Quantum computing is emerging as a promising complement to high-performance computing (HPC). Rather than replacing classical supercomputers, quantum processors can act as specialised accelerators for selected parts of a larger application, while CPUs and GPUs continue to perform the remaining computations. This combination creates a new kind of hybrid workload, in which classical and quantum resources cooperate to solve a problem. But integrating the two worlds is not straightforward.
Traditional HPC scheduling generally allocates resources to a job for a defined period of time. Once a job starts, its nodes remain assigned to it until the job terminates. This model works well for many classical HPC applications, but hybrid Quantum-HPC applications have a different execution pattern. In addition, quantum computers are still scarce resources, while classical HPC systems provide thousands of compute nodes. A hybrid workload may therefore require a large amount of classical computation and only occasionally access a quantum processor.
The challenge becomes particularly evident when classical resources are allocated for the entire duration of the application. While the application is waiting for its quantum computation, the HPC nodes it has reserved may have little or no useful work to perform. For a single application this may seem like a minor inefficiency. At the scale of a shared HPC facility, however, these unused resources can lead to substantial resource waste and significantly reduce overall system utilisation. This raises a simple but important question: How can we efficiently share a quantum computer among hybrid workloads while minimizing the time classical HPC resources remain idle?
StreamFlow application
A more efficient model is to allocate resources according to the current stage of the application. This is precisely the kind of problem that hybrid workflow systems are designed to address. StreamFlow approaches the problem by representing the hybrid application as a hybrid workflow, allowing each step to be associated with the infrastructure and resources required for its execution. When a classical step is ready, StreamFlow requests the required HPC resources. Once it completes, those resources can be released. When the workflow reaches a quantum step, the corresponding quantum resource is requested through a proper Connector. After the quantum computation finishes, the workflow continues with the next classical stage.
This is a natural extension of StreamFlow’s approach to hybrid workflow orchestration. StreamFlow is designed to bind individual workflow steps to their most suitable execution environment, allowing heterogeneous applications to run across different infrastructures without requiring the entire application to be deployed on a single system. In the Quantum-HPC case, the same principle can be applied to the boundary between classical and quantum computation: resources are allocated when they are needed, rather than reserved for the entire lifetime of the application.
Experimental results
The approach was evaluated using a hybrid clustering application combining k-means, DBSCAN, and hierarchical clustering. The three algorithms run in parallel on the classical side, and their results are combined through a Maximum Independent Set (MIS) formulation expressed as a Quadratic Unconstrained Binary Optimization (QUBO) problem. The resulting solution is then evaluated using the Silhouette score. The experiments used a dataset of 80,000 two-dimensional points, creating a workload with clearly distinct classical and quantum phases.
The workflow was first tested on Qluster, a dedicated Slurm-based testbed hosted by the E4 Computer Engineering company, and then on Leonardo, the EuroHPC Tier-0 supercomputer hosted by CINECA. StreamFlow orchestrated the application by allocating resources according to the individual workflow stages. For the Leonardo experiments, the quantum phase was emulated, allowing the study to focus on resource management rather than on the performance of a specific QPU.
The results show a substantial reduction in classical resource consumption, which becomes more pronounced as the quantum phase becomes longer. Across the Leonardo experiments, StreamFlow achieved 1.94× greater resource efficiency for short quantum jobs (less than one second) and 2.78× greater resource efficiency for two-minute quantum workloads, compared with the static baseline. The corresponding wall times remained comparable within the variability observed on the production system The result illustrates the central advantage of workflow-based execution: classical resources do not have to remain reserved while the application is performing work elsewhere.
Towards the Quantum-HPC continuum
Quantum computers are unlikely to operate in isolation. As they become integrated into HPC facilities, scientific workflows will increasingly combine classical and quantum resources in complex execution patterns, requiring a control plane capable of coordinating resources with fundamentally different characteristics. The hybrid workflow paradigm provides a natural abstraction for this heterogeneity: each computational step can describe what it needs and where it should run, while the workflow manager coordinates execution and data dependencies.
This is already the principle behind StreamFlow’s hybrid Cloud-HPC approach: bind each workflow step to the architecture best suited to it, rather than forcing the entire application onto a single infrastructure. Quantum processors can become another specialised resource within the same computational continuum, towards a future in which HPC, Cloud, and Quantum resources can participate in the same workflow, each being used where and when they provide the greatest benefit. This perspective can make the difference between simply connecting a quantum computer to an HPC system and actually sharing it efficiently.
M. Cipollini, S. Rizzo, S. Iserte, P. Viviani, G. Vitali, M. Barbieri, G. Bettonte, E. Boella, F. Ganz, R. Rocco, O. Spina, A. J. Peña, P. Sandås, I. Colonnelli, A. Scionti, C. Vercellino, E. Dri, J. Frassineti, S. Marzella, A. Muratori, D. Ottaviani, O. Terzo, B. Montrucchio, and D. Gregori, “Three ways to share a QPU: Scheduling strategies for hybrid Quantum-HPC applications,” in Future Generation Computer Systems, vol. 185, p. 108699, 2026. doi:10.1016/J.FUTURE.2026.108699