Horses for courses: IBM trots out CLOPS as measure of quantum performance

Wolfram Language 12.1

The IBM Quantum team is proposing a new metric for the performance of a quantum computing system, which it calls Circuit Layer Operations Per Second, or CLOPS.

According to the IBM Quantum team, performance in a quantum computing system requires three critical attributes: scale, quality, and speed. For scale, IBM measures progress by the number of qubits in its systems, while for quality the firm uses a measurement it calls Quantum Volume, which takes into account the capabilities and error rates of a particular quantum processor.

However, where speed is concerned, an appropriate system-agnostic metric that captures all dependencies across hardware and software of circuit executions has not so far been defined, IBM claims. This is where CLOPS comes in, as detailed in a post on the IBM Research blog.

According to the Big Blue, CLOPS is a metric correlated with how fast a quantum processor can execute circuits, where a circuit is a computational routine consisting of coherent quantum operations, which can be thought of as comparable with an algorithm in classical computing. Specifically, the metric measures the speed at which the quantum processor can execute layers of a parameterised model circuit of the same sort used to measure the Quantum Volume.

The benchmark used for CLOPS requires execution of many instances of this model circuit with different parameters generated at runtime. Various parts of the combined hardware/software stack contribute to CLOPS, including the repetition rate of the quantum processor, the speed at which gates run, the runtime compilation, the amount of time it takes to generate the classical control instructions, and the data transfer rate among all units.

The IBM Quantum team has found that initial timing measurements using the benchmark have identified several previously unknown speed bottlenecks in its own systems. Not to be deterred, IBM said that improvements in quantum hardware will reduce circuit delay times, and further advances in the runtime architecture will reduce initialisation times for data loading as well as improved runtime compilation.