NVIDIA cuQuantum 23.10

NVIDIA cuQuantum is an SDK that provides optimized libraries and tools for accelerating quantum computing workflows. It allows developers to speed up quantum circuit simulations based on state vector and tensor network methods using NVIDIA Tensor Core GPUs. The latest version, cuQuantum 23.10, introduces several new features.

Support for NVIDIA Grace Hopper systems

The new release now includes support for NVIDIA Grace Hopper systems. These systems offer improved performance compared to other CPU and GPU architectures when used in conjunction with NVIDIA CPUs. For example, a simulation of 40 qubits that previously required 128 NVIDIA H100 80GB GPUs can now be performed using just 16 NVIDIA Grace Hopper systems. This not only leads to significant speedups but also results in cost and energy savings for each workload.

Tensor network high-level APIs and gradients

cuTensorNet provides high-level APIs that enable quantum simulator developers to program in an intuitive way and take advantage of its capabilities. This technology abstracts away specific tensor network knowledge, making it easier to build tensor-network-based quantum simulators. It covers various elements such as expectations, measurements, and samples, enabling drastic speedups for quantum machine learning (QML) and adjoint differentiation-based workflows.

Other optimizations

In addition to the above features, cuQuantum 23.10 includes various API-level and kernel-level optimizations to further enhance performance. These optimizations contribute to faster runtimes on Grace Hopper systems compared to other CPU and GPU architectures.

Getting started

To get started with cuQuantum, documentation is available to assist users in setting up their environment. Users are encouraged to check out marketplace listings for each major CSP (Cloud Service Provider) if they are running a CSP. Additionally, a benchmark suite is provided on GitHub for validation and testing of GPU engagement in benchmarks. Users can also reach out with any questions, requests, or issues on GitHub.

Overall, NVIDIA cuQuantum 23.10 offers significant acceleration for quantum circuit simulations, making it an essential tool for developers working in the field of quantum computing.