cuQuantum: Accelerating Quantum Circuit Simulation on GPUs

Quantum circuit simulation plays an important role in quantum computing, as it enables algorithm research at a scale and performance level not possible on today's quantum computers. We present cuQuantum, a set of libraries to accelerate quantum circuit simulations on GPUs. cuQuantum contains libraries for state vector and tensor network simulation, and has already been integrated as a backend into several leading frameworks. We benchmark cuQuantum against current state-of-the-art CPU and GPU-based simulation solutions. Using cuQuantum on NVIDIA's Selene supercomputer, we are also able to simulate a variational quantum algorithm circuit to solve a MaxCut problem with thousands of qubits, an order of magnitude increase over the previous largest solution. We expect and hope that cuQuantum will open the door for novel algorithms research, at scales not previously possible.