Authors: LAURA SCHULZ and Prof. Dr. MARTIN SCHULZ
By its very nature, high-performance computing (HPC) resides at the forefront of technology, with a history of advances punctuating its evolution—from the introduction of multi-core systems in the mid-2000s, the use of GPUs in the last decade, to specialized artificial intelligence (AI) processors in recent years. Each of these waves crashes upon new boundaries of maximal performance that drive new, previously unattainable scientific discoveries. It is no surprise that the HPC community now looks to the emerging capabilities of quantum computing as the next wave of progress to continue pushing the field as a whole forward.
To fulfil this potential, though, quantum computers—as any other accelerator technology before—must be tightly weaved into the HPC realm. Due to its radically different approach to computing, quantum integration is a complex, multi-layered and interdisciplinary endeavour. For one, quantum computing systems are just now emerging from the experimental world of physics laboratories with a long road towards maturation ahead. Second, most types of quantum systems, in their current form, have notably different environmental needs and infrastructure requirements than established HPC systems. Third, quantum systems are heterogeneous themselves and require elaborate embedded control systems to achieve the quantum state necessary for computing–it is these controls systems that need to integrate with HPC. Finally, the software requires integration across the entire stack: from programming models to tools as well as integrated scheduling environments and operating system extensions.
As part of the Munich Quantum Valley (MQV), funded by the state of Bavaria as part of its Hightech Agenda, and its associated projects Q-Exa, DAQC, MUNIQC-SC, MUNIQC-ATOMS, and QuaST,the Leibniz Supercomputing Centre (LRZ) investigates solutions to these challenges. MQV-associated projets are funded by the German Federal Minstry for Education and Research (BMBF) and the German Federal Ministry for Economic Affairs and Climate Action (BMWK).
These projects lay the foundation for the efficient integration and use of quantum accelerators in HPC. As a nexus point for these efforts, the Quantum Integration Centre (QIC), established in 2021 at LRZ, houses a research-rich environment with multiple quantum systems and a hefty HPC testbed that directly connects to the quantum systems. This includes an inaugural cryostat, in collaboration with IQM deploying its superconducting technology, and will soon add other quantum technologies and systems. The HPC resource, a medium-sized cluster with additional non-quantum accelerator technologies, is currently being setup in close physical proximity, providing the opportunity to experiment with different network configurations and placement of the control systems. This will provide the deeply integrated, low-latency environment needed for many quantum applications as well as enable novel research in widely heterogeneous systems spanning multiple technologies. This testbed environment further offers a unique platform to merge the two currently widely different software environments.
In order to overcome these challenges, computer and domain scientists require new programming models and approaches that bridge the gap between the two disparate software stacks and enable simplified programming accessible to a broader audience without advanced physics degrees. These models must then be integrated into the existing world of HPC programming models and languages coupled with high-level abstractions to open the door for usage by the broader HPC user community. As a starting point, LRZ and its partners are currently devising extensions to OpenMP that would allow quantum offload through OpenMP’s “target” environment, which is a first step to realizing this vision. Additionally, the HPC and the QC system software must be adapted to operate the two system types together in a unified fashion.
Further yet, we require a new generation of schedulers that combine HPC and QC schedules and enables clean co-scheduling of resources and facilitates efficient sharing of QC resources from multiple processes within a large-scale job or even across jobs. In this area, LRZ heavily leverages the work and expertise from several HPC projects (in particular the EuroHPC/BMBF projects REGALE and DEEP-SEA) on dynamic and malleable schedulers with co-scheduling opportunities and extends this work to include quantum acceleration. This will benefit QC acceleration and help transform the HPC landscape as a whole towards more dynamic resource utilization and heterogeneous execution across any accelerator.
With all these efforts aligned in concept and execution, LRZ helps push forward a powerful next wave in HPC’s continued evolution, bringing the raw power of quantum mechanics and the extraordinary benefits of quantum-based acceleration to advanced scientific computing and the discoveries and breakthrough results lying in wait.