“Lots of tiny tasks” workloads, mostly. For more coarse-grained parallelism the cache thing matters a lot less. I still favor work-stealing schedulers as the de

“Lots of tiny tasks” workloads, mostly. For more coarse-grained parallelism the cache thing matters a lot less. I still favor work-stealing schedulers as the default, since robustness across different types of workload is a very good thing, but they’re definitely not always needed.