- Hardware
- Also: circuit layer operations per second
CLOPS (Circuit Layer Operations Per Second)
CLOPS is a quantum computing throughput metric measuring how many layers of quantum circuit operations a quantum processor can execute per second, complementing quantum volume as a hardware performance benchmark.
CLOPS (Circuit Layer Operations Per Second) is a benchmark metric introduced by IBM Quantum in 2021 to measure the speed at which a quantum processor can execute circuits. While quantum volume measures the largest random circuit a device can run reliably, CLOPS measures how quickly a device can run many circuits, which matters for practical workloads that require large numbers of shots.
Why CLOPS Was Needed
Quantum volume captures quality (circuit depth and width before errors dominate) but says nothing about throughput. Two devices with the same quantum volume might differ dramatically in execution speed. A device running 100 shots per second and a device running 100,000 shots per second have the same quantum volume, but the latter finishes a variational optimization in minutes rather than days.
For applications like VQE, QAOA, and quantum machine learning, where thousands of circuit executions are required per optimization step, throughput is as important as fidelity.
Definition and Measurement
CLOPS is measured by running a specific benchmark: a parameterized version of the quantum volume circuit, with parameters updated between circuit executions to simulate the real-time parameter updates common in variational algorithms.
The benchmark runs V layers of circuits, each with D = V qubits, using W = 100 circuit updates (different parameter sets) and K = 100 shots per update. The CLOPS value is:
CLOPS = (V * K * W * S) / T
where V is the quantum volume square size, S is the number of independent samples, and T is the total wall-clock time including all classical overhead (parameter updates, transpilation, communication).
This means CLOPS includes the full software stack overhead, not just raw gate speed.
Typical Values
At time of publication, IBM superconducting devices ranged from approximately 1,000 to 15,000 CLOPS. Trapped-ion devices from IonQ and Quantinuum typically score lower (hundreds of CLOPS) due to slower gate times, despite having higher gate fidelities.
For context, a VQE experiment with 1,000 optimization steps, 100 observables, and 1,000 shots per observable requires approximately 10^8 circuit layer executions. At 10,000 CLOPS, this takes about 10,000 seconds (roughly 3 hours). At 100 CLOPS, it takes 300 hours.
CLOPS vs Quantum Volume
Quantum volume and CLOPS measure orthogonal properties:
- Quantum volume measures circuit quality: depth, width, and fidelity.
- CLOPS measures throughput: how quickly circuits can be submitted and executed.
A high quantum volume with low CLOPS means accurate but slow. A low quantum volume with high CLOPS means fast but shallow circuits only. The ideal hardware has both.
Limitations
CLOPS is defined for a specific benchmark circuit type and may not generalize to all workloads. Circuits with many classical feedforward operations (dynamic circuits) have lower throughput than parameter-sweep circuits. The CLOPS metric does not capture latency, which matters for real-time feedback applications like quantum error correction.
As hardware architectures evolve (particularly with the shift to modular and networked quantum computers), CLOPS will need to be extended to account for inter-module communication overhead.