computer. So, you have to create some software that does it in a quantum mechanical way. And that's not easy, that's why you need companies like us to help you do that. But again, it's not all problems that are inherently useful for a quantum computer. So, that's why we mentioned quantum chemistry, which is a thing that most people think will be the first application of quantum computers.

James Lewis: I've read a lot recently about, for example, you know, machine learning being able, like, deep fold. If you come across deep fold and then, like, machine learning has solved protein folding, apparently now this is a pretty exciting thing. Is it those sorts of problems that you're looking to address that?

Stig Elkjær Rasmussen: Not usually because machine learning problems usually have a lot of inputters of data and quantum computers are not very good at handling data because data is usually classical. That is something we know from the classical world, and that's not quantum mechanical. So, therefore, we need something that is quantum mechanical and not just a lot of data. Molecules are quantum mechanicals, so, therefore, data-driven problems are not suitable for quantum computers. You need something with just more...it's suited in a different way that you need this molecule structure that you can plug into the quantum computer.

James Lewis: I remember looking at quantum a little while ago, back in 2018, because I thought that was kind of interesting, and while I sort of with my classically trained program, my brain went, "Oh, okay. That's kind of..." So, how does the programming paradigm change between what I would consider, you know, kind of classical programming for even via HBC or whatever versus what you're having to do with quantum computers? Is there a complete shift of mind, mind shift?

Søren Gammelmark: You can program more on the quantum computer, but there's some really radical differences. For example, all calculations has to be reversible, which means that if you do something, you should be able to undo it. So, you cannot just multiply two numbers because if one of them is zero, you lose information. So, there's a lot of constraints. You cannot copy information. You used to be able to make a clone or whatever. You cannot do that on a quantum computer. It's physically impossible to do. So, there's a lot of practical... I mean, it's just a completely different way.

Stig Elkjær Rasmussen: Right now quantum computers are very small, so you have, like, 50 bits or quantum bits, and that means that you're doing a basic operation, like, a naught operation. So, that's what I'm coding is basic operations at the moment.

James Lewis: So, when you say basic, you don't mean, like, as in BASIC, the language? You mean...

Søren Gammelmark: No, no.

James Lewis: ...the basic, like, almost assembler, essentially plotting in. It's in logic exactly.

Søren Gammelmark: It's even more low level than assemblers, right, because you're addressing individual physical bits with individual logic operations, you don't have virtual memory, you don't have a stack. There's almost nothing. No abstraction.

Stig Elkjær Rasmussen: But always that is probably hopefully something that will come in the coming 5 to 10 years. But right now it's very basic.

Søren Gammelmark: But that's also because we don't know what the good abstractions are going to be yet.

James Lewis: So, could you foresee a time when myself with a tinkerer's interest in this kind of thing will be able to maybe write a classical program and it would be, like, transpiled into code that can run on the quantum computer, like, that kind of thing? Or is that way off in the future, we don't even know if that's possible to do yet, kind of thing?

Stig Elkjær Rasmussen: I think at some point, yeah. But right now, you have to have a problem that's very suited for a quantum computer, and that is probably the way it's gonna be the next couple of years. So, unless you're doing chemistry or something else, which is very suited for a quantum computer, you're probably not gonna write something that will be transpiled to a quantum computer.

Søren Gammelmark: And even if you took the path of having a more classical, programming language that you could transpile to the quantum computer, you would need some different language paradigms in order for it to make sense on a quantum computer as well.

How does it work?

James Lewis: Okay. So, like, into the nuts and bolts, how do these things actually work? Right? So, you've got 50 qubits, you're writing logic operations that are addressing each individual bit, and then you turn the thing on, right? I mean, how does it actually work? What happens? I'm intrigued by this. I've never seen one. It's super exciting.

Stig Elkjær Rasmussen: At the moment, there aren't really any best form of platform for quantum computers, but there are few different ways. One of them is called superconducting circuits, which is basically a normal computer chip that you supercool. Then it has some inductors and some capacitances and a type of non-linear inductor called a Josephson junction. And this Josephson junction is basically a superconducting wire with a hole in the middle. And then you would think, all right, now electrons can't pass through this hole, but a quantum mechanical effect is tunneling.

James Lewis: Sometimes they can. 

Stig Elkjær Rasmussen: In the quantum mechanical machine, they can tunnel through. And then depending on where the electron is, that's the state of the qubit.

James Lewis: Oh, right. And so, you don't actually know until you observe it. It's whether the electrons one side or the other, and that's the different approach.

Stig Elkjær Rasmussen: That's more or less the idea. Yes.

James Lewis: That's super cool. Is that the sort of thing you've been working with?

Stig Elkjær Rasmussen: That is one of the computers that we're accessing. We're also using someone called trapped ions.

James Lewis: Oh, is this outta Duke in... So, there's some friends of mine who actually were working on. This is in the U.S. is it or...?

Stig Elkjær Rasmussen: Yeah, there is the ion cube. They're trapping charged ions in magnetic fields, and then these ions, because they're atoms, they're quantum mechanical and they are building qubit because it's a two-level system. And then they can manipulate these ions and work them as quantum mechanical bits.

Recommended talk: Introduction to Quantum Computing • Jessica Pointing

James Lewis: You've then got the potential controversy around what Google are calling quantum.

Stig Elkjær Rasmussen: Yes. Google are using superconducting circuits.

James Lewis: Right. They are as well.

Stig Elkjær Rasmussen: There are lot of different platforms. Those were just two. There are at least five more on the top of my mind. I think those two are the most common one. Those are the ones that you can access on Cloud hardware providers such as AWS or Azure.

Current challenges in quantum computing

James Lewis: And just we've got these multiple different types of quantum computing system, I guess. And then I hear a lot about, I read a lot about, or at least I see a lot of fiz.org., you know, lots of questions about things like error rates, error correction, and how, hey, we've got this amazing ability to do this calculation, but the state it's only coherent for, like, a fraction of a millisecond. What challenges, I guess, are you facing at the moment with these systems?

Stig Elkjær Rasmussen: Right now they're very fragile. They're decohering very fast. So, you can do a number of gates operations on each qubits, and then they just start doing random stuff that you don't want them to do. So, therefore you need error correction. It's sort of, like, you have in a normal computer, you just don't notice it anymore because they're so good. But we need that at some point, and then we'll make perfect qubits that doesn't do random stuff. But that is also a couple year down the line.

Søren Gammelmark: Should be mentioned that error correction is a more complicated subject on a quantum computer because if you were to measure or read out the result of the computer in order to see whether or not a bit has been flipped, you destroy the state. You need to be able to measure in kind of an aggregate way over several bits in order to see if a parity is obtained or some other feature of the sequence of bits. And then you can do some operation based on sort of that measurement. So, that's also more complicated on a quantum computer. Everything is more complicated. Everything...

James Lewis: I mean, everything is more complicated, I would say. Sounds it. I mean, what's the why? I mean, you know, A, it's super difficult science, superconducting. Hey, we're just gonna super chill stuff down to, like, minus 250, like, C or something, right? I mean, that's not easy to do. So, the hardware is, like, super expensive. It's a difficult way to program. You have to learn a whole new approach to doing things. What's the benefit you think we're gonna be able to...I'm being hyperbolic here, right? I mean, this is the time to be hyperbolic. So...

Stig Elkjær Rasmussen: Hopefully, it's gonna open up possibility to calculate, for example, new molecules that we cannot calculate on normal computers because they become too complex. So, as I think Feynman said it when he proposed using quantum computers, that if you want to simulate something quantum mechanical, you better make a quantum mechanical simulator. And that is basically the idea of a quantum computer.

Søren Gammelmark: Also, because they're likely going to be slower than normal computers for a long time. There's this famous case of factoring. In principle, we should be able to factor enormously large numbers much faster than a quantum computer. That's of course a very interesting, from a maybe theoretical point of view if you're cracking codes. I mean, if you want to build new stuff, then being able to calculate on more different types of materials more accurately, it doesn't necessarily have to be faster, but it can be more accurate, for example. So, there's definitely a potential there.

James Lewis: Because I've read a lot about quantum-safe encryption, their future not perfect for security is no such thing. Right? Certainly when quantum computers get good enough at factoring fast enough, I guess, as you say. And is this gonna be another real area when we see lots of changes you think in the future in terms of security, privacy, encryption, these kinds of things?

Stig Elkjær Rasmussen: I think quantum computers are probably at some point, going to break the RSA encryption because they can factorize large prime numbers. But luckily, quantum computers has also have a built-in way of encryption. So, it will just be that our encryption would have to change, but it has to do that all the time. Encryption is not something static.

James Lewis: Yes, sure.

Stig Elkjær Rasmussen: So, at some point, quantum will just play a bigger role in encryption.

Søren Gammelmark: And it should be mentioned that even though RSA is vulnerable to quantum attacks, there are lots of other systems that are more resilient. So, there are classical ways around it. I mean, it's going to be important at some point, but it's not going to break the world in that way.

Quantum in practice

James Lewis: There's one that's on the tip of my tongue, but I can't remember it now. It'll come to me in a minute. Cool. That's super interesting. So, we've got a range of problems that can't classically be addressed by normal classical, like, systems. But the hardware there to do it, it's obviously hugely expensive, fusion complicated, and only about four places in the world can afford to do it. Right. But, hey, you can access it over AWS self-service. Fantastic. So, and we talk about things like chemistry and risk. So, I guess the whole stuff in finance around pricing models, there's a lot of quants already in banks who are already using these kind of the similar sort of quantum equations to model risk and things internally. What about things like slightly chunked up a bit? So, because biology is a classic example of modeling actual biological systems, so one level above the chemical. And that's super, super difficult, right? I mean, we obviously have to be able to do that by observation almost. Is that another thing that is...?

Stig Elkjær Rasmussen: That is hopefully some of the things that the quantum computer will be able to do, which the classical computer cannot do at the moment.

Søren Gammelmark: Because especially in pharmacological applications, sometimes you have a very small active site on an enzyme or a protein where if you could increase the accuracy of that calculation, you would be able to much more accurately predict, for example, which kind of molecules could have some desired effects. So, the quantum computer's mission to be useful at some point.

James Lewis: And are you seeing interest now from big pharma, from these organizations into the approaches that you are sort of championing?

Stig Elkjær Rasmussen: Yes. We're talking to several big companies and also the EU are backing us.

James Lewis: Nice smile. And the EU, just saying that, you know, that's awesome. That's super cool. That's super cool. So, I guess there's just a wholesome ecosystem around this stuff as well, right, in terms of making the hardware things, like, cheaper. There's been a lot of research. I've seen recently breakthroughs into not room temperature, superconduction, but being able to bring the temperature with new materials that are superconducting at higher temperature. Is that sort of stuff all feeding in? Do you think there's gonna be this sort of blossoming, like, a sort of compound effect where you get lots of things coming together, and suddenly we are gonna see this quantum computing?

Stig Elkjær Rasmussen: Well, yes. There's a lot of small steps that needs to be taken and room-temperature superconducting circuits would be very nice. There'll be a huge step underway. But currently, the biggest challenge is simply to make the qubits more stable against noise.

James Lewis: Okay. And so, how many qubits are we talking about? Is it hundreds?

Stig Elkjær Rasmussen: I think the record right now is around 400. Yes.

James Lewis: That's super cool. 

Søren Gammelmark: But they're not all connected, right?

Stig Elkjær Rasmussen: No, they're not all connected.

Søren Gammelmark: So, you cannot do arbitrary things with them.

Stig Elkjær Rasmussen: That's another problem that when you do cubits, you have to lay them out in a grid and they have to be close to each other before they can talk to each other. So, that's a difficulty of making everyone talk to each other.

Søren Gammelmark: It limits the number of which bits can you do.

Recommended talk: Quantum Computing in Practice • Murray Thom • GOTO 2021

James Lewis: Right. I see. Yes. Okay. Right.

Stig Elkjær Rasmussen: I can't just swap two bits if they're far away from each other.

James Lewis: Oh, right. It's one of these like an Uberist-type thing where what you need is actually a quantum computer to help design a quantum computer, right?

Stig Elkjær Rasmussen: That would be a very...

James Lewis: Like the kind of grid qubits or some kind of strange geometry that's gonna allow you to kind of more easily program them.

Søren Gammelmark: But there's no reason to think that's not going to be an ingredient in the future. Why not use the tool that you're developing to improve the tool? But it's unclear, right.

What’s coming up in the world of quantum?

James Lewis: So, what are you most excited about? That's stuff that's coming up.

Stig Elkjær Rasmussen: I think I'm most excited about the fact that I'm actually working with quantum computers. I mean, that's a childhood dream of mine. I saw it in science fiction movies. I was like, "That sounds really, really cool," and now I'm doing it.

James Lewis: And is it really, really cool?

Stig Elkjær Rasmussen: It is really, really cool.

James Lewis: How about yourself, Søren?

Søren Gammelmark: Well, I mean, I also did a PhD within related areas, not quantum computer specifically, but since then I've been doing normal computer software engineering. But it's extremely nice to get back to an environment that's very research-oriented and doing stuff with quantum mechanics that I know from my studies. And also I really enjoy delivering high quality, robust software for customers that just works the way they expect it to. So, I think taking the prototypes that people do in quantum computing away from, or bringing them out of academia and into industry so that they're actually usable in a very friendly way, that really triggers me. I think that's super interesting. And then at the same time, being allowed to use thousands of computers at the same time to do a calculations it's always fun.

James Lewis: That's always fun. Yeah. I mean, that's actually a really interesting point. I mean, I'm really curious to be, how do you test this stuff? For example, we talk about industrialization engineering, that is always...

Søren Gammelmark: That is a really good question.

James Lewis: Why?

Søren Gammelmark: Yes. Because I've been thinking a lot about it because I care a lot about tests because I want to deliver a robust software, and it's not clear at all because as we can attest to, running stuff on a quantum computer right now is super expensive. So, you're not going to put your quantum algorithm into a continuous integration test because that would just ruin you. And you cannot run the test on an input that makes sense. As soon as you go over 40 qubits, you basically do not have enough memory. Maybe you could do 50, I don't know what the limit is, but then you run out of memory on planet Earth in order to simulate it. So, maybe some kind of formal verification system could make sense, or maybe we just have to do extremely careful copy, to begin with. I don't know. But it's unclear what the good solutions are. I don't know if there are any good solutions right now.

James Lewis: That's really interesting actually. Anyone out there interested in testing quantum computing? There's obviously some opportunities. I mean, actually, at Thoughtworks, we've built an entire company more or less on the idea of, like, industrializing, making sort of new paradigms more robust and stable and kind of. So, for example, machine learning, that's a classic kinda Wild West kind of full of say cowboys. That's way too strong. But, you know, lots of data scientists who are applying really interesting statistical models and writing interesting algorithms and so on. Building interesting models, but without some of the sort of the techniques you would expect from modern software engineering, things like continuous integration or delivery or even versioning or, you know, even being able to read. You reuse the same model twice in your, etc. But there's been lots of advances in that recently. Things like, you know, continuous delivery from machine learning and feature stores and all these kind of cool things. So, I guess there's lots of opportunities or will be.

Søren Gammelmark: I think it's also an interesting question just for high-performance computing, because a lot of the existing tools out there are either tools with a very isolated domain where you kind of can test that domain. But if you're doing a larger calculation over a huge chemical system where nobody knows what the answer is, how do you test that? Because just running a single calculation might take a month or weeks, and, of course, to have any degree of confidence you need to test things in isolation. But the next step depends extremely sensitively on the step before. So, there's also a lot of difficulties in delivering robust software for that kind of set-up as well.

Recommended talk: Quantum Computing in Action • Johan Vos & Preben Thorø • GOTO 2021

James Lewis: So, I guess in things like meteorological kind of research, those sorts of things, where you're looking at cell-based kind of Monte Carlo-like algorithms?

Søren Gammelmark: For example, yes. Yeah. So, that's also difficult to test, I find.

James Lewis: You mentioned Feynman, of course, Monte Carlo was adjacent to him, right? Because that was for Neumann in Los Alamos. You came up with that with standards, you know? So, what was I gonna say? So, yeah. I mean, I'm intrigued by this idea now, sort of industrialization kind of process and testing the idea of a formal methods. But I guess to be the downside generally with formal methods is they take a long, long time themselves, right? But is it something that maybe is gonna be worth it actually investing in that sort of thing?

Søren Gammelmark: It's a good question. I mean, we have some people working roughly in that area, but it's not really clear if it makes sense. I mean, if the quantum program is complicated enough, you of course cannot verify it formally. So, there's definitely limits to what you can do. But if you can verify some of it, I mean, you just need to do what can be done instead of doing what's perfect.

James Lewis: Super interesting. Yes. So, in terms of high-performance compute, these are the sort of problems that you are working on already, the sort of chemistry spaces, these sorts of spaces, or is it...?

Søren Gammelmark: Yeah. So, right now, chemistry, pharma and financial applications. So, Monte Carlo and optimization, and chemical calculation stuff.

James Lewis: That super cool. Well, I think probably that's a good point to come to the end of this conversation. So, thank you so much, Stig and Søren, I murdered your names again, and I'm terribly sorry about that. But thanks very much for joining us for this episode of GOTO Unscripted. I've been James Lewis and see you another time. Bye-bye.