
Nvidia Claims 13,000x Faster QEC Decoder as Fermilab Platform Announced
Nvidia, IBM and Fermilab announced a co-located quantum-GPU codesign platform at SC26, paired with a decoder Nvidia says is up to 13,000 times faster and new Slurm quantum scheduling.
- By
- Sophie Lindqvist
- Filed
- Channel
- AI & Compute
- Read
- 4 min read
Nvidia says its new quantum error-correction decoder runs up to 13,000 times faster than existing leading algorithms, cutting a 12-week research computation to 45 minutes. The company released the tool, TCBP-JOBS, on Nov. 16 in its CUDA-Q QEC 0.9.0 library, alongside a plan with IBM and Fermilab for a modular quantum codesign platform announced at SC26 in Chicago, running Nov. 15 to 20.
The speedup figures are Nvidia's own. They apply to research workloads. Nvidia did not present the result as real-time decoding on a running quantum computer. That distinction matters, and the announcement did not spell it out.
What would the Fermilab platform do?
The partners will co-locate Nvidia DGX B200 systems with IBM quantum computers, including machines powered by IBM Quantum Nighthawk, according to Nvidia. Fermilab frames the project in conditional terms: the partners "intend to collaborate," and a "proposed initial platform" would put the hardware in one place.
Neither announcement gave a launch date, a location or a dollar figure.
Fermilab's stated target is modularity: linking several quantum processors so they can work together on calculations too large for one device. A first phase would emulate communication between modules, measuring how link errors, latency and feedback affect performance. Those findings would guide work between IBM and Fermilab's SQMS Center on physical quantum interconnects. A later phase adds superconducting cavity hardware, tested as quantum memories and for bosonic error correction.
The division of labor, per Fermilab:
- IBM contributes superconducting processors, software and quantum networking expertise.
- Nvidia brings decoding, simulation, optimization and real-time feedback computing.
- Fermilab leads physics applications and coordinates researcher access.
"This would be an important step toward the capabilities and community needed for a future DOE quantum computing user facility," said Anna Grassellino, Fermilab's chief technology officer and SQMS Center director.
IBM's blog tied the platform to a DOE initiative aiming to deploy a fault-tolerant, scientifically relevant quantum computing capability by 2028. A separate codesign platform from IBM, AMD and Oak Ridge National Laboratory launched on Nov. 5, pairing IBM Nighthawk r2 processors with AMD-based classical systems in Poughkeepsie, New York.
How fast is the decoder, and by whose measure?
Error correction spreads information across many physical qubits. A classical program, the decoder, reads repeated measurements to identify errors so they can be fixed. Simulating decoders against huge numbers of error patterns is heavy classical computing.
Nvidia says TCBP-JOBS speeds up the intensive parts of that research by up to 13,000 times while keeping high accuracy. The company also said Fermilab used CUDA-Q to complete studies in three weeks that would normally take five months. Both claims come from Nvidia without independent verification.
The same day, IBM and AMD released Pelagos, an open-source decoder-testing tool built on the Triton framework. It runs on both AMD and Nvidia GPUs, and IBM said it ran decoder tests up to 97 times faster than a single CPU core.
Can Slurm schedule a qubit like a GPU?
Nvidia expanded Slurm, the open-source scheduler used widely at supercomputing centers, so quantum processors can be scheduled like CPUs and GPUs. Nvidia acquired Slurm's lead developer, SchedMD, in December 2025, according to IBM.
Nvidia said nearly 30 supercomputing centers worldwide are deploying more than 50 quantum processors, and that each new processor or site previously meant a new integration project. The release lets a hardware maker write a single interface for its machines, building on the QRMI and QDMI interfaces.
Per IQM, Slurm 26.11 adds plugin support for Hierarchical Resources (HRES), letting a center describe a quantum processor as a shared, cluster-level resource. IBM is integrating QRMI with HRES, so a researcher can request a QPU in a job script much as they would a GPU. Finland's IQM supports the effort; its partner MQSC built an open-source orchestrator using HRES and QDMI to schedule IQM systems.
What did the chemistry runs show?
IBM, RIKEN and Nvidia researchers ran a sample-based quantum diagonalization calculation of Fe4S4, an iron-sulfur cluster found in proteins, on ROQUO, RIKEN's new supercomputer built on Nvidia GB200 NVL4 systems. Starting from a broken-symmetry reference state, the result came within 4.3 millihartree of the density-matrix renormalization group benchmark. IBM says that is 79 times closer than the team's earlier SQD result on the same cluster.
Separately, work by Cleveland Clinic, RIKEN and IBM modeled protein-ligand complexes of up to 12,635 atoms with IBM Heron processors and classical computers. It is a finalist for the 2026 ACM Gordon Bell Prize. Nvidia said the system is more than 40 times the size of the previous 303-atom quantum-classical benchmark.
The SC26 releases extend Nvidia's quantum push from earlier this month, including an FPGA controller design with Altera and Oxford Quantum Circuits and its Rubin HPC science GPU, due in 2028. Until the Fermilab platform gets a launch date and a budget, the decoder speedups and the Slurm integration remain the only deliverables already in users' hands.
Source: mlq.ai
More from Sophie Lindqvist
Show full bio
News editor covering money, funding and market shifts around Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.
1,055 articles
Related articles
infleqtion-and-nvidia-hit-5-4-physical-qubits-per-logical-qubit-27e8c610
Infleqtion and Nvidia Hit 5.4 Physical Qubits Per Logical Qubit
nvidia-ships-gpu-library-aimed-at-ai-inside-chip-design-and-fabs-aeb4d368
Nvidia Ships GPU Library Aimed at AI Inside Chip Design and Fabs
ionq-12-cores-decode-408-logical-qubits-in-simulation-31416165
IonQ: 12 Cores Decode 408 Logical Qubits in Simulation
chipmakers-turn-to-nvidia-cuda-x-libraries-for-ai-defined-fabs-619c3edb
Chipmakers Turn to NVIDIA CUDA-X Libraries for AI-Defined Fabs



