ORNL, IBM and AMD pair a 120-qubit quantum processor with GPU supercomputing for error correction
The co-design platform at IBM's Poughkeepsie site links IBM Quantum Nighthawk with ORNL's Umbra system on AMD Instinct MI355X GPUs and EPYC CPUs for decoding research; IBM says the field needs a roughly 230-fold rise in hard operations to hit DOE's late-2028 target.
Why it matters
Scaling quantum error correction is now limited as much by the classical computing that decodes errors in real time as by qubits, and this platform puts a GPU supercomputer next to IBM's most powerful processor to test that link. IBM says its experiments have reached 70 logical qubits and 438 hard operations, while DOE's Q Competition targets 100 logical qubits running 100,000 hard operations by late 2028. That roughly 230-fold gap is the number to watch, along with whether decoding latency on Umbra is low enough for live correction.
Key facts
- Platform sited at IBM in Poughkeepsie, New York
- IBM Quantum Nighthawk: 120 qubits
- ORNL Umbra uses AMD Instinct MI355X GPUs and 5th Gen EPYC CPUs
- IBM experiments: up to 70 logical qubits, 438 hard operations
- DOE target: 100 logical qubits, 100,000 hard operations by late 2028
- Roughly 230-fold increase needed