THE ESSENTIALS IN 30 SECONDS

  • NVIDIA DGX B200 systems would be co-located with IBM quantum computers, including Nighthawk-based machines, at Fermilab's SQMS Center; Fermilab describes it as a proposed platform with no start date or cost given.
  • NVIDIA says its new TCBP-JOBS decoder in CUDA-Q QEC 0.9.0 runs up to 13,000 times faster than a leading decoder in one benchmark comparison.
  • A vendor-neutral interface, QRMI, is meant to let supercomputing centres schedule quantum processors as ordinary Slurm jobs, with Slurm 26.11 targeted for late November.

NVIDIA said on November 16, at the SC26 supercomputing conference in Chicago, that it will work with IBM and the U.S. Department of Energy's Fermi National Accelerator Laboratory on a platform for designing quantum and classical computers together. The announcement came with two software pieces aimed at research centres: a faster error-correction decoder and quantum support in the Slurm job scheduler.

What is planned

The idea is to install NVIDIA DGX B200 GPU systems next to IBM quantum computers, including machines built on IBM's Nighthawk processor, at Fermilab's Superconducting Quantum Materials and Systems (SQMS) Center. Unite.AI reports that Nighthawk handles circuits of up to 7,500 two-qubit operations on 120 qubits. Fermilab would lead the error-correction and physics work, coordinate access through its HEPCloud infrastructure and focus on high-energy and nuclear physics problems such as lattice gauge theories and neutrino interactions. A later phase would test Fermilab's superconducting cavities as quantum memories.

The project is still at an early stage. Fermilab's chief technology officer, Anna Grassellino, calls it a proposed platform, and The Next Web notes that no start date or cost has been given. The first step, per Unite.AI, is modelling how separate quantum modules would communicate, including link errors and latency, before any hardware is built.

Where AI and GPUs come in

Quantum processors make frequent errors, and correcting them requires classical computers to decode measurement data very quickly. NVIDIA says its TCBP-JOBS decoder, released in the CUDA-Q QEC 0.9.0 library, is up to 13,000 times faster than existing approaches in one comparison and cut one design study from 12 weeks to 45 minutes. Unite.AI says the baseline is a configuration of the Tesseract decoder on bivariate-bicycle codes, and describes the method as learning fault probabilities to steer GPU searches. These are vendor figures that have not been independently checked. Grassellino argues that "accelerated computing and superintelligence will play an important role" in designing such algorithms.

Scheduling quantum like any other job

NVIDIA also plans to add a quantum interface to Slurm, the open-source scheduler used by most supercomputers. IBM is contributing its Quantum Resource Management Interface (QRMI), which Unite.AI says was developed with NVIDIA and Pasqal as a vendor-neutral layer. The aim is for sites to book quantum processors alongside CPUs and GPUs with existing commands. CASRAI reports that Slurm 26.11 is targeted for general availability in late November 2026; NVIDIA has not given a firm date for the full feature set.