Phasecraft ran 3,000 quantum emulations on Nvidia GPUs

Phasecraft has produced what it says is the largest known database of variational quantum eigensolver results, more than 3,000 emulations across 13 molecular systems.

The work ran on Nvidia Hopper processors hosted at the University of Nottingham, using Nvidia’s cuQuantum toolkit, and the company reports a 15-fold speedup over earlier results for modelling many-body systems.

The emulations covered circuits of four to 32 qubits, concentrated in the 24-to-28 range. That detail is the one to hold on to, because it places the whole exercise inside the territory where classical computers can still simulate quantum ones.

No quantum hardware was used. Phasecraft built a picture of what a quantum computer would calculate, on machines that are not quantum computers, which is currently the only way to generate data at this volume.

Doing it this way is reasonable, and the company is not hiding it. Quantum chemistry is the application most often promised for quantum computing, and useful hardware does not yet exist.

Building the training data now, so that quantum-enhanced density functional theory has something to work from when machines arrive, is a defensible strategy rather than a workaround.

Phasecraft cites no paper and no preprint for the 15x figure, and does not name the benchmark it improved on beyond “prior results” for many-body systems.