if i want to improve performance through server selection, is memory or compute optimized better for dask?

I think that will really depend on your workload and where you’re bottlenecked. Dask is essentially just distributing your workload across many machines; the requirements of the workload (memory, CPU) will be the same on the remote machine as they are locally. Dask itself isn’t too resource intensive, so I don’t think you need to worry too much about for example the Dask Scheduler

ok ill review the bottlenecks and see what is needed