Cloud Hosting

What an interesting panel, Quantum Advantage Where are We and What is Needed? While the panelists looked slightly weary theirs was, after all, one of the last panels at ISC 2024 the discussion was fascinating and the panelists knowledgable. No such panel would be complete without also asking when QA will be achieved. The broad unsurprising answer to that question is not especially soon.

The panel included: Thomas Lippert, head of Jlich Supercomputing Centre (JSC) and director at the Institute for Advanced Simulation; Laura Schulz, acting head of quantum computing and technologies, Leibniz Supercomputing Centre; Stefano Mensa, advanced computing and emerging technologies group leader STFC Hartree Centre; and Sabrina Maniscalco, CEO and co-founder, Algorithmiq Ltd. The moderator was Heike Riel, IBM Fellow, head of science & technology and lead of IBM Research Quantum Europe.

Missing from the panel was a pure-play quantum computer developer that might have added a different perspective. Maybe next year. Topics included quantum-HPC integration, the need for benchmarks (though when and how was not clear), the likely role for hybrid quantum-HPC applications in the NISQ world; familiar discussion around error mitigation and error correction, and more.

Of the many points made, perhaps the strongest was around the idea that Europe has mobilized to rapidly integrate quantum computers into its advanced HPC centers.

Schulz said, The reason that our work in the Munich Quantum Valley (MQV) is so important is because when we look at the European level. We have the EuroHPC Joint undertaking. We have the six quantum systems that are going to be placed in hosting centers that European wide, and we all [have] different modalities, and we all have to integrate. We have to think about this at the European level for how were going to bring these systems together. We do not want multiple schedulers. We do not want multiple solutions that could then clash with one another. We want to try to find unity, where it makes sense and be able to amplify the user experience and smooth the user experience European-wide for them.

The idea is to connect all of these EuroHPC JU systems and make them widely available to academia and industry. LRZ and JSC, for example, have already fielded or are about to field several quantum computers in their facilities (see slides below).

Lippert emphasized that, at least for this session, the focus is on how to achieve quantum advantage when we talk about quantum utility, when this becomes useful, then the quantum computer is able to solve problems of practical usage significantly faster than any classical computer [based on] CPUs, GPUs, of comparable size, weight and power in similar environments. We think this is the first step to be made with quantum-HPC hybrid type of simulation, optimization, machine learning algorithms. Now, how do you realize such quantum advantage? You build HPC-hybrid compute systems. We have the approach that we talk about the modular supercomputing architecture.

Our mission is to establish a vendor-agnostic comprehensive public quantum computer user infrastructure integrated in to our modular complex of supercomputers to . [Its] is a user friendly and peer reviewed access. So like we do with supercomputing.

Schulz drilled down in the software stack being developed at LRZ in collaboration with many partners. On the left side of the slide below are traditional parts co-scheduling, co-resource management, all those components that we need to think of, and that we do think of with things like disaggregated acceleration, said Schulz.

When you get to the right side, she noted, we have to deal with the new physics environment or the new quantum computing environment. So we have a quantum compiler that we are developing, we have a quantum representation moving between them. Weve got a robust, customized, comprehensive toolkit with things like the debuggers, the optimizers, all of those components thats built with our partners in the ecosystem. Then we have an interface, this QBMI (quantum back-end manager interface) and this is what connects the systems individually into our whole framework.

Now, this is really important. And this is part of the evolution. Weve been working on this for two years, actively building this up, and were already starting to see the fruits of our labor. In our quantum Integration center (QIC), we are already able to go from our HPC environment, so our HPC testbed that we have, using our Munich quantum software stack, we are able to go to an access node on HPC system, the same hardware, and call to the quantum system. We have that on prem, it is co located these systems, and it is an integrated effort with our own software stack. So we are making great strides, Schulz said.

The pan-European effort to integrate quantum computing into HPC centers is impressive and perhaps furthest along worldwide. Its emphasis is on handling multiple quantum modalities (superconducting, trapped ion, photonic, neutral atom) and approaches (gate-based and annealing) and trying develop relatively-speaking a common easy-to-use software stack connecting HPC and the quantum.

Mensa of the U.K.s STFC zeroed in on benchmarking. Currently there are many efforts but few widely agreed-upon benchmarks. Roughly, the quantum community talks about system benchmarks (low and middle level) that evaluate a systems basic attributes (fidelity, speed, connectivity, etc) and application-oriented benchmarks intended to look more at time-to-solution, quantum resources needed, and accuracy.

No one disputes the need for quantum benchmarks. Mensa argued for a coordinated effort and suggested the SPEC model as something to look at it. The SPEC Consortium for HPC is a great example, because its a nonprofit and it establishes and maintains and endorses standardized benchmarks. We need to seek something like that, he said

He took a light shot at the Top500 metric not being the best approach, noting it didnt represent practical workloads today, and added the You know that your car can go up to 260. But on a normal road, we never do that. Others noted the Top500, based on Linpack, does at least show you can actually get your system up and running correctly. Moreover, noted Lippert and Schulz, the truth is that the Top500 score is not on the criteria lists they use to evaluate advanced systems procurements.

Opinions on benchmarking varied, but it seems that the flurry of separate benchmark initiatives are likely to continue and remain disparate for now. One issue folks agree on is that quantum technology is moving so fast that its hard to keep up with, and maybe its too early to settle on just a few benchmarks. Moreover benchmarking hybrid quantum-HPC systems becomes even more confusing. All seem to favor use of a suite of benchmarks over a single metric. This is definitely a stay-tuned topic.

Turning to efforts to achieve practical uses, Maniscalco presented two use cases that demonstrate the ability to combine quantum and HPC resources by using classical computing to mitigate errors. Her company Algorithmic Ltd, is developing algorithms for use in bioscience. She provided a snapshot of a technique that Algorithmic has developed to use tensor processing in post-process on classical systems to mitigate errors on the quantum computer.

HPC and quantum computers are seen almost as antagonists in the sense that we can use, for example, tensor network methods to simulate quantum systems, and this is, of course, its very important for benchmarking, said Maniscalco. But what we are interested in is bringing these two together and the quantum-centric supercomputing idea brought forward by IBM is important for us and what we do is specifically focused on this interface between the quantum computer and the HPC.

We develop techniques that are able to measure or extract information from the quantum computers in a way that allows [you] to optimize the efficiency in terms of number of measurements, this eventually corresponds to shorter wall time overhead overall, and also allows to optimize the information that you extract from the quantum computer, and importantly, allows in post processing, she said. (best to read the associated papers for details)

At the end of Q&A, moderator Heike Riel asked the panel, Where will we be in five years? Here are their brief answers in the order given:

Read more here:
Europe's Race towards Quantum-HPC Integration and Quantum Advantage - HPCwire