Workshop: The 2nd International Workshop for Software Frameworks and Workload Management on Quantum-HPC Ecosystems
Authors: Chandrachud Pati and Yogesh Simmhan (Indian Institute of Science, Bengaluru)
Abstract: Quantum processors increasingly act as accelerators in HPC workflows, where the runtime schedules the work. Many search problems split into independent parts, but the runtime can exploit this only if the workflow exposes those parts and a per-part validity check. SemQBench compares seven execution policies on the 20-qubit IQM Garnet processor, varying one choice at a time: whether to amplify, whether partial results combine across runs, and what a retry keeps. Amplifying the whole problem gives no clear gain over random sampling at 6.7-104x more device time. Collecting valid partial results across a batch cuts executions for 95% confidence by 2.20-4.00x and wins in 99/108 overhead settings. Fewer executions need not mean less device time: one workload drops from thirteen executions to four as device time rises from 7.44-9.81 μs. Exposed structure gives the runtime a strong default and the evidence to drop it when it does not pay.
Back to The 2nd International Workshop for Software Frameworks and Workload Management on Quantum-HPC Ecosystems Archive Listing
Back to Full Workshop Archive Listing