Rework the benchmarking code to use a deterministic set of work items with a known number of diff share nonces at regular spaced intervals