IBM has released a Quantum-Centric Centric Supercomputing (QCSC) reference architecture to extend quantum to HPC environments. IBM Spectrum LSF and IBM Spectrum Symphony integrate into the QCSC system‑orchestration layer by using the Quantum Resource Management Interface (QRMI) abstraction to present Quantum Processing Units (QPUs) as schedulable, monitorable accelerators; this lets existing LSF and Symphony workflows dispatch hybrid quantum–classical jobs with minimal code changes while preserving centralized scheduling, accounting, and telemetry in production HPC environments.
How LSF and Symphony maps into the QCSC reference architecture
The IBM QCSC reference architecture defines a four‑layer stack: applications → middleware → system orchestration → hardware infrastructure, with the system orchestration layer responsible for resource discovery, scheduling, and cross‑platform workflow coordination. QRMI sits in that orchestration boundary as a vendor‑agnostic API that exposes QPU capabilities, queueing models, and telemetry to higher‑level schedulers.
LSF and Symphony are mature workload managers that already schedule heterogeneous accelerators (e.g., GPUs & FPGAs) and can be extended to treat QPUs as first‑class resources. In practice, this means adding a QRMI connector or lightweight wrapper so that job descriptors can request a quantum accelerator (e.g., qpu=ibm:heron:1) alongside classical resources. The orchestration layer then coordinates pre‑ and post‑processing on CPU/GPU nodes and submits quantum payloads through QRMI.
End‑to‑end technical flow
A typical hybrid workflow begins with a classical stage that prepares data and compiles quantum circuits. The scheduler, LSF or Symphony, launches that stage on classical nodes, then uses QRMI to discover eligible QPUs, matching qubit count, topology, and queue length, and to submit quantum jobs. QRMI returns execution status and fidelity/telemetry metrics; the scheduler uses those signals to decide retries, fallbacks, or to resume dependent classical stages. This flow was demonstrated in an LSF Sample‑based Quantum Diagonalization (SQD) demo where the monolithic algorithm was split into four LSF jobs—map, optimize/transpile, execute on QPU, and post‑process—each orchestrated and monitored through LSF with QRMI‑style device queries.
Practical benefits for industrial and commercial users
By integrating via QRMI, organizations gain minimal application change (i.e., scheduler annotations rather than full rewrites), unified SLAs and chargeback across quantum and classical resources, and telemetry‑driven placement that adapts to device fidelity and queueing. This enables production workflows—such as chemistry simulations and optimization pipelines—to be run under familiar operational controls and audit trails while leveraging QPUs where they deliver a real measurable advantage.