Quantum Volume
Quantum volume is a single benchmark that folds together qubit count, connectivity, gate fidelity, and crosstalk into one number.
A holistic metric
Counting qubits alone overstates a machine's power: a hundred noisy, poorly connected qubits may run less than a dozen good ones. Quantum volume (QV), introduced by IBM, is a hardware-agnostic benchmark that captures the largest square random circuit a device can run reliably, combining number of qubits, connectivity, gate fidelity, compiler quality, and crosstalk into one figure.
How it is defined
Run random circuits on m qubits with m layers each (a square circuit of width and depth m). Each layer pairs qubits randomly and applies random two-qubit unitaries. A device passes at width m if its outputs match the ideal distribution well enough, measured by the heavy-output probability exceeding two-thirds with statistical confidence. The quantum volume is 2 raised to the largest passing m.
What it rewards
- Higher gate fidelity, since deeper circuits demand fewer errors
- Better connectivity, since random pairings need less routing
- Good compilation that minimizes added SWAPs
- Low crosstalk when many gates run at once
Uses and limits
Because it is a single comparable number across modalities, QV became a popular headline metric, and reported values have climbed rapidly. Its limits: it saturates when circuits are limited by width rather than depth, it rewards all-to-all connectivity, and being exponential it exaggerates linear progress. It also does not measure speed.
QV is best read as one axis of performance. Speed benchmarks like CLOPS and application-specific tests complete the picture, and no single number substitutes for measuring the actual algorithm one intends to run.