Quantum – Are We There (or Close) Yet? No, Says the Panel

By John Russell

November 19, 2022

For all of its politeness, a fascinating panel on the last day of SC22 – Quantum Computing: A Future for HPC Acceleration? – mostly served to illustrate the wariness of quantum computing felt by significant portions of the HPC community and the enthusiasm felt by those directly involved in quantum computing development.

About the only thing really agreed upon was quantum computing remains very young and that there has been a tremendous amount of hype surrounding it. Even the moderator was tough – Sven Karlsson of the Technical University of Denmark said, “We’re here to discuss quantum computing and discuss its application for acceleration for HPC. And you’ve probably also noticed the question mark at the end here, right, so we are going to have an honest discussion about this. I’m a Swede and Swedes are quite direct.”

Sven Karlsson, Technical University of Denmark

“One challenge is to go from quantum computers being, in essence, an appliance, perhaps even lab equipment, if I may be so rude again, and go into a system where we actually integrate the quantum computers into the systems and make them a part of the system that can be done with GPUs. We have two separate software stacks. But at least we have a very developed HPC software stack. And we have a less developed software stack for the quantum side. Now we need to see, should we integrate these together? Should we develop common high level abstractions, perhaps even programming models and programming languages to program these systems,” noted Karlsson.

“We have several different communities. We have the physics community, we have the math community, and we have the theory community. We have engineering, we have models of computation, we have algorithms, etc. How do we make these people talk together using the same language? So there are there are several grand and general challenges here. And I’m hoping that we can discuss some of them here in the panel,” said Karlsson.

Then, of course, there was the question of timing and whether quantum computing was still too nascent to tackle integration with HPC.

The conversation was wide-ranging. Here are excerpts from their introductory comments, in the order in which panelists presented (lightly edited, apologies for garbling). Understandably, panelists working in quantum research focused most on nearer term job-related operational and research strategies while those looking in from outside brought a more qualitative external perspective.

Laura Schulz, Leibniz Supercomputing Centre head of Quantum Computing and Technologies

Laura Schulz, Leibniz Supercomputing Centre

“We opened the Quantum Integration Center in March of 2021. Since then, we’ve been working to establish a system to seat systems. We have a superconducting qubit system that has just come in. [It’s] five qubits with our strategic partner IQM. It’s Friday, so the QPU might actually be getting placed in today and we’re starting coarse calibration. We’re also looking at different [qubit] modalities to bring in.

“We are a user facility [and] looking at how we’re bringing these two communities (HPC and quantum) together. We’ve already experienced interesting challenges in terms of vocabulary understanding, process, workflows, again, like other user communities that come and start to utilize more HPC resources, we have to go through a little bit of an acclamation period to figure out how their workflows, how their behaviors can be translated or morphed into an HPC environment where HPC can accommodate them.

“Is it the right time? You know, I need to say that it is absolutely the right time. Are we too late? I think no. But are we too early? I also think no. What we’ve already seen is, like in the early days of HPC, we see there’s kind of a stack approach, this monolithic approach reaching from the hardware to the software. I think it’s really time that we start asking some of the tough questions of the quantum community. We need to be able to understand how we can fold this in as they’re designing and developing. We need to be in those conversations to help them with things like the scheduling, with the compilers – we definitely see that they’re going in a particular direction and we need to be part of that conversation early on.”

Anne Matsuura, Intel director of Quantum Applications & Architecture, Intel Labs

Anne Matsuura, Intel

“Our quantum computing research really focuses on quantum practicality and scalability. We’re trying to bring quantum out of the physics lab and into a commercial reality. So, we’re using our state-of-the-art process capabilities to build qubits that are fully compatible with our high-value manufacturing. Part of this bringing [quantum computing] into commercial reality is also where do where do we place the quantum machine eventually. I think it’s an accelerator in an HPC center. We design our qubits in the lab right next door to the same fab where we have our latest process node development in Oregon. [This] leverages our manufacturing capabilities of transistor and silicon expertise, which we think is important to being able to realize the promise of quantum computing.

“We’re focusing on silicon quantum dot spin qubits that look like one electron transistors, so they’re perfect for our manufacturing. We’ve dedicated a 300-millimeter wafer line specifically for fabricating silicon, spin qubit chips, and we’re leveraging the rest of the quantum community collaborating with partners throughout the world, such as Technical University of Delft. So again, I see a quantum machine as an accelerator in the HPC center. I envision the quantum system as a coprocessor, an accelerator with tight coupling between the classical and the quantum machines.

“How we actually connect the HPC center and the quantum machine is a very important question today. I think one of the biggest challenges is balancing and scheduling at scale. So, as we think about these machines, as they scale up, there’s much research needed into the right balance of quantum computing and classical computing nodes. How do you efficiently interconnect them? … The current method for doing quantum error correction would require a really large amount of classical computation constantly feeding back error syndrome information to the quantum machine. What does that look like? And we need very short latency there. Things we need to consider: things like compilation, mapping and scheduling for large scale, quantum computing systems will be demanding a lot of classical resources for optimization for performance and will be increasingly resource hungry for systems as we’re running larger scale applications.”

Natalia Vassilieva, Cerebras Systems director of product,

Natalia Vassilieva, Cerebras Systems

“I’m Russian, I’m also very direct. When Laura invited me to the panel, I confessed that I’m a big skeptic of quantum computing. And yes, I’m a scientist. I’m curious what it can bring, but I don’t see any benefits in the near-time and will be asking hard questions and challenge everyone on the panel. Although I’m a skeptic, I believe there are certain applications where quantum computing will be probably helpful in the future. Also, I think that there will be applications which we don’t even think about right now, which will be solved by quantum computing in another fifteen years from now. So, things like understanding the ground states of very complex molecules, solving the problems that are related with the quantum dynamics, and trying to understand the physical properties of the universe. [Using] quantum machines for solving things like that is really plausible.

“For the other applications – and my background is machine learning – I know that there are some explorations of how quantum computing can be used for machine learning. But frankly, I believe that the most probable use of ML is in the other direction, so where machine learning might be used to help design quantum computers, and understanding what kind of problems quantum computing can be used, but I don’t see in the near future [that] quantum computing [will] replace digital computers.

“The next question was about how to find people? Do we need to train and to have special programs in universities to bring more people into this field? My statement is, yes, we need to train people. And I think right now, quantum computing is in still very early stage, and in order to even program simple things, you need to have a very good understanding of quantum mechanics. I certainly believe that in order to train the classical machine learning programs, you better know statistics and probability. But at the same time, there are many users in today’s machine learning, classical machine learning, that just use Pytorch and TensorFlow routines without properly understanding how back propagation works. What’s the math behind the things? Because we are in a state where those tools are usable, where you don’t necessarily [need to] understand what’s underneath. I think we’re very, very far away from that state in quantum computing.”

Torsten Hoefler, ETH Zürich, is professor of computer science

Torsten Hoefler, ETH Zürich

“I’m a very pragmatic person. I like to actually know before I’m investing money into something, what I’m going to do with it. So, let me get right to the point. I looked at quantum algorithms. And I’m glad to see David Keyes here. [He] gave a wonderful keynote talk [recently] where he stated that in the mathematical field, about every 10 years, there’s a major algorithmic breakthrough. And the last one is overdue so we as a community should start thinking about it. In quantum, it’s exactly the same thing. In quantum, about every 10 years, there’s a major algorithmic breakthrough. The unfortunate piece about this is that there is only a handful of algorithms that provide speed-up (over classical) and every other algorithm is basically composed out of those. There is a quantum algorithm zoo but that’s really about five basic algorithms. And let me go through them. So, we have amplitude amplification. We have phase estimation, we have quantum random walks, we have quantum Fourier transform, [and] we have Hamiltonian simulation. Whenever you hear that there’s some kind of physics simulation, that is Hamiltonian simulation, basically. Whenever you hear that there’s some kind of Grover’s speed up, that’s about 90% of the quantum algorithms. There are some others that I personally don’t care about, because those are things that are not HPC related. This is quantum teleportation and random number generation.

“I can get exponential speed up sometimes, and then you win. But actually, as I mentioned, 90 to 98% of the algorithms give you a quadratic speed up (only). So now let’s assume the two-week runtime – so I give these guys two weeks to battle it out. How many instructions can the oracle function contain such that the quantum computer outperforms the A100 (GPU). The unfortunate fact for Grover or quadratic speed-ups is that we can do 0.1 floating point operations in our oracle. Okay. Well for cubic speed up we can actually get something reasonable. That is the main message I wanted to make here. We have to be extremely careful when we plug in the constants. Quadratic speed up is great. But if we need a problem size that’s larger than the universe and it takes 10 years to solve, our quantum computer will decohere before and we will just not do anything. So that was of course exaggerated. By the way, everything I said is exaggerated. I’m supposed to be controversial and start a discussion. If that was way too fast for you, there will be an article appearing in the CACM (Communications of the ACM) journal very soon, with all these numbers and analyses, because I had only five minutes. And I think I’m over time.

“I wanted to put that challenge out there; we as a community need to focus on finding better algorithms. And we need to specifically focus on finding algorithms that have super quadratic speed-up because these quadratic speed algorithms are not going to be useful in my lifetime… Are we still too early [trying to integrate quantum and classical]? Too late? I don’t know. What are the right metrics to compare classical quantum computers? When is quantum computing most effective? Well, just by chance, my colleagues at Microsoft published a paper this week on assessing the requirements to scale practical quantum advantage. There is also the Azure quantum resource estimator that was just posted yesterday, or the day before yesterday. So, there are people working on seriously assessing these constants because it’s all about the constants. Theory is great. But if it doesn’t happen in your lifetime, it’s not so great.”

Travis Humble, director at the Department of Energy’s Quantum Science Center, Oak Ridge National Laboratory

Travis Humble, ORNL

“From my perspective, the current systems today are very much focused on discovery. Whether you’re a physicist or an engineer, or a computer scientist, or the application end users, the idea of having a new tool, which surpasses all previous limitations that we knew about is a remarkable thing to play with. What we’re doing at the moment is we’re playing with these toy quantum computers. We’re trying to learn how to use them to solve problems. I would say that at the moment, we’re down in the lower left (slide below), so the lower stages of evolution, if you will, where we have, effectively, single links to these experimental quantum physics systems. They’re basically wrapped in Python, or some other language stack, in which we can execute the most rudimentary controls that we can imagine. Even then, we’ve already been able to demonstrate that these devices are sufficiently capable of solving problems such as chemistry, materials, simulation. Now, they’re not surpassing any of the problems that have been solved on our modern supercomputers. They certainly don’t adapt to theoretical approximations that those systems use in their algorithms. But that’s an open area for them to develop.

“It was also noted earlier that many of the devices we have access to today are limited by their noise, their quality. But moreover, they’re limited by how they’re operated. There are quantum error correction methods for redundantly encoding information and protecting against errors. And there are fault tolerant protocols by which those methods ensure that those operations can scale up. But the current devices are incapable of doing this. So, there has to be a balance between the size of these systems and the amount of control that goes into their performance in order to make this a worthwhile return on investment for system design.

“We’re learning now how to play with our toys; we’ll develop over time how to turn them into more sophisticated cultures, including the culture of high-performance computing. But ultimately, I believe that this is a new technology that will help us surpass what the field is capable of doing today. Now, I should say, I’m not so optimistic to think that quantum computing is a replacement for conventional computing. This is but yet another technology. And we’ll need to learn how to integrate that as part of the system that we develop, just as we integrate every other technology that comes along. So, there are going to be [the] difficult, difficult task of understanding what is the balance between different accelerators? How do we manage heterogeneity when quantum is a piece of that? How do you program in these different computational models, not just the data models and the language models, but even when the computation itself is distinct? And then ultimately, what was the application that made all that worthwhile?”

It would have been nice to have at least one quantum computer company in the mix; one with a working system (IBM, IonQ, Rigetti, D-Wave, etc.) that’s being used by early users. It would have been interesting, for example, to hear about at least one proto-application. Many of those applications, perhaps, would not be pure-play HPC or even blended HPC-AI, but they might have shed light on real-world uses.

The participating panelists were solid and frank, but at this chicken-or-egg stage for quantum computing, their perspectives tended to mostly reflect their existing work focus.

In one sense, Cerebras is trying to do on a lesser scale what the quantum computing crowd hopes to do in a broader way. Its wafer-scale engine is changing ideas, at least in the very high-end community, but that success hasn’t yet translated into broader market traction. Cerebras systems were used in an impressive collaboration that won this year’s Gordon Bell Special Prize (see HPCwire coverage, Gordon Bell Special Prize Goes to LLM-Based Covid Variant Prediction).

Vassilieva drew a comparison with the nascent quantum landscape. “It’s very challenging to break the status quo and convince people to use our architecture instead of very familiar Nvidia GPUs [because] their ecosystem is very well understood. The only reason for people to try out our new big shiny chip is to show that we are orders of magnitude faster than existing computing, and that we are orders of magnitude faster for the problem that they’re trying to solve right now. So, if you’re coming and saying, we can train very large language models really, really fast – and we can do that – the sparsity people who are not interested in that, who don’t understand why do they need learning models say, Okay, fine. Go do that.”

“I think the same is applicable to quantum computing,” she said. “[F]or that field to be very well recognized, you need to have people who believe and who have problems which cannot be solved today with classical methods. I think right now, with quantum computing, we are in a state where an achievement is when you have a toy problem programmed with a quantum computer, and achievement is that the quantum computer solves that problem. When the quantum computer is the right tool to make the discovery, then we will see mass adoption for quantum computing. Near term benefits near term? Sorry, I believe there is none. But that doesn’t mean that we should stop working on that,” she said.

Hoefler wondered if the flow of funds in the young quantum landscape wouldn’t better used in traditional HPC.

“If I give you a quantum computer today that has two properties I mentioned, that is a very powerful computer, I wouldn’t know what to do with it,” said Hoefler. “It may be that we’re diverting a whole lot of our attention to very early on technology, which is a physics experiment, and we start integrating the physics experiment into our HPC ecosystem. And we start building machines with the physics experiment. Then at the end, we realized that the experiment didn’t work. So, we have built all this infrastructure around and we spend a whole lot of our attention that we could have spent on driving the more traditional aspects of high performance computing forward, new architectures, the next generation CPUs, next generation physical integration of HBM the next generation memory systems chip systems chiplets.”

Workforce scarcity is a common clarion call in HPC and it is also a problem in quantum computing. Schulz repeated several times that the Quantum Integration Center at LRZ is actively looking for people and she is hiring.

Everyone agreed the hype around quantum computing is likely creating unrealistic expectations. Then again, hype around any new technology isn’t a new thing. Hoefler suggested a Quantum Bubble could burst.

Quantum computing stirs people in many ways. Hate the hype. Love the idea. It will be transformative. It will never Launch. It’s just getting started; temper your criticism. What’s it supposed to do again? Where’s all the money going? Shouldn’t we be investing more in traditional HPC? It’s policy infatuation, not practical science!

All of the sentiments listed above appeared, at least briefly, the discussion today though without rancor.

Humble offered the following:

“There is a story concerning Michael Faraday. Many of you may know him as a scientist from the 19th century, famous for his work in currents and electricity. After once demonstrating the phenomenon of electricity, someone asked him what is it good for. Faraday’s response, or so the story goes, is “of what uses a child?” Faraday was not trying to provide proof or evidence that this new phenomenon was in fact, ready to mature to evolve into the force that it is today. But he recognized the potential opportunity that was there. My opinion is very similar when it comes to quantum computing. It is at a very early stage of development, we have recently discovered it as a phenomenon. But we have not yet evolved it to the point where it is, in fact, a useful tool. But at least for me, personally, I can see a future where that may happen.”

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industry updates delivered to you every week!

Slide Shows Samsung May be Developing a RISC-V CPU for In-memory AI Chip

June 19, 2024

Samsung may have unintentionally revealed its intent to develop a RISC-V CPU, which a presentation slide showed may be used in an AI chip. The company plans to release an AI accelerator with heavy in-memory processing Read more…

ASC24 Student Cluster Competition: Who Won and Why?

June 18, 2024

As is our tradition, we’re going to take a detailed look back at the recently concluded the ASC24 Student Cluster Competition (Asia Supercomputer Community) to see not only who won the various awards, but to figure out Read more…

Qubits 2024: D-Wave’s Steady March to Quantum Success

June 18, 2024

In his opening keynote at D-Wave’s annual Qubits 2024 user meeting, being held in Boston, yesterday and today, CEO Alan Baratz again made the compelling pitch that D-Wave’s brand of analog quantum computing (quantum Read more…

Apple Using Google Cloud Infrastructure to Train and Serve AI

June 18, 2024

Apple has built a new AI infrastructure to deliver AI features introduced in its devices and is utilizing resources available in Google's cloud infrastructure.  Apple's new AI backend includes: A homegrown foun Read more…

Argonne’s Rick Stevens on Energy, AI, and a New Kind of Science

June 17, 2024

The world is currently experiencing two of the largest societal upheavals since the beginning of the Industrial Revolution. One is the rapid improvement and implementation of artificial intelligence (AI) tools, while the Read more…

Under The Wire: Nearly HPC News (June 13, 2024)

June 13, 2024

As managing editor of the major global HPC news source, the term "news fire hose" is often mentioned. The analogy is quite correct. In any given week, there are many interesting stories, and only a few ever become headli Read more…

Slide Shows Samsung May be Developing a RISC-V CPU for In-memory AI Chip

June 19, 2024

Samsung may have unintentionally revealed its intent to develop a RISC-V CPU, which a presentation slide showed may be used in an AI chip. The company plans Read more…

Qubits 2024: D-Wave’s Steady March to Quantum Success

June 18, 2024

In his opening keynote at D-Wave’s annual Qubits 2024 user meeting, being held in Boston, yesterday and today, CEO Alan Baratz again made the compelling pitch Read more…

Shutterstock_666139696

Argonne’s Rick Stevens on Energy, AI, and a New Kind of Science

June 17, 2024

The world is currently experiencing two of the largest societal upheavals since the beginning of the Industrial Revolution. One is the rapid improvement and imp Read more…

Under The Wire: Nearly HPC News (June 13, 2024)

June 13, 2024

As managing editor of the major global HPC news source, the term "news fire hose" is often mentioned. The analogy is quite correct. In any given week, there are Read more…

Labs Keep Supercomputers Alive for Ten Years as Vendors Pull Support Early

June 12, 2024

Laboratories are running supercomputers for much longer, beyond the typical lifespan, as vendors prematurely deprecate the hardware and stop providing support. Read more…

MLPerf Training 4.0 – Nvidia Still King; Power and LLM Fine Tuning Added

June 12, 2024

There are really two stories packaged in the most recent MLPerf  Training 4.0 results, released today. The first, of course, is the results. Nvidia (currently Read more…

Highlights from GlobusWorld 2024: The Conference for Reimagining Research IT

June 11, 2024

The Globus user conference, now in its 22nd year, brought together over 180 researchers, system administrators, developers, and IT leaders from 55 top research Read more…

Nvidia Shipped 3.76 Million Data-center GPUs in 2023, According to Study

June 10, 2024

Nvidia had an explosive 2023 in data-center GPU shipments, which totaled roughly 3.76 million units, according to a study conducted by semiconductor analyst fir Read more…

Atos Outlines Plans to Get Acquired, and a Path Forward

May 21, 2024

Atos – via its subsidiary Eviden – is the second major supercomputer maker outside of HPE, while others have largely dropped out. The lack of integrators and Atos' financial turmoil have the HPC market worried. If Atos goes under, HPE will be the only major option for building large-scale systems. Read more…

Comparing NVIDIA A100 and NVIDIA L40S: Which GPU is Ideal for AI and Graphics-Intensive Workloads?

October 30, 2023

With long lead times for the NVIDIA H100 and A100 GPUs, many organizations are looking at the new NVIDIA L40S GPU, which it’s a new GPU optimized for AI and g Read more…

Nvidia H100: Are 550,000 GPUs Enough for This Year?

August 17, 2023

The GPU Squeeze continues to place a premium on Nvidia H100 GPUs. In a recent Financial Times article, Nvidia reports that it expects to ship 550,000 of its lat Read more…

Everyone Except Nvidia Forms Ultra Accelerator Link (UALink) Consortium

May 30, 2024

Consider the GPU. An island of SIMD greatness that makes light work of matrix math. Originally designed to rapidly paint dots on a computer monitor, it was then Read more…

Choosing the Right GPU for LLM Inference and Training

December 11, 2023

Accelerating the training and inference processes of deep learning models is crucial for unleashing their true potential and NVIDIA GPUs have emerged as a game- Read more…

Nvidia’s New Blackwell GPU Can Train AI Models with Trillions of Parameters

March 18, 2024

Nvidia's latest and fastest GPU, codenamed Blackwell, is here and will underpin the company's AI plans this year. The chip offers performance improvements from Read more…

Synopsys Eats Ansys: Does HPC Get Indigestion?

February 8, 2024

Recently, it was announced that Synopsys is buying HPC tool developer Ansys. Started in Pittsburgh, Pa., in 1970 as Swanson Analysis Systems, Inc. (SASI) by John Swanson (and eventually renamed), Ansys serves the CAE (Computer Aided Engineering)/multiphysics engineering simulation market. Read more…

Some Reasons Why Aurora Didn’t Take First Place in the Top500 List

May 15, 2024

The makers of the Aurora supercomputer, which is housed at the Argonne National Laboratory, gave some reasons why the system didn't make the top spot on the Top Read more…

Leading Solution Providers

Contributors

AMD MI3000A

How AMD May Get Across the CUDA Moat

October 5, 2023

When discussing GenAI, the term "GPU" almost always enters the conversation and the topic often moves toward performance and access. Interestingly, the word "GPU" is assumed to mean "Nvidia" products. (As an aside, the popular Nvidia hardware used in GenAI are not technically... Read more…

The NASA Black Hole Plunge

May 7, 2024

We have all thought about it. No one has done it, but now, thanks to HPC, we see what it looks like. Hold on to your feet because NASA has released videos of wh Read more…

Google Announces Sixth-generation AI Chip, a TPU Called Trillium

May 17, 2024

On Tuesday May 14th, Google announced its sixth-generation TPU (tensor processing unit) called Trillium.  The chip, essentially a TPU v6, is the company's l Read more…

Intel’s Next-gen Falcon Shores Coming Out in Late 2025 

April 30, 2024

It's a long wait for customers hanging on for Intel's next-generation GPU, Falcon Shores, which will be released in late 2025.  "Then we have a rich, a very Read more…

GenAI Having Major Impact on Data Culture, Survey Says

February 21, 2024

While 2023 was the year of GenAI, the adoption rates for GenAI did not match expectations. Most organizations are continuing to invest in GenAI but are yet to Read more…

Nvidia Shipped 3.76 Million Data-center GPUs in 2023, According to Study

June 10, 2024

Nvidia had an explosive 2023 in data-center GPU shipments, which totaled roughly 3.76 million units, according to a study conducted by semiconductor analyst fir Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing po Read more…

A Big Memory Nvidia GH200 Next to Your Desk: Closer Than You Think

February 22, 2024

Students of the microprocessor may recall that the original 8086/8088 processors did not have floating point units. The motherboard often had an extra socket fo Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire