Accelerating Large-Scale Simulation of Quantum Processing Units on Graphical Cards via cuQuantum SDK

Efficient simulation/emulation of Quantum Processing Units (QPU) at the fundamental quantum-mechanical level is crucial for improving their design and practical performance. Existing software packages, like QuTiP and scQubits, already provide broad functionality required for computing the energy spectrum and quantum dynamics of a given QPU design. While very efficient for small-size simulations, these software packages struggle with scalability and performance for larger simulations due to the lack of efficient support for GPU computing. To address the scalability and performance challenges exhibited by these software, NVIDIA's cuQuantum SDK offers the cuDensityMat library, providing scalable GPU-accelerated building blocks for quantum simulator developers. Specifically, the cuDensityMat API allows offloading of the critical time-consuming computations to NVIDIA GPUs, including support for both single- and multiple-/many-GPU processing. The cuDensityMat library accelerates quantum master equation solvers by orders of magnitude, as recently demonstrated by the integration with the QuTiP and CUDA-Q software packages. Similar acceleration can also be achieved for the extreme eigen-value solvers used in the design of the superconducting QPU via the scQubits software package, for example. To enable even larger quantum dynamics simulations, the cuDensityMat library now provides the APIs for implementing the MPS-TDVP quantum dynamics solvers, opening a route to a more efficient emulation of the latest-generation ion-trap and neutral-atom based QPUs.