Fugaku Lost To A Seven-Hour Quantum Run
The Seven-Hour Answer
Japan's Fugaku supercomputer spent roughly 700,000 processor-hours this summer trying to reproduce a single materials simulation, and it still could not match what an IBM quantum computer did in seven hours. Fugaku is not a toy: it was built for exactly this kind of physics problem. The task, run under a partnership between IBM and quantum-software firm Qedma, modeled how a material responds when struck by a laser: the electrons get knocked out of their resting state and the surrounding lattice starts to oscillate, and predicting the shape of that oscillation is precisely what classical simulators are supposed to be good at. IBM's quantum chip, called Heron, encoded the material's quantum state directly onto its own qubits (the basic quantum bit, which unlike a normal bit can hold a mix of 0 and 1 at once) instead of approximating the physics step by step the way a classical machine must. Fugaku ground through that approximation for the equivalent of roughly 80 years of one processor's time and never landed on the right oscillation. Eighty processor-years against seven hours is the number that should worry whoever is budgeting the next generation of classical supercomputers built to handle exactly this kind of physics problem.
The Open Scoreboard
IBM turned the demonstration into a public dare by publishing the underlying circuits, not just the results, so any outside team can try to reproduce or break them. Alongside the Qedma materials run, IBM and the University of Chicago executed a separate circuit built from 70 logical qubits (qubits wrapped in extra hardware so one stray error does not ruin the whole calculation), firing 2,415 two-qubit logic gates in about 15 minutes with error rates ten times lower than the raw hardware underneath. A third demo, run with quantum-software firm Algorithmiq on disordered materials relevant to battery electrolytes and catalysts, has sat on IBM's public Quantum Advantage Tracker for eight months; Algorithmiq co-founder Sabrina Maniscalco says roughly half the classical methods thrown at it have failed outright, and none has matched it in full. More than 30 outside submissions have now tried to break one of the three results. Unlike a paper that stands or falls once at peer review and then gets cited for a decade regardless, IBM built a scoreboard that Google's quantum team, or anyone else, can walk up to and challenge on its own schedule. For the engineers who keep Fugaku running, and the ministry that funds it, the eight months the Algorithmiq result has stood unbroken against more than 30 outside submissions is the fact on the board now.
None of the three demonstrations ran on a real drug molecule or a real battery chemistry. They ran on problems built to be hard for classical computers and comparatively easy for these particular quantum chips. That is not a small caveat. The open tracker proves nobody has broken the claim yet on IBM's chosen ground, on engineered test problems rather than industrial materials or chemistry, and the open question is whether that will ever translate into a paying commercial use case, or whether the goalposts simply move again once someone tries.
Sources
- IBM Newsroom (newsroom.ibm.com)
- Next Platform (nextplatform.com)
- IBM Quantum Blog (ibm.com)
- Yahoo Finance (finance.yahoo.com)
- Postquantum.com (postquantum.com)