Hot AI News

IBM and Nvidia want quantum chips scheduled like GPUs, and Fermilab joins in

IBM, Fermilab and Nvidia plan to pair DGX B200 systems with Nighthawk quantum computers, and an open-source Slurm layer would let supercomputer sites schedule QPUs like GPUs.

IBM and Nvidia want quantum chips scheduled like GPUs, and Fermilab joins in
Image: IBM

IBM said on Monday it plans to build a modular quantum computing co-design platform with Nvidia and Fermilab, placing Nvidia DGX B200 systems next to IBM quantum computers, including those running the Nighthawk processor. The work is described as plans, not a launched product.

A platform that pairs DGX B200 with Nighthawk

According to IBM, the platform would co-locate Nvidia DGX B200 systems with IBM quantum machines. Nighthawk is described as the most powerful processor IBM has built: it runs circuits of up to 7,500 two-qubit operations on 120 qubits, at up to 100,000 samples per second.

The IBM Quantum Nighthawk r2 processor
Image: IBM

Fermilab’s Superconducting Quantum Materials and Systems (SQMS) Center would supply the testbed. The aim is to explore what is needed for several quantum processors to work together on large numbers of samples, while classical compute scales alongside them. Fermilab and IBM also plan to study what it would take to run calculations in high-energy and nuclear physics that classical systems cannot handle alone, such as lattice gauge theories, neutrino interactions and nuclear simulations.

Anna Grassellino, Fermilab’s chief technology officer and SQMS director, called the effort “an important step toward the capabilities and community needed for a future DOE quantum computing user facility.” IBM’s Jerry Chow framed it as using real scientific problems to shape the modular, interconnected architectures of future machines.

IBM Quantum Nighthawk r2 processor held in a hand
Image: IBM
Advertisements

Slurm gets a native way to schedule quantum hardware

The more immediately practical piece is software. Slurm, the scheduler behind many of the largest supercomputers, had no built-in concept of a quantum processor, so sites attached them through plugins outside its core resource model. IBM, Nvidia and Pasqal worked with the open-source community on the Quantum Resource Management Interface (QRMI), an open-source, vendor-agnostic layer that plugs into Slurm’s Hierarchical Resources (HRES) scheduling.

A GPU lives inside one machine, but a QPU can serve several nodes at once. According to IBM, HRES lets administrators treat that pooled device as a first-class resource with its own caps per component of the cluster. Slurm can then hand out, book and bill QPU time under the same policies sites already apply to CPUs and GPUs. In IBM’s words, a hybrid job becomes an ordinary Slurm job, with no separate queue or login.

IBM and Nvidia want quantum chips scheduled like GPUs, and Fermilab joins in
Image: IBM

A chemistry result: 79 times better accuracy

IBM also reported new work with RIKEN and Nvidia on Fe4S4, an iron-sulfur cluster found in proteins whose interacting electrons make it hard to simulate. Using sample-based quantum diagonalization, the team started from a deliberately imperfect but more realistic guess of the molecule’s state. They ran circuits of up to 4,000 two-qubit operations on 72 qubits, collected 5 million samples, and used Nvidia GPUs, including RIKEN’s ROQUO supercomputer.

A 79-fold accuracy gain, within 4.3 millihartree of the classical DMRG method

IBM says that is a 79-fold accuracy improvement over earlier runs of the same experiment, and within 4.3 millihartree of the density-matrix renormalization group (DMRG), a leading classical method. Seiji Yunoki of RIKEN said it brings accuracy close to DMRG for this cluster. The work is backed by Japan’s NEDO.

Where this sits in IBM’s Nighthawk run

The announcement lands in a busy week for IBM’s quantum roadmap. On November 6 we reported that IBM put a Nighthawk-powered Quantum System Two in Chicago, where Fermilab was already named as a partner on high-energy physics and distributed quantum computing. Fermilab’s role here extends that thread.

IBM’s November 11 Nighthawk post explains why modularity matters. It says four l-couplers linked processors at 99.3% state transfer fidelity, against a 99.9% target, and it plans to link two next-generation r3 processors into a 1,040-qubit system in 2027. IBM also says partners have run 34 billion circuits on the current r2 chip. The co-design platform is where the interconnect and classical-side questions behind that 2027 system could be tested.

For background on why hardware scale alone is not enough, see our explainer on logical qubits and error correction.

Advertisements

What the announcement leaves unsettled

Nothing here is available to outside users yet. The release gives no launch date, no location for the co-designed system, no budget, and no access terms for researchers. The phrase “intend to launch” is the only timing language. Fermilab’s remark about a future DOE user facility suggests the aim is a community resource, but no such facility is announced.

The Slurm work is the part sites could adopt soonest, since QRMI is open source and vendor-agnostic, but the release does not say which Slurm versions include HRES support or which centers run it today. The Fe4S4 figures come from IBM’s summary of a paper, and the 79-fold gain compares against earlier runs of the same experiment, not against a classical method. DMRG still comes out ahead, so the result narrows a gap rather than closing it.

Sign Up for the Daily AI Pulse

One email a day. All the stories that matter.

Daniel ReedAI Research, Safety & Ethics Analyst

Daniel Reed currently works as an AI Research, Safety & Ethics Analyst at Aiholics, writing about how changes in artificial intelligence are affecting and will affect scholarship,…

Leave a Reply

Your email address will not be published. Required fields are marked *

You may also like