How to build a hybrid supercomputer: IBM details how to build classical-meets-quantum system

IBM has created a new reference architecture for hybrid supercomputing systems that bring together both classical and quantum elements.

The reference architecture, published earlier this month, shows how quantum processors could be integrated with classical computing elements, including CPU and GPU clusters, to tackle complex workloads such as those used in scientific research.

The reference architecture sets out a schema for hybrid systems’ hardware, comprised of three layers: a quantum layer; a layer of colocated programable CPUs and GPUs that can work as a testbed for quantum error correction; and a separate layer of cloud or collocated CPUs and GPUs that will deal with the classical workloads that support the QPUs’ (quantum processing units) execution, such as pre or post processing.

It also lays out details on how system orchestration in a supercomputer using both quantum and classical approaches could work. As well as borrowing from other heterogenous architectures, the system orchestration includes a specific quantum resource management interface to handle the quantum elements in a hybrid supercomputing system.

The reference architectures also describes an application middleware layer that includes parallel and distributed processing protocols such as MPI, OpenMP and SHMEM, alongside quantum-friendly middleware. It also recommends using quantum programming models including Qiskit or Tket.

IBM’s aims for hybrid supercomputer architecture

“The architecture is designed to be open and composable, relying on open software, standard interfaces, and modular system configurations so that quantum capabilities can plug into existing HPC workflows, schedulers, and facilities rather than requiring an entirely new computing stack,” IBM said.

IBM hopes that its reference architecture for using classical computing alongside quantum infrastructure will speed up the adoption of quantum computing as well as offer better accuracy in critical use cases.

The company says the blueprint is less a “prescriptive blueprint” and more a “forward looking path” on how hybrid systems could be used for scientific work, as quantum systems continue to become more popular and greater numbers of algorithms to take adventures of them are developed.

IBM has already been working with customers on systems for quantum-centric workloads. It recently published a research paper detailing a joint project with US healthcare provider Cleveland Clinic to create a quantum-centric workflow to predict the molecular electronic structure of a miniprotein made of 300 atoms.

According to the paper, the work suggests that a mix of classical and quantum computing can be used to simulate complex proteins ultimately running to hundreds or thousands of atoms.

Researchers from IBM and Japanese research science and technology organisation Riken also published a paper last year detailing the largest quantum simulation of iron-sulphur clusters using a hybrid supercomputing system.

Related articles

• Future of chip design: Four things we learned from the Synopsys CTO Prith Banerjee at MWC 2026
• Nu Quantum gets $60m funding for mission to scale quantum computers
• IBM and AMD join forces to create basis of quantum-centric supercomputing