In June of this year, Xanadu released version 0.31 of its Pennylane quantum programming platform which had a new feature that supports multi-node/multi-GPU calculations. This capability requires explicit installation of the NVIDIA cuQuantum SDK but when that is installed it opens up the capability to perform large state-vector simulations. The Brookhaven National Laboratory is current running simulations that use up to 256 NVIDIA A100 Tensor Core GPUs to simulate about three dozen qubits and Xanadu has an end of year a goal of supporting simulations that have up to 1000 nodes (or 4000 GPUs) to provide state vector simulations of more than 40 qubits. This may not seem a great many but it is important to remember that every time one adds one more qubit to a simulation, the hardware resources needed to perform a classical simulation will double. Both Xanadu and NVIDIA have posted blogs about their work with multi-GPU simulations with Pennylane. The NVIDIA blog can be seen here and the Xanadu blog is available here.
September 12, 2023