Benchmarking State Of The Art Ising Machines

As Moore's Law comes to an end, high performance computing has often turned to dedicated hardware capable of solving a specific family of problems for increased computational power. Quadratic Unconstrained Binary Optimization (QUBO) problems, or equivalently, two-body Ising problems, have been the focus of a multitude of new dedicated hardware devices of different types: fully classical (Fujitsu, Toshiba, STATICA, MemComputing, etc.), semiclassical (Coherent Ising Machine), and quantum (D-Wave). Development of QUBO-specific algorithms has likewise been prolific. Benchmarks across these QUBO solvers have not been uniform, partly due to differences in characteristics like connectivity, precision, maximum size, etc. Important questions like the current state of quantum QUBO solvers vs. classical QUBO solvers, and more practically which solver to use when, have gone only partially addressed. In this work, we seek to provide answers to these questions by uniform benchmarks across a wide range of state-of-the-art solvers, both classical and quantum, including the establishment of scaling comparisons on planted solution instances.