Harnessing the Power of HPC and AI: How Do They Work Together?

Harnessing the Power of HPC and AI: How Do They Work Together?

High-performance computing (HPC), artificial intelligence (AI), and quantum computing (QC) are each touted as technologies that can individually benefit commercial and scientific interests to significant extents. HPC is the most mature of these technologies and is already in deployment around the world, with HPC applications powering advancements in numerous industries and academic fields. AI is in numerous deployments in its own right but is still quite experimental and in need of continued research. Quantum computing is the most nascent of the three technologies, but could arguably become the most disruptive of the three by solving certain problems that are intractable for classical computing, even high-performance computing.

In addition to their individual applications, these three technologies are often proposed in pairs. If HPC and AI are so promising individually, for example, imagine the achievements that could be advanced by leveraging their strengths together. The same proposition has been forwarded for both HPC and QC as one pair, as well as AI and QC as another. But if these technologies can be combined synergistically into these pairs, imagine the potential benefits of using them as a trio. Not only may their consolidated strengths be amplified, but their weaknesses may be addressed and alleviated.

One advantage of discussing these technologies together is that AI and QC, in particular, raise concerns about ethical and responsible computing. Boosting them together, as well as with HPC infrastructure, understandably ought to boost those concerns further. Imagine an autonomous AI, for example, with the ability to break various encryption schemes. Any measures that are put in place to mitigate these risks, therefore, will be applicable to these technologies both individually and in pairs.

As an example of these technologies being proposed in pairs, we have previously published an article ourselves titled “Harnessing the Power of HPC and Quantum Computing.” This article addresses the integration of HPC and QC, but no mention was made of AI at that time. Roughly half a year later, AI has matured enough to be added to the conversation.

Understanding High-Performance Computing

Imagine a laptop computer with a single-core processor and a couple of gigabytes of memory. That’s far more computational power than was used for the Apollo moon landings, and yet it quickly becomes insufficient when trying to solve numerous problems of commercial, scientific, and even academic interest. Upgrading your laptop to one with a multi-core processor and more memory results in noticeably faster computation.

Now imagine an extreme version of this upgrade. Foregoing the physical constraints of a laptop, imagine taking a larger case and building it into a powerful computer with multiple multi-core processors and many more gigabytes of memory. But that’s not enough. Now interconnect this one powerful computer with other powerful computers, each with its powerful processors and large memories. The resultant high-performance computer is considerably more powerful than any laptop and focuses its concentrated computational power on solving humanity’s most complex problems at blazingly fast speeds.

Some of the most noteworthy characteristics of HPC include:

• High-performance components, each of which seek to maximize speeds and throughputs while minimizing latency – delays in communication – to optimize the entire cluster’s performance and not cause any bottlenecks that would slow the cluster down; examples of such components include networking equipment, data storages, memories, and even file systems.
• Parallel computing, which means breaking down large tasks into many smaller tasks and then simultaneously executing all of these tasks across many processors or even many servers; the extreme version of this is called massively parallel computing and can involve thousands or even millions of processors.
• Clusters, with which a centralized scheduler manages parallel computing across a network of servers, called nodes, each offering powerful GPUs or multi-core CPUs for the tasks to which they are assigned.
• Cloud computing, which is not guaranteed to be available for every HPC cluster, but lowers the cost of HPC and greatly increases the accessibility of HPC when it is made available; coincidentally, these benefits have also helped democratize QC, which is mostly accessed through the cloud, although on-premises options are gaining in popularity.

HPC is heavily in use for scientific research, financial services, engineering, and other business problems. It enables involved organizations to solve problems that might only otherwise be solvable with supercomputers. However, supercomputers are rare, expensive, and might not be optimized to solve the given problems. HPC, therefore, has more flexibility and can be adapted to a broader range of problems.

‍

{{Newsletter-signup}}

‍

The Convergence of HPC and AI

The convergence of HPC and AI with QC is a nascent field that promises to be even more revolutionary than these component fields, whether individually or in pairs. There is not one set configuration, however, but rather different ways these fields could be implemented to boost one another.

AI, for example, needs to process vast amounts of data. This currently benefits from HPC, but could potentially benefit from QC in the near future. If QC is used to exponentially compress data and perform computation on it, HPC will remain useful for classical pre-processing and post-processing. Together, the so-called time-to-insight could be significantly reduced.

QC already implements Machine Learning, or ML, up and down its stack. ML is a subfield of AI. Therefore, AI is already in place to optimize control system operations, quantum circuit compilation, and more. Latency in executing these tasks would slow down the overall process of executing a quantum circuit and performing quantum computation, so these tasks are boosted, where possible, by HPC.

HPC, as noted, involves workflows and the scheduling of potentially large numbers of classical computing resources. This is not just an optimization problem, but rather a series of optimization problems. Solving these optimization problems could potentially rely on either QC-boosted AI or AI-enabled QC.

These scenarios are not meant to be all-inclusive. If HPC, AI, and QC are thought of as three points arranged in a triangular pattern, each pair of points can be drawn with a bidirectional arrow between them. Each field can boost each of the other two fields, as well as be boosted by the other two fields.

Key Benefits of HPC Integration

Fully integrating HPC with QC is still in development, but progress has already been shown nonetheless. Some of the areas where significant benefits have been realized include:

• Accelerating the synthesis, transpilation, and compilation of quantum circuits
• Accelerating the execution of hybrid classical-quantum algorithms, which utilize classical mid-processing in between executions of quantum circuits
• Increasing the sizes of quantum computers, as measured in qubit counts, that can be simulated using classical resources
• Solving “toy problems,” or small proofs-of-concept, of quantum algorithms that are designed to solve classically intractable problems

The benefits of QC to HPC, as well as to AI, still need to be realized. Again, QC is the most nascent of the three fields. To decrease latency, multiple initiatives are underway to integrate QC into data centers, eliminating the cloud between quantum and classical resources. To learn more, be sure to read “What does it mean for quantum computers to be HPC ready?”

Implementing HPC-AI Integration

Adding AI to the HPC integration with QC would offer several additional benefits. A non-exhaustive list of these benefits includes:

• Analyzing whether optimal computation should be with HPC, QC, or neither
• Analyzing whether HPC should accelerate QC or QC should accelerate HPC
• Optimizing workflows that include classical and quantum computational resources
• Through frameworks and libraries, simplifying the design of complex quantum algorithms
• Integrating HPC, QC, and AI frameworks, none of which are inherently compatible
• Optimizing the synthesis, transpilation, and compilation of quantum circuits
• Reducing the number of iterations required to solve hybrid classical-quantum algorithms
• Characterizing quantum computer hardware to suppress errors before they can occur
• Characterizing the noise of quantum computers to mitigate measurement errors

For an overview of the benefits and challenges of integrating just HPC and AI, check out the article “HPC and AI: Better Together” by Run:AI. “Integration of High-performance Computing And Artificial Intelligence” is a similar article by FS, except that it includes illustrations.

The most exciting aspect of using HPC, AI, and QC synergistically is that AI and QC are both relatively young fields that have not yet achieved their full potential individually, let alone together. The challenges of integrating these three technologies are well understood, but benefits may as yet be discovered that have not been imagined yet. Using AI to write code for quantum circuits is a very recent phenomenon, for example, and is destined to improve over time. But a few short years ago, it wasn’t an option at all. As AI and QC mature, this article is destined to get longer.

‍