Home » Quantum Computing » NVIDIA Says New Decoder Speeds Up Quantum Error Correction 13,000x

NVIDIA Says New Decoder Speeds Up Quantum Error Correction 13,000x

Fermilab’s own chief technology officer calls the new quantum computing platform a proposal, not a reality, even as NVIDIA’s release describes it as already running.

Abstract red circuit board chip rendering symbolizing a quantum error-correction decoder
  • NVIDIA says its new TCBP-JOBS algorithm cuts quantum error-correction research time from about 12 weeks to 45 minutes.
  • IBM and NVIDIA are pairing new DGX B200 systems with IBM’s fastest quantum computer, Nighthawk, as part of a codesign platform with Fermilab.
  • A separate IBM and RIKEN chemistry project in the same announcement is a 2026 ACM Gordon Bell Prize finalist, for modeling a 12,635-atom protein system.

NVIDIA, IBM and the Fermi National Accelerator Laboratory announced a three-way research push on November 16, 2026, from Chicago. The news came in a press release NVIDIA distributed itself. Interesting Engineering published a same-day writeup checking NVIDIA’s numbers against the rest of the announcement. Two things in this announcement are real and usable today. One is a new, faster quantum error-correction decoder for research use. The other is an expanded quantum-plus-GPU hardware setup at IBM’s facilities. A third piece, the joint research platform with Fermilab, is still being built.

What NVIDIA’s Quantum Error-Correction Decoder Actually Speeds Up

Quantum computers make mistakes constantly. Engineers protect against that by spreading one reliable “logical” qubit across many physical qubits. The approach is called error correction. A classical computer program called a decoder reads the error signals and figures out what went wrong. Decoding research takes enormous amounts of computer time. Researchers test new error-correction codes and decoder designs against each other to see which ones actually work.

NVIDIA’s new algorithm, TCBP-JOBS, targets that research step specifically. The company says it cuts the research time for a new quantum error-correction design from about 12 weeks down to 45 minutes. That’s a claimed speedup of up to 13,000 times over existing leading decoders. The algorithm is available now through NVIDIA’s CUDA-Q QEC 0.9.0 software library. Researchers can download it and run it on their own hardware.

That distinction matters. TCBP-JOBS doesn’t make a live quantum computer less error-prone today. It makes the research into better error-correction designs faster. That’s a more modest claim than the “13,000x” headline might suggest.

Fermilab’s Role, and What’s Still a Plan

Alongside the software release, IBM and NVIDIA are placing new DGX B200 systems next to IBM’s quantum computers. That includes machines built around Nighthawk, IBM’s fastest quantum processor to date. That hardware pairing is already happening.

The three organizations also describe a “modular quantum codesign platform” built with Fermilab, the Department of Energy’s particle-physics lab near Chicago. Fermilab would contribute its expertise in high-energy and nuclear physics. That expertise would help define what a future fault-tolerant quantum computer actually needs to do. But the language in NVIDIA’s own release is inconsistent about how far along this part is. One line says the hardware pairing “is supporting” the platform already. Anna Grassellino, Fermilab’s chief technology officer, called it something different. In her quote, the platform is “proposed,” one that “would combine” the three organizations’ strengths. Proposed and already running are not the same thing, and the release does not resolve which is accurate.

A Bigger Molecule, Not a New Idea

The same announcement points to a separate, independently confirmed result. IBM, RIKEN and Cleveland Clinic ran a joint chemistry project, one of several teams now pairing quantum and classical computers to simulate hard molecules. That team modeled a protein-ligand system containing 12,635 atoms. The result is more than 40 times bigger than the team’s own 303-atom benchmark from less than a year earlier. They used IBM’s quantum processors alongside RIKEN’s Fugaku and ROQUO supercomputers.

The work was named a finalist for the 2026 ACM Gordon Bell Prize, one of supercomputing’s most prestigious awards. Cleveland Clinic and RIKEN each confirmed the finalist status in their own newsrooms, separately from NVIDIA’s release. The winner will be announced at the SC26 conference in Chicago on November 20, 2026. That’s four days after this announcement.

Who Benefits and How

None of this reaches an ordinary consumer product this year, or probably for several more. The direct beneficiaries are researchers at universities, national labs and quantum hardware companies who already use NVIDIA’s open-source CUDA-Q tools. The faster decoder is free to download today. That matters most for a small academic team with no budget for months of supercomputer time.

The indirect case is the one that eventually touches more people. Error correction is what stands between today’s small, noisy quantum computers and machines reliable enough to help with drug discovery, better batteries or cheaper fertilizer. Other teams are chasing the same goal from different angles. Faster research tools don’t guarantee that future arrives any sooner, but they remove one of the slower steps along the way.

Where the Evidence Stops

Every number in this announcement comes from NVIDIA, IBM or their named partners, not from an outside lab. The 13,000x decoder speedup is NVIDIA’s own benchmark against its own choice of comparison algorithms. Nothing in the release points to independent reproduction yet.

The codesign platform with Fermilab is the shakiest part of the story. NVIDIA’s own release describes it two ways: already supporting research, and only a proposal. There’s no public timeline for whether, or when, it moves from one status to the other.

The chemistry result is on firmer ground. Cleveland Clinic and RIKEN have each confirmed the Gordon Bell finalist status in their own announcements. The prize itself, though, hasn’t been awarded yet.

SC26 runs through November 20, 2026, in Chicago. That’s the nearer deadline worth watching. Will the Gordon Bell judges agree this quantum-assisted chemistry work beats the all-classical competition for one of supercomputing’s top prizes?

Sources

Related Reading