Quantum Computers: Behind The Scenes At IBM

Mark Fielding (00:00)
Scott, welcome to Thinking on Paper. Thank you for thinking on paper with us today.

Thinking on paper.

We speak about all technologies and at the beginning it was quite surprising for us that nothing divided our audience like quantum computing. There was a very noticeable for and against camp when we posted any videos on social media.

For the people who in our audience question quantum computing, what's your message for them today?

Scott Crowder (00:39)
Sure, so I guess the message today is that, quantum computers are real. It started about, the concept started about 45 years ago, so it's not an old concept. I mean, I'm older than that, so maybe I have a different perspective,

It's only been 45 years since even the concept of a quantum computer came to life. It's been about 10 years. It's been 10 years in a couple of weeks, actually, since IBM put a really baby five qubit quantum computer on the cloud and people could actually run a program on a quantum computer and get a result. So was kind of like sci-fi come to life a little bit moment. ⁓ And then I think there's a lot of hype and confusion in the field, which I think contributes

to that reaction. There's a lot of people talking about quantum computing that have never built a quantum computer, but still hype it like they have. And then there's some of us who have built quantum computers and made them available for people to use. ⁓ Where we are today is that we now have quantum computers that can run a quantum algorithm that is too complex to run on any classical hardware.

⁓ the confusing part, and this is why you get the bipolar response, is that doesn't mean it's better than running a classical algorithm that can run on that classical computer to solve the problem in a different way. And that's like the next step in quantum computing is when the quantum way of doing it is better than the classical way of doing that. And because we're not there yet, you still have two camps, in my personal opinion. ⁓

And until we hit that threshold and prove it to people beyond a reasonable doubt, then those camps will converge into one.

Jeremy Gilbertson (02:24)
Man, if we could only converge camps from a global human perspective, we would be, we better off in a lot of ways. But so, uh, I want to talk through what's really interesting. So recent paper reference architecture for a quantum centric supercomputer, some really cool things in, that announcement. want to start kind of high level when new technologies come out, there's always a need to kind of figure it out. And someone has to kind of call the ball and be like, Hey, this is where we are. This is what's working today.

Mark Fielding (02:31)
You

Jeremy Gilbertson (02:54)
This is where we're headed. mean, we have, and I want to ask you a couple of questions historically, why certain reference architectures worked and why certain reference architectures got pushed to the side. example, TCP IP, everyone got on board, right? ⁓ Unix, everyone kind of got on board. OSI, everyone kind of got on board. Vettiver Bush's, Memex, not so much. It was early.

and but it ended up being the kind of the same architecture that Berners-Lee did with the World Wide Web. So what do you think makes a reference architecture powerful, compelling, and one that'll serve as scaffolding moving forward?

Scott Crowder (03:34)
Yeah, I think it's probably two things. One, it actually has to solve a problem. So that's kind of fundamental. And some of the examples that you gave did solve a problem, but it didn't have the second one is it needs to get adopted. So it needs to actually solve something, and then it needs to get adopted. ⁓ And I think ⁓ for us, we're in the fairly early days of leveraging quantum computation to solve

problems in a practical, real way. And for us, the quantum-centric supercomputing reference architecture is our perspective on what computational resources and how you put them together do you need to actually run a problem and solve a problem. And because it's early days in quantum computing, think part of the fuzz, like you mentioned before in the two camps, ⁓

You got a similar fuzz in terms of what a quantum computer is. ⁓ Because for a lot of people, we're still in this like, sci-fi come to life, you the excitement about flux capacitors and spooky action in a distance and teleportation, like being used to do computation.

People lose sight of the fact that at the end of the day, all we're doing here is building a different kind of computer that runs a different kind of algorithm, which is for particular subroutines and a larger workflow. And quantum-centric supercomputing for us is how do you take that acceleration for that subroutine that quantum computation is really good at and make it part of a larger workflow?

⁓ When you get like just completely sucked into like the sci-fi come to life part, people are only focusing on that like subroutine piece of it and not thinking about the full problem. We need to start thinking about the full problem if quantum computing is really going to become useful for a lot of different applications. And we're starting to see the proof of that right now, but it's a bigger picture thinking that wasn't there a couple of years ago.

Mark Fielding (05:36)
Can I just ask?

at Jeremy's

examples, just to give me a frame of reference going forward on the, this conversation, it has to solve a problem and it has to be adopted. Were those examples the best examples of solving the problems that they solved?

Scott Crowder (06:05)
⁓ There's always debate there. mean, it depends on who you talk to. ⁓

I mean, you can do the OS 2 versus Windows thing if you wanted to, like giving an old IBM reference. But ⁓ yeah, the best technical solution isn't always the best adopted because of the ease of adoption, et cetera, et cetera. Frankly, like in our program, this is why we're so heavily fixated on making the technology externally available. Actually, in my background, like I'm a long-term IBMer. I started out in semiconductors. Then I was head of corporate strategy for technical strategy.

Then it was the CTO for IBM Systems, which is our infrastructure, server, storage, blah, blah business. Actually, the reason I got so geeked up about quantum computing 10 years ago, yes, I was excited about the sci-fi come to life part of it, but I got really attracted to it because IBM research took the technology and put it out for everybody else to do research on it, as opposed to having it in-house for IBM research to do it, which I think... ⁓

Historically, IBM has been really good at creating technology. I think we're a underappreciated for that. We haven't historically been as great as getting people to make it easily adopted. And that's personally why I got so passionate about it because it was like baby technology, like first steps in the real world. Can we really heavily lean on the...

making it accessible, getting people to use it, et cetera, et cetera. So yes, I'm really proud that we've got the largest fleet and the most powerful system and blah, blah, blah, blah. But I'm actually more proud that almost 6,000 research papers have been published by people using our technology. And since it didn't even exist 10 years ago, that's a pretty staggering statement.

Mark Fielding (08:02)
Is quantum different to those? And we will get to this announcement because I'm very excited about that. But quantum, you have these different modalities. You have trapped iron, you have spin qubits, you have superconducting. Again, those examples that Joe mentioned, one survived, the didn't. Is quantum computing the same in that one modality will survive or is it different in that all the modalities could come together and solve those problems as one global quantum computer or will they

Scott Crowder (08:10)
Yep.

Yep.

Yep. I think that probably maybe it's my background, but to me the best analogy is probably silicon versus germanium versus gallium arsenide versus indium phosphide, et cetera, et cetera. ⁓ And it's not like germanium, which was the original basis for transistor radios back in the 50s, like ever completely went away or indium phosphide for specialized things ever really went away.

Mark Fielding (08:31)
by the wayside too.

Scott Crowder (08:55)
⁓ But the silicon transistors, CMOS transistors just became dominant. I'm old enough to just join IBM at the end of the debate between bipolar and CMOS. ⁓ And I think it's more similar to something like that, where in my opinion, yes, one of the modalities is probably going to become dominant. ⁓ And once it does, it's going to be hard for the other modalities to catch up just because of the

like infrastructure, et cetera, et cetera, that's built up around it. They've all got pros and cons. And again, it depends if your headset is sci-fi come to life or your headset is I'm building a computer. Our headset is we're building a computer. So fundamentally for us, the primary thing is computational throughput. Because at the end of the day, that's going to drive cost efficiency, practicality, et cetera, et cetera.

⁓ So that's why we think certain modalities are more likely to ⁓ So for us, superconducting qubits is the right trade-off of being fairly easily manufacturable because it uses silicon fabrication techniques and also fast, which makes it feasible as a computational platform.

Jeremy Gilbertson (10:13)
Got it. Mark, just a quick note. You're crackling. You may want to check your connections ⁓ over there. So maybe do that while we're moving to this question. ⁓ So Scott, yeah, I love your reference back to the silicon adoption. That became the path. That became the spot. ⁓ How important, you mentioned, you referenced this too, how important is accessibility to the technology?

⁓ in the means of adoption. Let's talk through that a little bit.

Scott Crowder (10:45)
Yeah, from our perspective, it's critical. So overly simply, from our perspective, there's two things we need to do to take quantum computing and make it to unlock its value. The first is we need to build more and more powerful quantum computers that's both hardware and software. That's kind of obvious. That's kind of mostly on us and other vendors that are out there. But the second one was equally important is

algorithmic discovery of like stuff you run on quantum computers. I think we take it for granted that for binary math, we've been developing algorithms for centuries as humans.

And when digital quantum digital classical computers came along in the 40s and the 50s, there was this huge increase in algorithmic development in the 50s and 60s that kind of like underpin a lot of what we run on classical computers today. Fast Fourier transform, blah, blah, blah, A lot of examples. ⁓ We're really early stages in humans thinking about algorithms that run using quantum information science and how to apply them for applications. ⁓

So it's a huge part of it and part of it is the algorithmic discovery. So part of it is getting the computers out there so people can explore and make sure that it works. But I think what is, I think a little bit less appreciated, and bear with me, this is probably a long answer, but a little bit less appreciated is how do I think about the problem?

and break up the problem so I can leverage this tool effectively. And there's been a huge, like, sea change in that in the last couple of years that I think people are not appreciating quite as much as they should. So that's kind of, what's underneath the need for quantum-centric supercomputing is, OK, I've got this new computational tool as quantum. It's good at certain things, but...

It's not big enough yet to run the entire problem on the quantum computer. Does that mean I wait for 20 years or 10 years or however long? From our perspective, no.

because people didn't wait for classical for that either. ⁓ And what you're seeing now is people are starting to run real world scale problems on quantum computers. So for example, Cleveland Clinic published a paper where they basically show that they can simulate a protein of interest in life sciences with 303 atoms.

And they did that by breaking up the problem and taking the really hard parts and using a quantum computer to simulate the really hard parts and using classical to basically break up the problem and stitch it back together.

So they can basically simulate that protein, which is important for life sciences because the oxidation of that protein is actually one of the things that causes vaccines to lose their shelf life. So it's a real problem that they're really trying to simulate. And they can simulate it on quantum computers today. ⁓ So it's that.

work on how do I think about the problem? How do I apply quantum algorithms? And how do I break up the problem to run it? And you can't do that without access to computers. It's kind of like the AI world, right? You can't do chat GBT without access to GPUs today, mean, primarily. Like it's the same thing. You're not gonna be able to develop the quantum enabled applications, quantum algorithms without access to the underlying.

Mark Fielding (14:32)
how complex of a protein is that Cleveland study simulating?

Jeremy Gilbertson (14:32)
So Mark, we.

Scott Crowder (14:39)
So that one's 303 atoms. Let make sure I get the pronunciation right. It's tryptophan. And like I said, it has real-world application. There are people who ⁓ actually want to simulate that for real-world application. And like I said before, the confusing part here and the reason why there's two camps is now we're showing that we use an app-on-large metaphor. And I don't know if it's going to go over.

know, quantum computers run quantum algorithms. Classical computers run classical algorithms. So it's not like a GPU where you're running the same algorithm on a GPU and a CPU. It's just you're parallelizing it on a GPU so it runs faster. You're running different algorithms. So you're comparing a quantum orange with a classical apple.

What I'm saying is that right now we've come to the point in respond computers is that that orange, you can't run it on the thing that runs apples anymore. It's too complex to run using binary arithmetic. And this is the important part I'm trying to stress. People are now have figured out how to use that orange to run real world problems that they would like to solve. Where we are on the cusp of, but have not yet claimed that the orange is now better than the apple.

We're now saying that you compare the orange and the apple, but we're not yet saying the orange is better than the apple. And that's why you still have the two camps, right? But we're really close.

Mark Fielding (16:10)
don't know on protein simulation,

I don't know how many 303 atoms, I don't know how complex that is. So if you wanted to compare it to...

Scott Crowder (16:18)
⁓

It's a relatively moderate size protein. But the exciting thing is that you can leverage the same approach to run 10,000 atoms. So now that people have figured out how to break up the problem and leverage the quantum computer, they can actually scale it significantly larger.

I highly anticipate that you're going to start seeing publications around those larger scale very soon, like in the matter of month.

Mark Fielding (16:51)
Is this kind of

Jeremy Gilbertson (16:51)
Let's.

Mark Fielding (16:51)
like learning

how the tool works or is it the computational power of what you're able to do at the moment that is preventing or keeping it at 300 before it gets to 10,000?

Scott Crowder (17:02)
So it's a little bit of both, but it's primarily the first. So like I said, before people were thinking of this as, how do I fit the whole problem into a quantum computer? This is like five years ago. And algorithms to do that.

People are now thinking about, now that the computational tool is real, that the quantities of quantum computers are real, people are thinking about how do I best leverage them and realizing that if I use similar approaches that I use classically, I can use this quantum computational tool more efficiently and more effectively. So like the Cleveland Clinic example, they were doing like eight atoms, 14 atoms as recently as nine months ago. And then they had this breakthrough in

algorithmic discovery on a better algorithmic approach, which improved the accuracy.

And then they had a breakthrough in figuring out how to break up the problem. So how do I construct the problem differently, which allowed them to increase the size of the atoms. And there's no reason why they can't, now that they've figured this out, go from 303 to 11,000, 12,000, and now not just simulate this protein, but simulate that protein insolvent in a much more complex thing, which is what is needed to do the real-world digital twin kind of stuff that they want to do in health care life sciences.

And at the same time, we're continuing to improve the quality of the computer, which will improve the accuracy of the simulation and ⁓ make that crossover point of crossing over so the orange is better than the apple, if that analogy holds.

Jeremy Gilbertson (18:40)
I think it yeah, I think it does that. This all threads the needle for me a little bit that there's a tremendous opportunity in the quantum market just in general to figure out how to break up those problems and how to reconnect those problems. think that's a big, I wish I ⁓ had the capacity to do that because I think there's an interesting thing there. But let's talk more about Cleveland Clinic and what they did because I think whenever someone talks about computers and proteins, they tend to gravitate right towards, ⁓ Alpha Fold,

Scott Crowder (18:52)
Yes. ⁓

Jeremy Gilbertson (19:10)
did this thing and they did this structure modeling and ⁓ in but I think what is interesting about what Cleveland Clinic is doing, they're taking it beyond structure and they're actually actually talking about the and I did some basic research on this. The model works on the electronic structure between the molecules working together. It's more complex than just hey, here's the studs and the framing. It's actually how things communicate. Can you unpack that a little for us?

Scott Crowder (19:35)
Yeah, I I might be oversimplifying this a little bit, but you can think of they're using quantum simulation to kind of like do more of like a digital twin-ish kind of thing of like simulating it. Whereas alpha photo and AI is basically taking the known information that we've got and trying to interpolate and find better ways to like use that data to make a better prediction of what might work.

So in one case, you're basically doing the actual simulation of the molecule. In the other case, you're trying to use known human data that we've like wet benched or whatever, and basically try to predict what might work better based on that. So the way that we think that those things will work together in the future is quantum focusing on simulating the world better, AI basically trying to extend.

find ⁓

predict better basically based on what we know what might work. ⁓ There's more complex ways that quantum and AI can work together, but the healthcare life sciences is probably the easiest like hand wavy way I can explain.

Jeremy Gilbertson (20:47)
So we're here to talk about this exciting announcement that you guys have. so it's a three phase architecture, ⁓ kind of based on what I understand. Can you help us go through this phase by phase and help us understand it from a high level?

Scott Crowder (20:51)
Yeah.

Yeah, mean, I think the phases are more based on the maturity of basically the connectivity between the HPC elements and also the classical elements and the quantum elements in a quantum-centric supercomputer. ⁓ But at a really simplistic level, basically you can think about it as we're going to have quantum processing units or quantum computers that are going to

run those quantum subroutines. And we're going to have classical resources that we know and love today, CPUs and GPUs, that are going to continue to be good at what they're good at. So how do you build an architecture that allows you to run the subroutines on the underlying processing units in a way that's efficient? ⁓

And that's fundamentally at its heart what the architecture is about. So our kind of...

Mark Fielding (22:10)
So just on that, so do you mean

the choice between what part of the problem is done by which by either the QPU or the GPU? that what you mean?

Scott Crowder (22:24)
Yes, exactly. Exactly. Exactly. And there's levels here. And this is why it makes it a little bit complex. But there's levels of the onion here. There are things that need to be done to create a large scale fault tolerant quantum system that need to run on the quantum processing unit. And there's pieces that need to be run on.

classical accelerators, whether they be ASICs or GPUs or CPUs. So that's kind of like the inner part of the onion, which is a mix of quantum processing units and classical resources. ⁓ That's kind of like, think of it as inside the box, inside the first box. ⁓ But then there's a second layer of classical resources that need to run ⁓

kind of like the classical parts of the workflow. ⁓ So you're very similar to CPUs and GPUs working together today, or GPU-heavy nodes and CPU nodes. So you need some way to basically coordinate the work across those. And you need some kind of architecture that kind of lays out how are these connected, what bandwidth requirements are there, what latency requirements are on there, those kinds of things.

So the architecture is really basically saying, I've got these piece parts. This is how we're going to put them together. And from a fundamental point of view, ⁓ we're trying to lean into making it accessible, making the quantum resources accessible to the tools that people are already using to orchestrate ⁓ computational resources today.

Jeremy Gilbertson (24:14)
So the fault

tolerant scalable thing is kind of the down the road, like, hey, we're trying to point that direction. What's happening in the near term? Like not today, but like in the next year or so that is exciting and about this architecture. And we can even go in the technical weeds a little bit here. Like, what are you excited about in that realm?

Scott Crowder (24:22)
Yes.

Yeah, so what I'm excited about is some of the work that we did with RIKEN, which is a scientific research institution in Japan that has Japan's largest HPC cluster. ⁓ And we actually have a quantum computer in the same building, which allows us to have direct connectivity between the quantum.

system and the classical systems, you to really explore this QCSC architecture. ⁓ And what they're doing today is figuring out how do we orchestrate these two resources so they can work well together. And it's a little bit more challenging in the sense that

A lot of the orchestration today is you've got a cheap resource and an expensive resource, and you're trying to don't care too much about managing the cheap stuff, but how do I optimize use of the expensive thing? In this case, in a quantum-centric supercomputer, you've got your HPC, AI resources, and your

quantum resources and they're both kind of expensive. So you basically want to come up with a way to orchestrate them so you're not wasting time on either side of it. And I know I'm way in the operational weeds here, but this is the kind of thing that's important to make it real. And what we worked on with Riken was the workflow and the orchestration for a chemistry experiment similar to what I described with Cleveland Clinic of how do you run a chemistry workload.

that roughly runs about half the time on the classical resources, runs half the time on the quantum resources in such a way that I'm not letting either side of them just sit idle. And I'm fully using the computational on both sides.

So there's a couple pieces of the architecture that are important. One of them is just the pure technical. How do I optimize the workloads so it's most efficient, gives you the best answer across both of those. Like that's the splitting up the problem thing we were talking about earlier. And optimizing each side of that, the software on each side of that so it runs most efficiently. But there's also the piece of it of like, how do I get quantum plugged into the orchestration software they use to orchestrate their HPC stuff?

so that I can orchestrate both the HPC stuff and the quantum stuff so neither of them are sitting idle. So that's the other piece of the work that was done there and they were able to demonstrate that A, they could run the chemistry experiment and get good results ⁓ and B, that they could do it in such a way that the two resources were working together and orchestrated so it was sufficient for the computation.

Mark Fielding (27:26)
I've got a Jeremy thought it's okay.

Scott Crowder (27:26)
Sorry for the long answer, but

that's the kind of thing that we're doing short term. then, know, like I don't want to overcomplicate, but the phases are like as the quantum computer gets more complicated under the covers, you know, how does that work? And as we get to more and more, ⁓

longer programs, like bigger problems, like how do we even look at more tightly connecting that as opposed to loosely coupled orchestration, if it's needed.

Mark Fielding (27:56)
Jeremy, I've got a thought, I've got Jeremy thought that I just want to dump out there in a minute. Before I get to my thought, does the phase one, phase two, phase three, does that correlate in any way to the IBM chip? So you're talking about Starling in 2032, I think is on the roadmap. Do the phases relate to that?

Jeremy Gilbertson (27:56)
So Mark, you got so.

Scott Crowder (28:13)
⁓

2029. yeah. It's sooner than you think. Yeah. So, yes, because we're trying to intercept our ⁓ roadmap for the quantum piece of the quantum-centric supercomputer with the classical piece of the quantum-centric supercomputer, how they work together.

Mark Fielding (28:18)
Thank you for correcting me. I'm too pessimistic.

Scott Crowder (28:43)
Yes, the phase three is kind of aligned with the Starling 2029.

Mark Fielding (28:49)
Is that still on?

Because 2029 is rolling up very quickly as I look at my calendar in the corner. Are you still on track for 2029?

Scott Crowder (28:54)
Yeah, that's why I basically... Yeah. Yeah. So

we are... So we... For the reasons we mentioned before, like you've got two camps, you've got a lot of hype. There's lots of different kind of like confusion hype, like I mentioned, like...

you know, apples to oranges confusion. There's the fault tolerant, what really fault tolerant is confusion. There's the, do you really have a quantum computer or you just have PowerPoint and like making it sound like you've got a quantum computer ⁓ hype. ⁓ But we've put out basically a roadmap of what we're releasing to our clients year by year by year.

And we put out a roadmap of key internal development milestones year by year by year. So you can like follow along at home of like, we on track for 2029?

So right now, we're on track for 2029. The question I always get is, what's your error bar? And what I would say the error bar is, I don't think we're going to pull it in more than six months. I think there's some really challenging engineering work necessary to put it all together. And I don't think we're going to miss by more than a year. And I don't think it's going to be fundamental. It's going to be like, there's a lot of engineering work that we've got to do.

So I would say minus six months plus one year is my like 80 % like bars.

Mark Fielding (30:18)
That's going to be all over the internet, Scott. my little thought, and then I'll leave it to you, Jamie. I was just thinking, we started the conversation speaking about solving a problem and then how that technology is adopted. And I'm thinking about chemistry and material science and molecular simulation and where that happens. So it happens in the universities. It happens in the big pharmaceutical companies. It happens in clinics.

Jeremy Gilbertson (30:20)
You

Mark Fielding (30:50)
And then I try to connect that to the technology that you're building. And I think about learning curves and I think about removing what already in place and what's been working for these research institutes for so long. And I think about the scientists in the lab and how much say they have in what the, how the computations are run and I, and how do you change, update, evolve that system that seems very embedded.

culturally or maybe I'm completely wrong on that because I'm not a scientist, but that's what I was thinking as you were speaking.

Scott Crowder (31:27)
I think there are two levels here, right? So there's the level of the people who are building the fundamental algorithm approaches and building the software assets or wherever you want to call them that ⁓ instantiate that that are repeatable. And then you've got the people who are going to leverage those models or leverage those software assets, right? So.

⁓ We're currently at the state where we're still in the algorithmic discovery and application research piece, where you do need to basically... ⁓

still improve building those assets. So like in the Cleveland Clinic, part of what they did was kind of use raw concepts. We may have had some software assets that made it easier, blah, blah, blah, blah, but it wasn't like a black box that they could just put their input, get the answer out and.

and use. We're going to get there. Like we're going to get to the point, you know, probably in the next three, four years where you're going to have like chemistry solvers that are more like black boxes that like a wider set of computational chemists can basically just use as a tool as opposed to using, know, pick a tool today, Julie, blah, blah, blah, blah, you know, tool today that they use. So I think we'll get there. But in the short term, really the work is

⁓ having the people who have domain expertise and math expertise building these algorithmic approaches for quantum, proving that they work, running them on real quantum computers, et cetera, et cetera. ⁓ So I think it's gonna be, I apologize for the complicated answer, but I think it's gonna be stages.

So we're at the first stage like we were in the 50, 60, 70s, you know, in classical computing where people are going to need to build those black boxes and the people who can build those black boxes themselves are going to get early mover advantages in, you know, simulating chemistry or optimization for finance or whatever. But eventually, eventually like any other, it really is just a

different computations. At the end of the day, it really is just a different computational tool. just like we're more like in the 50s, 60s, 70s than the like 80s, 90s where I think a lot of us have our head trained of like how computers are used. We're back to, at the point still where understanding compilation still can give you an advantage as an example.

Jeremy Gilbertson (34:09)
makes a lot of sense. I've got one more kind of, and this could be the probably the dumbest analogy I've ever thrown out on this show, but it might land, you never know. So with this plan, with this ⁓ reference architecture, you imagine, you know, classical compute on one side, quantum on the other, and they're walking down the same road. And like the initial phase, they're kind of walking about 10 feet apart. And then as we move to that third phase, maybe they start holding hands a little bit.

And then maybe to the end result, maybe they become kind of one thing together. Is that where we're going to eventually see is like these two things merging to the next version of what computers are?

Scott Crowder (34:51)
I think you're going to have a merge in this quantum-centric supercomputing thing. So you're going to run certain things on that. ⁓ I think you're also going to have additional classical resources, classical computing environments that are going to also interact with it. So I think it's going to be ⁓ not to over-complicate things, but I think it's going to be like layers of an onion. ⁓

And then, so yes, classical is going to work much more closely with quantum and vice versa. ⁓ And the boundaries of where...

the supercomputer boundary is versus what you run outside of the supercomputer is going to depend a lot on how the workloads evolve. Just like, you know, there are more than one computing environment today in classical. It's not like the GPU environments are the only compute environments and they've sucked everything else into like the AI mega data center, right? So it's going to be similar. You're to have a quantum centric supercomputer that's good at certain things. It's also going to have to work in an overall

workflow with other classical. ⁓

Jeremy Gilbertson (36:06)
Got it. Got it. Cool. Thank you. ⁓ Scott, let's have some fun with this one.

Mark Fielding (36:08)
Jeremy,

wasn't the dumbest analogy you've ever done, but it was up there.

Jeremy Gilbertson (36:13)
I may we can unpack what the

what the dumbest one was in the future. ⁓ Scott, let's have some fun with this one. So let's imagine somehow Richard Feynman comes back to life and he comes across this paper and he reads the paper. Imagine what how he would react to it. Let's just have some fun. There's no wrong answer here. Like how would he react?

Scott Crowder (36:16)
Yeah.

I think you probably, so it's not, this one maybe less Feynman, but like, you know, they announced the Q4 bio winners today. They mentioned that Cleveland Clinic one. So I think those, I think you would be more excited about because like you had this postulation back in 1981 that like.

Let's use quantum information science for computation. ⁓ The fact that people are doing that for real world examples today, I think.

I think it would be a little mind blowing for him, honestly, even though he was the one who like postulated it. You know, there's the half-mobius thing, there's the neutrons kind of thing. There's now a lot in the last, this was kind of my big point before, in the last like five months, you've seen like a major uptake in like people simulating stuff on quantum computers that are real problems they want to solve. So, you we put out this paper on this half-mobius molecule thing that like the...

researchers at Zurich had this build this molecule that doesn't exist in nature with this crazy half-mobius property. You go around, like goes halfway around like, ⁓ crazy stuff, but they then simulated on a quantum computer to kind of prove that they had built it.

You've got the neutron scattering experiments at like DOE where they use the quantum computer to simulate the experiment that they ran. You've got the Cleveland Clinic thing simulating the protein. I think this is the thing that's like Feynman's vision come to life. I mean, I didn't know him. Like, I don't know how excited he would be about like, know, optimization for finance, but I'm just guessing he'd be more excited about the, like his concept that you.

Mark Fielding (38:21)
Did Feynman paint this? Feyn-

Scott Crowder (38:29)
can use this to really compute stuff that physicists and computational chemists are interested in has come to fruition.

Mark Fielding (38:37)
Did did Feynman write about this classical quantum unification or was he very much of that you'll have quantum computers and there'll be a separate idiom on their own? Or would he be disappointed by the reliance on the classical infrastructure?

Scott Crowder (38:59)
Hmm, I doubt it. Like I said, I didn't know him, complete speculation. ⁓ But I think just the fact that you're leveraging quantum information science in order to simulate quantum mechanics on a real physics problem, I think would be pretty damn exciting to him because that was the big picture concept.

Jeremy Gilbertson (39:25)
Let's, can we transition to data centers for a second, Mark? Are you good with that?

Mark Fielding (39:30)
And before data centers, just one more question before we get on to data centers. Then, Have you changed your mind about anything in the past year about quantum computers that you didn't think you would?

Scott Crowder (39:43)
I don't think I've changed my mind. ⁓

I think I've been surprised by how quickly some of these algorithmic approaches to do larger scale problems have really picked up. ⁓

But I mean, I'm living it daily. I've been less surprised by other things. I am continually surprised by some of the hype and how the hype is interpreted. I shouldn't be at this point.

Mark Fielding (40:14)
Or yours, Jeremy.

Does it does

something sometimes you just shake your head at all of the hype and all of the nonsense and say, oh, man.

Scott Crowder (40:29)
Probably can't comment on those, but yeah.

Mark Fielding (40:33)
Jeremy, data centers.

Jeremy Gilbertson (40:35)
Yeah. So, so you spent, mentioned being CTO of, uh, I guess what was it? The infrastructure side server compute stored all that kind of stuff for, for quite a long time. So you're well aware of the impact of it infrastructure on data center infrastructure and the correlations and that sort of thing. So today there are data, there are billion dollar data centers being built today that aren't really incorporating some of these requirements in. And I lived in the data center world. I developed.

Scott Crowder (40:42)
Yep. Yep.

Jeremy Gilbertson (41:04)
requirements, technical programs, that sort of thing. Like, what is this architecture going to do to the future of data centers is my question.

Scott Crowder (41:15)
Yeah, I mean the good news, bad news is that ⁓ the rise of AI mega data centers ⁓ have made it really easy for you to plop a quantum computer in any of those data centers. ⁓ Our requirements are tamed by comparison, actually.

I think people assume quantum computing, really complex, et et et cetera, that it must be massively power hungry, like all these requirements, et cetera, cetera.

⁓ Basically, we require water cooling for our systems and not to get the heat out primarily, but to keep the temperature steady across everything because variations in temperature leads to variation in signal propagation, which leads to timing kind of situations. So that's why we water cool. ⁓ It requires.

very minor power compared to AI. So rule of thumb is like a state of the art quantum computer, is about the same amount of power as one rack of AI. And that's true today, it's little under actually today. And that's true for the system in 2029 also. You this large scale quantum computer.

Mark Fielding (42:30)
What's one rack of H100s

and like how much is that draining off the grid?

Scott Crowder (42:35)
God,

when you get into like 2029 timeframe, you know, they're estimating well over a megawatt per.

Jeremy Gilbertson (42:43)
Okay, let's pause right there. Just just for reference, Mark, like, I don't know, probably 10 years ago, five kilowatts a rack was like, chunky was like pretty big, right? And then we've gotten to 20 kilowatts and 50 kilowatts, 100 kilowatts, but you're talking about a megawatt in a single rack, which used to be like the whole data center, right? Yeah.

Scott Crowder (43:05)
Yeah, it's a little bit insane. Yeah.

Mark Fielding (43:07)
What about the water?

Scott Crowder (43:07)
So.

Mark Fielding (43:08)
What about the water consumption? If we get on the cooling, could we get a figure on that?

Scott Crowder (43:13)
I don't know what the AI data centers, what our consumption are these days, ⁓ ours is fairly tame. Like, you know, if you've got water cooled in your data center, we're like a blip on the.

Mark Fielding (43:16)
or the quantum yours if you put it in a day.

Scott Crowder (43:28)
And then the weight of the system, again, is the weight of these AI racks is intense. So if you can solve that problem, the weight for our systems is not an issue. So the only constraint that we've run into a couple times is our systems are higher.

Jeremy Gilbertson (43:28)
Less than the water for your coffee pot, right? Yeah.

Scott Crowder (43:50)
⁓ Again, so these AI mega data centers, no problem whatsoever, but in like a traditional data center, and even a traditional big data center, but like a departmental data center-ish kind of thing, our systems tend to be a little bit higher. Just just height, physically height, that's it. They're taller. ⁓ So in most of our deployments, it hasn't been an issue at all, but it's usually the one that is the question mark. The other ones are...

Jeremy Gilbertson (44:05)
What do you mean by higher? like...

Mark Fielding (44:06)
Physically higher.

Is that a design

choice or is that just a necessity?

Scott Crowder (44:23)
It's a little bit of both. It's a lot easier to get the signals up on the top, up over the top, because ⁓ there's some advantages of basically your cryogenics like hanging down as opposed to coming up. ⁓ There's somewhere behind me.

Mark Fielding (44:45)
if you have some show and tell Scott, we'd love to, you can, don't know you can. ⁓ yes. Yes. So what are we, what are we looking at Scott?

Scott Crowder (44:48)
There we go. So

that's kind of like an example of the thing that hangs down inside what we call the fridge. So that's like where the qubits live. ⁓ So that's where the qubits live. And they live there because, like, overly simplistically, they need to be isolated from the rest of universe. So you do this quantum information science by...

Mark Fielding (45:10)
I love that sentence so much. love this. But my favorite sentence in quantum, think.

Scott Crowder (45:15)
So this is the sci-fi come to life part. So for us, it's like 1.5 millikelvin, 15 millikelvin. ⁓ So it's really, really freaking cold. But it's also ⁓ light isolation, vibrational isolation, et cetera, et cetera, et cetera. Yeah, it's like a Bond movie. Either you're cryogenic at some point, like even the trapped ions are cryogenic at some temperature.

And then it's either just cooling it or you shoot friggin' laser beams at it. It's like one of the two in order to get the entropy out. So that's kinda like the sci-fi come to life part of it. But anyway, so in order to basically get the signals most efficiently in and out, it's easier to do it over the top. If you're doing it over the top, it means that it's a little bit higher.

Jeremy Gilbertson (46:06)
So, so Cubits, Cubits are very distractible. They require, they require some focusing, right? Which you were, which you're talking about and, and you're talking.

Scott Crowder (46:11)
Yep. Well, the trick

Mark Fielding (46:12)
their own universe.

Scott Crowder (46:14)
is you want them to be distractible because if they're not distractible at all, you can't program them or they're really, really slow. So it's trying to find the right balance of how distractible do you want your qubits and giving them the right medication so they don't lose focus.

Jeremy Gilbertson (46:33)
What a reference.

Wow. ⁓ all right. Well, so speaking, speaking of being higher and having the, ⁓ having the computation happening higher, let's talk about orbital, orbital data centers and let's talk about quantum in space. And, know, I, I know you probably don't think a lot about that, but my job is to weave and connect the dots on the show. We talk a lot about space tech, talk a lot about orbital data center, star cloud and others. Is there a place for quantum computing in space?

Mark Fielding (46:34)
That's great.

Scott Crowder (47:04)
I think it would be similar. like overly simplistically, I don't think it's for cooling or anything like that. So the real question is, you know, space, energy, et cetera, et cetera, et cetera. You know, honestly, it's not something we've looked at very carefully. ⁓ But our.

Mark Fielding (47:29)
So when Elon

Musk is on Twitter saying, they should put quantum computers on the South Pole of the moon in the dark shadows of the craters. Is he doing that? Is he saying that just to get likes or is he saying that because he believes that that's actually a good idea?

Scott Crowder (47:47)
I actually believe this is not a good idea. don't know why Elon only uses it. I mean, in fairness, I would need to think through his hypothesis for why that is. There's a bunch floating around of helium sources, et cetera, et cetera, et cetera. ⁓

Mark Fielding (48:14)
helium 3, I'm sorry to interrupt your train

of thought, but is helium 3 needed in the quantum computing industry?

Scott Crowder (48:21)
It is for the, you're using cryogenics to isolate it down to the temperatures that we are, it is, you know, it's a real thing. It is something that we need to address in the long term. It's not a short-term problem. It's more of a, if you believe this technology is going to scale and you've got...

Mark Fielding (48:37)
Do you have a shortage of it?

Scott Crowder (48:45)
lots and lots and lots of quantum computers by the middle of the next decade, then it is something that we need to consider.

⁓ So the answer to question is yes, it's not a problem today, but it is something that we need

Jeremy Gilbertson (48:55)
Excellent.

Well, let's, let's, let's try to, ⁓ like you said, all planes come down eventually, right? Let's try to, let's try to land the plane a little bit. ⁓ we, talk about not just technology, but what, technology means for humans, human solving problems, humans doing work, that sort of thing. do have one question that, ⁓ Kevin Kelly left us, to ask all of our guests. And I want to, I want to ask you this question. Little bit, little bit social question, little bit tech question, but what do we want?

humans to be and how does technology like quantum computing potentially help us get there?

Scott Crowder (49:43)
I would say we humans to be happy and productive. And I think it's... ⁓

Again, for me, quantum computing is just a new computational tool. We don't really get at the heart of it. So the question is, what kinds of applications are you going to be able to do better with quantum computers? So from my side, I think there's a lot of societal benefit in doing a better job of simulating materials, where that's chemistry for life, chemistry for fertilizer, et cetera, et cetera, et cetera, et cetera. ⁓ And then there's the, it doesn't sound so great.

for societal good. I'm not sure it be on the UN list of things that they want to use a quantum computer for, but there are a lot of things in optimization that can save money ⁓ or make more money for a lot of industry. Where if they could optimize better, they'd have happy customers, higher return portfolios, lower risk, all those kinds of things. ⁓

⁓ So I think it's a combination of those kinds of things. ⁓ But yeah, I I think it's more in the line of just the more meta question of how can we continue to leverage computation to make humans happier and productive ⁓ with all the like ethical thinking about how to not use computers.

for the opposite.

Mark Fielding (51:23)
I don't know if you have still got five minutes. you got five minutes, Jeremy, if you got five minutes, Scott, do you have to jump? Jeremy, two things I'd like. I mean, if you have any other show and tell quantum show and tell some anything that you could hold physically in your hand to show us, that would be stunning because that always is well appreciated.

Scott Crowder (51:30)
Yep, I'm good.

They took the chips out of the closet, so Steve Crisale took them, so I don't know where they're at. He hasn't responded to me. So not today. I mean, we have the cryostat. We have actually some of these things over here. Do you me to the five, the block? Do you know about this? Is there any object in here that you want to show? Because we're also on the cusp of the 10 years of quantum and the cloud.

Mark Fielding (51:54)
Okay.

Scott Crowder (52:11)
Sure. So I don't know if there's a... Yeah, you're pretty smart. We're good. Got it.

Mark Fielding (52:16)
Excellent. It's very excited

because we can't be in New York. So it's like the next best.

Scott Crowder (52:20)
⁓ Yeah,

Yeah. I mean, think the challenge is like I can turn the camera, but I think Dave's worried about quality. So ⁓ yeah, you're editing this, right? So.

Mark Fielding (52:29)
Yeah.

Yeah,

yeah, this is all, this is all, it's not live, so.

Scott Crowder (52:40)
So Dave's going to kill me here. ⁓ So State-of-the-art quantum computer circa 2018. So this is system one. This is the first one that we did from a design point of view to show it all in one effective really, really pretty box to show that basically this is a computer. It's not a lab experiment. ⁓ In the early days, we've got like

archival pictures of you know, waveform generator equipment and like wires flying all over the place and it looks like a lab experiment. ⁓ Yeah, I mean like... You could just talk about scaling. Yeah, like this block which was like at the bottom of that quantum computer back in the day. ⁓

you know, has turned into the thing that I'm showing back here, which has like flex cables, you know, you know, hundreds of lines in and out of this thing in order to program it. ⁓ Yeah, I've got the flex cable. We multipurpose here at IBM. So this is actually the flex cable that we're talking about, like a prototype of it.

Mark Fielding (53:49)
and the heron.

Scott Crowder (53:58)
One of the things that we need to do is get more microwave signals in and out of the cryostat. And also for manufacturing reasons, you can't have people one by one plugging in wires like this anymore when you've got thousands or 10,000s of lines coming out of your computer. So we need to come up with ways that we can significantly increase the... ⁓

the amount of connections coming in and being able to just like click it into the computer from a manufacturing point of view as opposed to like hand connect it. Yeah, yeah, yeah, yeah, yeah, yeah.

Jeremy Gilbertson (54:31)
That's like a ribbon cable, right, that we were looking at with multiple connection. So what if,

so you mentioned Mobius before, what happens if you, if you had half turn full turn the ribbon?

Scott Crowder (54:41)
⁓ If it doesn't break, I think you're okay.

⁓ Yeah, I mean the trick on these things is that ⁓ it looks pretty standard, it needs to be superconducting and obviously it needs to not break as you're cooling it down to really low temperatures and heating it up occasionally. ⁓ So it has more demands than the typical flux cables. ⁓

Mark Fielding (54:53)
Yes.

Scott Crowder (55:12)
So it's like example of, kind of like an example I was giving of like, you know, where do I put the risk on 2029? You know, it's basically lots and lots of piece parts, very similar to that. It's not just one thing. It's like this entire engineering stack that we need to not only.

get the right quality on, but we need to get the right quality on at scale. And we need to do all of that together. And that's why it's taking us several years, even though we've demonstrated all the piece parts on our roadmap, we need to put it all together. Yeah, so this is kind of like a picture of what the 2029 system.

roughly is going to look like. So it looks kind of similar to what our system is today. But this is not a football field anymore. If you talk to people like five, 10 years ago in quantum, because of the overheads and error correction at the time, people were talking like, other vendors were talking about having to build this gigantic football field size quantum computer of like,

We don't need to do that anymore. That's why we are so confident in the 2029 date, because we believe we've come up with the architecture and all the piece parts to build something that is of reasonable scale for us to deliver.

Mark Fielding (56:35)
I've got one last question and then Dave might overrule this question as well. So if Dave thinks that you're not allowed to ask it, we won't ask it. But I don't think you can speak about quantum at the moment without speaking about Nvidia. They've just released some big news on their AI iPhone.

reading through this is the idea of this

architecture.

to remove the reliance on Nvidia or is it to build the relationship between IBM and Nvidia? What role do or will the Nvidia QP use when they play in this?

Scott Crowder (57:15)
Yeah,

Nvidia GPUs are a really important part of the larger computing thing. And GPUs will be a critical part of quantum-centric supercomputing. It is the right platform to run certain kinds of work.

those certain kinds of work are definitely part of building a fault-taught quantum computer and is part of building a quantum-centric supercomputer and is part of AI mega farms. And all of those are going to exist in 2029. This is not about competition, et cetera, et cetera, et cetera. The one thing I will say, though, is that we firmly believe

that in order to drive adoption, it needs to be based on truly open source software that isn't tied to any one back end, whether it be the GPU back end, the CPU back end, or the quantum back end. So that's why we would say that the layer

for QCSE from an orchestration point of view or from ⁓ how you program the quantum elements of it, needs to be open and needs to be, in our perspective, cloud native. Because a lot of the usage, accessibility to your point, making it accessible, means right now making it cloud accessible. It's just a lot easier to get board adoption if it's cloud accessible. From our perspective, it needs to be open.

Qiskit is, I still think this is a true statement, the only software development kit you can run on Google, Microsoft, and AWS. ⁓ Obviously, IBM also.

We fundamentally believe that that's the case both in how you connect down and how you connect up. Connecting into a video is absolutely part of what's going to have to be the case. We just want to make sure that quantum computing and how quantum computing links into all the other forms of computing is done in a way that people have flexibility in what hardware they choose for the different elements.

Mark Fielding (59:35)
choice is king. Thank you. Makes sense. Anything that we've missed, Jeremy?

Jeremy Gilbertson (59:43)
No, I think we're in good shape. This was a great run, Scott. Thanks for taking the time with us. Thanks for entertaining some of these crazy analogies that we have and letting us know what you guys are up to. Very exciting and stay in touch on all the great stuff you guys are doing.

Scott Crowder (59:57)
absolutely, anytime.

Mark Fielding (59:58)
Did we miss

anything really fundamentally important?

Scott Crowder (1:00:11)
Dave's typing. I'm thinking. I don't think so. Yeah, I mean, I think the question you hit on the accessibility and the algorithm development threads are really, really, really, really important. I think that's there's, ⁓ we obviously believe we're in the lead here.

think that we're ahead of everybody else. think we've right there. Technology choices, obviously we're biased, right? ⁓ And we have confidence we're gonna hit 2029. But there are enough other players with enough other money behind it that even if we're off.

Like, I don't think they're gonna be 10 years behind us. Like, so we do believe that there's enough money in this field that quantum computing and fault-on quantum computing is going to be a reality. It's just a question of who and when, not if anymore. And I think the big thing is going to be how much algorithmic discovery, how much application research, how much...

focus on how do you use this tool is really going to be the bigger unknown, in my personal opinion, which is why we're so passionate about getting the technology out there, getting people to use it, et cetera, cetera. ⁓ So there's been a big tick up, like I mentioned, in the last three years, but the curve is actually even increasing. But it's still a long way to go.

Like the number of people thinking about quantum algorithms is still small compared to the number of people thinking about classical algorithms.

Mark Fielding (1:01:59)
Scott Crowder, Vice President of IBM Quantum Adoption. Thank you for thinking on paper with us today. If you enjoyed the show, please subscribe where you're listening to it and share with one quantum curious friend, one quantum curious person in your life who needs to hear this. And until next week, stay disruptive, be curious.

Jeremy Gilbertson (1:02:20)
Keep thinking on paper.

Scott Crowder (1:02:23)
I have one more thing you want. Do you want Scott to explain the sound in the background? I don't know if you can hear this.

Mark Fielding (1:02:24)
Yep.

Have you got time, Jeremy?

Jeremy Gilbertson (1:02:31)
I've actually got, have to run, unfortunately guys, yes. ⁓ But ⁓ no, thank you. Thank you guys very much. Scott, do me a favor on your side. I'm gonna stop the recording now and.

Mark Fielding (1:02:32)
Yeah, okay.