Since our first integrations with supercomputing centers, Slurm has been the workload manager we encounter most often. IQM quantum computers run alongside classical systems at leading HPC centers worldwide, including Oak Ridge National Laboratory (ORNL), VTT, the Leibniz Supercomputing Centre (LRZ) and CINECA, across traditional HPC, cloud and AI infrastructure. At each one, Slurm is how researchers reach the QPU: they submit hybrid quantum-classical jobs through the same scheduler and queues they already use for classical work.
Every site is different, from its hardware and partitioning to its user communities and operating practices. This post shares how we schedule QPUs with Slurm today, where current options reach their limits, and why hierarchical resources in recent Slurm releases matter for quantum-HPC integration.
We start from a reference architecture and adapt it to the environment already in place, so operators keep their own way of working. Slurm offers several building blocks for this:
The QPU becomes one more resource in a scheduler operators already run. Users request it in their job scripts, and Slurm’s standard features apply to QPU jobs as configured locally, including accounting, fair-share, access control, multifactor prioritization and quality of service (QOS) settings.
Each model comes with a trade-off. GRES gives tight control but binds the QPU to specific nodes. Licenses are flexible but coarse: Slurm assumes a license can be used from every node, which does not reflect how a QPU is physically connected to part of a cluster. Custom definitions fit one site well and rarely carry over to the next, so each new integration involves rework.
Choosing the right combination for each customer’s workloads has been, and remains, central to our integration work. It is also why we follow Slurm’s development closely.
Slurm is moving in a direction that helps. NVIDIA acquired SchedMD, the company behind Slurm, in December 2025 and has committed to keeping Slurm open source and vendor neutral. We welcome NVIDIA’s investment in making quantum computers a schedulable resource in HPC, and we are glad to support it as a launch partner.
Hierarchical Resources (HRES), introduced as a beta feature in Slurm 25.05, let administrators define a resource next to their compute and associate it with groups of nodes arranged in layers. Jobs request the resource in the same way as a license, and Slurm checks availability at the relevant layer of the hierarchy.
This maps closely onto a QPU: a scarce, shared resource reachable from part of the cluster and owned by no single node. Compared with GRES and licenses, the standard approaches today, HRES brings two benefits. It allows finer granularity in how QPU capacity is managed, and it gives administrators control over which nodes can access the system. They can describe which nodes reach the QPU, at which level its capacity is counted, and how it is shared across partitions and workloads.
Slurm 26.11 takes this a step further. Its release candidate adds plugin support to HRES, so resource behavior can be written as Lua scripts and connected to external services [NVIDIA to confirm wording]. Device availability, for example, can feed directly into scheduling. When a QPU goes offline, new jobs that need it wait in the queue without holding classical nodes, and become eligible again once the device is back.
For quantum-HPC integration, this makes deployments reproducible. A resource model designed and validated at one site can be packaged and reused at another, with only site-specific parameters changed. Each new integration builds on the last, shortening the path from installation to users running hybrid jobs.
Our partner Munich Quantum Software Company (MQSC) shows what this looks like in practice. The MQSC Orchestrator combines HRES in the Slurm 26.11 release candidate with QDMI, the Quantum Device Management Interface, to schedule IQM systems such as IQM Emerald alongside other quantum providers. Read more in MQSC’s blog post and demo repository.
Our engineers are now exploring how HRES maps onto our current hybrid set-ups, and how it works alongside the mechanisms we already deploy. As centers upgrade their Slurm installations and the feature matures, we plan to evaluate it with them, validate it in staging environments, and bring it into our reference designs where it improves on today’s options. As always, we will work with what each facility runs.
“Slurm is how most HPC users reach our quantum computers, so we follow its development closely. Hierarchical resources give operators finer control over how a QPU is shared, and a scriptable resource model would let us carry a proven set-up from one installation to the next.” Álvaro Caride, Customer Solutions Engineer, HPC integration, IQM
Emilia Stuart is a content strategist and storyteller at IQM Quantum Computers, specializing in translating complex quantum computing concepts into engaging narratives. With a background in research and tech marketing, she understands potential customers and crafts stories that resonate. Emilia’s passion is making intricate technologies accessible to diverse audiences.
Search faster—hit Enter instead of clicking.