IBM's Quantum Co-Design Push with Fermilab and NVIDIA Puts QPUs on the HPC Scheduler
- Harold Fritts
- 947 views
IBM, Fermilab and NVIDIA are building a modular quantum co-design platform, while Slurm gains QPU scheduling and IBM reports a 79x accuracy gain on Fe4S4.

IBM, Fermilab and NVIDIA plan to build a modular quantum co-design platform that places NVIDIA DGX B200 systems next to IBM quantum computers, including systems running the IBM Quantum Nighthawk processor. This follows the error correction platform IBM set up with Oak Ridge National Laboratory and AMD. The new effort targets the modular architectures IBM expects fault-tolerant machines to use. Alongside it, IBM described a Slurm integration for scheduling quantum processors and a chemistry result with RIKEN that it says is 79 times more accurate than its earlier attempt.

Why the Co-Design Angle Matters
Most quantum announcements focus on qubit counts. This one focuses on the classical plumbing around the qubits, and that's the more useful signal for anyone running HPC sites. Fault-tolerant machines won't be single boxes. They'll be several processors wired together, fed by large GPU clusters for decoding and control. Building the testbed now lets the software and the scheduling questions get worked out before the hardware is at scale.
The platform's Fermilab piece is where the physics comes in. The work draws on the lab's Superconducting Quantum Materials and Systems (SQMS) Center, which contributes ultracoherent superconducting cavities, prospective interconnects, quantum memories and bosonic resources. Anna Grassellino, Fermilab chief technology officer and director of the SQMS Center, pointed to "SQMS expertise in ultracoherent quantum hardware" as the lab's contribution.

IBM says researchers will use the platform to test what it takes for multiple quantum processors to work together, run large numbers of samples and scale classical compute alongside them. The program covers new scientific applications, the integration of supercomputing and quantum resources, and components for future quantum computers. Fermilab and IBM also plan to study calculations in high-energy and nuclear physics, including lattice gauge theories, neutrino interactions and nuclear simulations. IBM hasn't disclosed a site, system counts or a launch date, so treat this as a research direction rather than a deployable product for now.
What Nighthawk Brings to the Table
IBM calls Nighthawk the most powerful processor it has built. According to IBM, it runs circuits of up to 7,500 two-qubit operations on 120 qubits at up to 100,000 samples per second. The current Nighthawk r2 revision is put at 25 times the throughput of IBM's Heron processor. IBM's roadmap targets 10,000 operations on 520 qubits with Nighthawk r3, but the company didn't say which revision the co-design platform will use. That's a detail worth pressing on in any evaluation conversation.
Throughput numbers like these matter more to HPC teams than raw qubit counts, because sample rate determines how much work a shared QPU can absorb in a scheduling window. Here's the thing: a fast processor with no scheduler integration is an expensive instrument sitting behind a queue of ad hoc scripts. That's why the Slurm work deserves as much attention as the hardware.
Scheduling Shared Quantum Processors in Slurm
IBM and NVIDIA, working with Pasqal and the open-source community, have brought quantum computers into supercomputing workflows through the Quantum Resource Management Interface (QRMI). QRMI is an open-source, vendor-agnostic layer, and it now ties directly into Slurm's Hierarchical Resources (HRES) scheduling. Until now, Slurm had no native concept of a quantum processor, so sites attached QPUs through plugins of varying maturity.
In practice, a QPU can be shared across multiple nodes. HRES lets administrators manage it as a shared resource and set limits for each part of the system. Slurm can then allocate, reserve and account for QPU time using the scheduling policies HPC sites already run, and a researcher can request a QPU from inside a job script. That last point is the one that changes day-to-day workflows: quantum time becomes one more resource in the same accounting model as GPUs and CPUs.
IBM didn't name a Slurm release. NVIDIA, which acquired Slurm developer SchedMD, says the quantum interface arrives in Slurm 26.11 with support for QRMI and QDMI. If you're planning around that, check the release notes when they land rather than assuming the feature set from the announcement alone.
Iron-Sulfur Chemistry: The 79x Result
IBM, RIKEN and NVIDIA researchers applied sample-based quantum diagonalization (SQD) to Fe4S4, an iron-sulfur cluster whose strongly interacting electrons make it hard to simulate classically. The team started from an intentionally imperfect but more realistic estimate of the molecule's state. That estimate steers the quantum computer toward the electron configurations that matter, rather than sampling blindly. The classical side ran on NVIDIA GPUs, including RIKEN's ROQUO supercomputer, which IBM says is built on NVIDIA GB200 NVL4 systems.
The runs used circuits of up to 4,000 two-qubit operations on 72 qubits and collected 5 million samples. IBM reports a 79-fold improvement in accuracy over the team's earlier runs of the same experiment, landing within 4.3 millihartree of the result from the density-matrix renormalization group (DMRG) method, a classical technique used here as the reference. These are vendor-reported results, and we haven't run or audited them. IBM didn't name the quantum processor used, which limits how far you can generalize the numbers.
The Fe4S4 research is supported by Japan's New Energy and Industrial Technology Development Organization (NEDO) through its project on quantum-supercomputer hybrid platforms. Separately, a protein-ligand project from IBM, RIKEN and Cleveland Clinic is a finalist for the 2026 ACM Gordon Bell Prize. NVIDIA highlighted that work at the show, and it also used ROQUO. It's a different experiment, so don't conflate the two sets of results.
Decoding and HPC Results at SC26
IBM is also showing Pelagos at SC26. It's an open-source decoder project from IBM and AMD, built on the Triton framework and running on AMD and NVIDIA GPUs. IBM says its decoder tests ran up to 97 times faster than a single CPU core. On GPUs, decoding scaled 7.5 times across eight AMD Instinct accelerators compared with one, measured on ORNL's Lux system. Those are IBM's figures from its own test conditions, so compare them against your own workloads before drawing conclusions.
Finally, the University of Illinois Urbana-Champaign, NCSA and IBM connected NCSA's DeltaAI supercomputer to an IBM quantum computer to model Cu3C6O6, a kagome-lattice metal. That's a small experiment on its own, but it shows the hybrid pattern the rest of this week's announcements are building toward: classical GPUs and a quantum processor sharing one workflow, with the scheduler deciding who gets what and when.