- NVIDIA DGX B200 systems will work alongside IBM quantum computers, including the 120-qubit Nighthawk
- Target workloads include neutrino interactions, nuclear simulation and lattice gauge theories
- A new Slurm interface lets HPC centers schedule quantum jobs alongside CPU and GPU work
IBM plans to work with NVIDIA and the Department of Energy’s Fermi National Accelerator Laboratory to build a co-design artificial intelligence-enabled platform for modular quantum computers. The platform pairs quantum processors with GPU-accelerated classical systems to study high-energy and nuclear physics problems, IBM announced Monday.
The company said the work covers application development and component design for future fault-tolerant systems, along with tighter links between quantum hardware and supercomputers.
What Will the IBM-NVIDIA-Fermilab Co-Design Platform Include?
Under the plan, NVIDIA DGX B200 systems will be co-located with IBM quantum computers, among them machines running IBM Quantum Nighthawk. IBM describes the 120-qubit processor as its most powerful to date, capable of executing circuits with as many as 7,500 two-qubit gates at a rate of up to 100,000 samples each second. Nighthawk also powers the IBM Quantum System Two unveiled in early November in Chicago and a quantum error correction co-design platform that the company also launched earlier with AMD and the Quantum Science Center at Oak Ridge National Laboratory.
The Fermilab effort will draw on the lab’s Superconducting Quantum Materials and Systems Center, or SQMS. It gives researchers a place to test how several quantum processors can run large sampling workloads in concert while classical computing scales with them. IBM joined SQMS as a partner institution in July 2024, with work spanning cryogenics, quantum interconnects and qubit noise reduction. Through SQMS, the company is also exploring how multiple quantum networking units could operate inside quantum data centers. That work is part of a broader effort with Cisco to demonstrate a network of fault-tolerant machines by 2030.
Which Physics Problems Will Fermilab & IBM Target?
The two organizations plan to examine what it would take to run physics calculations that exceed the capacity of conventional computers. Candidate workloads include neutrino interactions, nuclear simulations and lattice gauge theories.
Jerry Chow, an IBM fellow and chief technology officer for quantum-centric computing, said that scaling the two computing models together “will open new frontiers in high-energy and nuclear physics and beyond.” He added that the partners intend to let real scientific problems inform the interconnected architectures of next-generation quantum and accelerated classical machines.
How Do NVIDIA and Fermilab Describe Their Roles?
Krysta Svore, NVIDIA’s vice president of applied research for quantum computing, said quantum chips and accelerated computing must “evolve together as accelerated quantum supercomputers” for the technology to deliver on its promise. Fermilab has already used NVIDIA’s CUDA-Q Logical software to estimate the resources fault-tolerant computing would require.
Anna Grassellino, Fermilab’s chief technology officer and SQMS director, cited the lab’s ultracoherent quantum hardware expertise. She said the goal is to involve physicists in designing algorithms, applications and modular system layouts. She called the project “an important step toward the capabilities and community needed for a future DOE quantum computing user facility.”
How Will Slurm Integration Bring Quantum Into HPC Workflows?
Separately, IBM and NVIDIA worked with Pasqal and open-source developers on the Quantum Resource Management Interface, or QRMI. It plugs quantum processors into Slurm, the job scheduler behind many large supercomputers. Slurm previously lacked a built-in way to recognize a quantum processing unit, forcing sites to rely on add-on plugins.
QRMI works with Slurm’s Hierarchical Resources, or HRES, feature. Unlike a GPU, a QPU can serve several compute nodes at once, so HRES lets administrators treat it as a shared resource with per-component limits. Slurm can then reserve, allocate and track quantum time under the policies already applied to CPUs and GPUs. IBM said HPC centers can run hybrid jobs without a separate queue or login, and the vendor-neutral design supports processors from multiple suppliers.
What Did the IBM-RIKEN-NVIDIA Simulation Show?
In related research, IBM, Japan’s RIKEN and NVIDIA used sample-based quantum diagonalization, or SQD, to model Fe4S4. The iron-sulfur cluster is found in living cells, and its strongly interacting electrons are hard to simulate. The team started from a deliberately rough but more realistic estimate of the molecule’s state. It then ran circuits of as many as 4,000 two-qubit gates on 72 qubits and gathered 5 million samples, pairing quantum hardware with NVIDIA GPUs, including those in RIKEN’s ROQUO supercomputer.
The workflow produced a 79-fold accuracy improvement over the team’s previous runs. It landed within 4.3 millihartree of results from density-matrix renormalization group, a high-accuracy classical method. Seiji Yunoki, who leads RIKEN’s computational materials science research team, called the outcome “an important step toward applying quantum–HPC hybrid computing to scientific problems.” Japan’s New Energy and Industrial Technology Development Organization funded the research.
IBM said the scheduler work, application research and Fermilab platform together support its plan to embed quantum computers in existing classical environments while developing architectures for larger systems.


