Benchmarking GPU Clusters with the Jülich Universal Quantum Computer Simulator
, Forschungszentrum Jülich GmbH
Simulating quantum computers is a versatile way to benchmark supercomputers with thousands of GPUs. We'll explain quantum computer simulators from a linear algebra perspective, using the Jülich Universal Quantum Computer Simulator (JUQCS) as an example. We'll show how the memory-, network-, and computation-intensive operations of JUQCS can be used to benchmark high-performance computers. In particular, we'll illustrate the CUDA-aware message-passing interface communication scheme. We present results for benchmarking JUWELS Booster, a cluster with 3,744 NVIDIA A100 Tensor Core GPUs.