NVIDIA Claims 13,000x Decoder Speedup in IBM and Fermilab Quantum Plan
NVIDIA, IBM and Fermilab propose a modular quantum-classical platform; NVIDIA's 13,000x decoder speedup claim applies to one benchmark against unsparsified Tesseract, not all workloads.
By Olivia Hart
4 min read

What's News
- NVIDIA announced the plan on November 16, 2026, claiming up to 13,000x decoder speedup in one benchmark against unsparsified Tesseract.
- TCBP-JOBS cut one research computation from 12 weeks to 45 minutes; sparsified-preset speedups ranged 10x to 7,400x.
- IBM Quantum Nighthawk supports up to 7,500 two-qubit operations on 120 qubits at up to 100,000 samples per second.
- Fermilab frames the co-located DGX B200 platform as a proposal, not an operating facility.
- Slurm 26.11 with QRMI/QDMI quantum-HPC interfaces is targeted for later in November 2026.
By Olivia Hart
NVIDIA says its newly released GPU-accelerated quantum error-correction decoder delivered up to a 13,000-times speedup in a specific benchmark comparison. The claim, announced on November 16, 2026, sits at the center of a plan by NVIDIA, IBM and Fermi National Accelerator Laboratory to build a modular quantum-classical co-design platform. Readers should note the qualifier: the figure applies to one benchmark, not to all workloads.
What exactly did NVIDIA announce?
The decoder, called TCBP-JOBS, is available through CUDA-Q QEC 0.9.0. It targets the classical half of quantum error correction. Error correction encodes quantum information across multiple physical qubits; a decoder interprets measurement data to identify likely errors so they can be corrected. NVIDIA says the algorithm cut one research computation from 12 weeks to 45 minutes — a reduction that lets researchers evaluate candidate codes and decoder configurations far faster.
The fine print matters. NVIDIA Research's abstract specifies that the 13,000-times result was measured against unsparsified long-beam Tesseract on bivariate-bicycle codes, at logical-error-rate ratios of 0.97 to 1.57. Against Tesseract's sparsified presets, reported serial speedups ranged from 10 to 7,400 times. Tests covered surface, color and bivariate-bicycle code memory and surgery circuits. NVIDIA's own results do not establish a uniform speedup across decoders or workloads.
The method combines trajectory-calibrated belief propagation, which learns calibrated fault probabilities from belief-propagation trajectories, with jittered OSD beam search, which uses those probabilities to guide parallel GPU searches. NVIDIA Research reports logical error rates comparable to Tesseract and up to 4.5 times lower on the largest codes, plus batched execution for offline simulation and millisecond-scale latency in serial memory and logical-surgery experiments for QPUs with longer logical clock cycles. Comparable performance with much higher speed is the real result; the five-figure number is the marketing frame.
Is the platform real yet?
Not yet. Fermilab describes the initiative as a proposal rather than an operating facility. The initial design would physically co-locate IBM quantum computers with NVIDIA DGX B200 accelerated-computing systems, including machines powered by IBM Quantum Nighthawk. IBM calls Nighthawk its most powerful quantum processor, supporting circuits of up to 7,500 two-qubit operations on 120 qubits at speeds of up to 100,000 samples per second.
The scientific program would concentrate on high-energy and nuclear physics. Fermilab would lead engagement with researchers and coordinate software workflows and access through HEPCloud. The partners would develop algorithms, shared software and benchmarks for lattice gauge theories, neutrino interactions and nuclear simulations, using those problems to translate scientific needs into requirements for future fault-tolerant quantum systems.
Modularity is the stated goal: connecting processors for calculations too demanding for a single device. Researchers would first model and emulate communication between modules, studying link errors, latency and feedback. That work would guide IBM and Fermilab's Superconducting Quantum Materials and Systems (SQMS) Center on physical interconnects. In a later phase, SQMS intends to integrate ultracoherent superconducting cavities and investigate their use as quantum memories, entanglement buffers and resources for bosonic error correction.
Fermilab chief technology officer and SQMS director Anna Grassellino said the initiative would help establish the capabilities and research community needed for a future Department of Energy quantum computing user facility. That is an aspiration stated as a step toward another aspiration — a DOE user facility that does not yet exist.
How does scheduling change?
A separate integration changes how supercomputing centers allocate quantum hardware. Slurm, the dominant HPC scheduler, previously had no native concept of a quantum processing unit, so sites attached QPUs through plugins outside the scheduler's core resource model. IBM, NVIDIA and Pasqal worked with the open-source community on the Quantum Resource Management Interface (QRMI), a vendor-neutral layer that connects to Slurm Hierarchical Resources.
Unlike GPUs, which operate within individual computing nodes, QPUs can be shared across multiple nodes. Hierarchical Resources lets administrators manage that shared resource, allocate and reserve QPU time, and account for it under familiar high-performance-computing policies. A hybrid workload can therefore run as a normal Slurm job without a separate queue or login.
NVIDIA's Slurm announcement says support for the QRMI and QDMI quantum-HPC interfaces will let QPU builders reuse one integration across deployments rather than create bespoke software for every QPU-center combination. Listed adopters are Alice & Bob, IBM, IQM, Pasqal, Quantinuum and MQSC. Slurm 26.11 is targeted for availability later in November 2026.
What is the honest takeaway?
The proposed platform and software releases address three distinct constraints: evaluating error correction, scheduling shared hardware and connecting processors. The first interconnect studies are the central hardware test, because their measurements of link errors, latency and feedback are meant to establish what later fault-tolerant, multi-processor systems must handle — before Fermilab adds any cavity hardware. Until those measurements exist, the partnership is a well-specified plan, not a machine.
Source: unite.ai
More from Olivia Hart
Show full bio
Staff writer covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.
2,203 articles