CuQuantum SDK's cuStateVec: From Single-GPU to Distributed State Vector Simulation

NVIDIA's cuQuantum SDK cuStateVec component provides exact quantum circuit simulation using the quantum state representation. With common primitives, cuStateVec enables users to build state vector simulators accelerated by NVIDIA GPUs. Memory and computational requirements scale exponentially with qubit count, which demands high-performance computing resources. Through continuous development, cuStateVec delivers high performance across multiple generations of GPUs, including the latest architectures. To extend cuStateVec's capabilities, we have added Ex APIs, an abstraction layer for distributed simulations with multiple GPUs and multiple processes. The Ex APIs introduce the distributed state vector as the foundation object and are designed as a straight-forward extension of the cuStateVec APIs. This approach overcomes single-device memory limitations through transparent distributed computing. This presentation showcases performance data for single-GPU cuStateVec and scalability results for cuStateVec Ex APIs.