AMD Releases Open Software Stack for Hybrid Quantum Supercomputing

Connecting quantum processors directly to high-performance supercomputing infrastructure, AMD released an open software stack designed to manage hybrid execution paths across both architectures.

The Tech TL;DR:

• The open-source architecture handles both staged workflow orchestration and co-scheduled execution modes to bridge classical HPC resources with quantum hardware.
• Workload management relies on Slurm for classical job handling alongside targeted integration paths for Flux, while a Quantum Resource Interface abstracts heterogeneous backends.
• Demonstrations of the integration milestone are scheduled at upcoming industry showcases, featuring infrastructure powered by AMD Instinct MI355X GPUs and 5th Gen AMD EPYC CPUs.

The reference implementation organizes execution paths to connect application workflows directly to quantum resources through explicitly testable interfaces. Designed to support staged and co-scheduled hybrid workflows, the framework gives developers multiple ways to manage compute-intensive tasks without locking into proprietary topologies.

In the staged model, a workflow engine owns the application definition, preprocessing, quantum, and postprocessing stages, plus the data movement between them. Rather than maintaining a single shared execution window, classical and quantum capacities are individually allocated and freed up at each phase.

Classical preprocessing runs on workload-managed resources like Slurm today, producing outputs that flow into quantum admission steps. Once the quantum resource accepts the stage, the execution runs with reservation context and returns results for classical postprocessing.

For tighter coupling, the open-source stack exercises a co-scheduled model where classical and quantum resources are reserved for the same execution window. A scheduler-specific adapter connects the hybrid request to capacity evaluation, allowing runtime steps to execute inside that reserved window while classical compute continues in parallel.

# Example conceptual submission via Slurm integration for hybrid resource allocation
srun --partition=quantum-hybrid 
     --nodes=1 
     --gres=gpu:mi355x:4 
     --quantum-reservation=active 
     python3 submit_hybrid_workflow.py

Workflow orchestration layers like Tierkreis, StreamFlow, and Prefect express workflow logic, sequence stages, and handle data movement. Slurm currently manages classical resources, while QFw extends control toward quantum hardware by linking requests to reservation contexts and evaluating capacity through admission modules.

Compilers including MQT Core, Qiskit, and PennyLane prepare quantum programs for execution while keeping interfaces above provider-internal topology details. A common Quantum Resource Interface shim sits above QDMI and QRMI, providing a shared boundary for heterogeneous resources.

Current implementations provide access to IBM, IQM, and Pasqal resources through both interfaces, while Amazon Braket connects through QDMI. Munich Quantum Valley is developing QDMI-based access for federated setups across the Leibniz Supercomputing Centre and Amazon Web Services.

The SC26 milestone brings together national laboratories, cloud providers, hardware vendors, and universities, including Oak Ridge National Laboratory, IBM, Quantinuum, Hewlett Packard Enterprise, and Xanadu. The multi-organizational effort validates interface boundaries and standard definitions on real systems.

A new co-design platform established by AMD, IBM, and the Quantum Science Center at Oak Ridge National Laboratory serves as a proving ground. To build out the classical side, AMD supplies Umbra—an HPC architecture driven by AMD Instinct MI355X GPUs alongside 5th Gen AMD EPYC CPUs—which operates in conjunction with IBM quantum hardware.

CST.

Always consult with certified IT and cybersecurity professionals before altering enterprise networks or handling sensitive data.