> In this situation there are a lot of processes running, for which increased cores-count and memory availability seems worth it. It's always the workload type

> In this situation there are a lot of processes running, for which increased cores-count and memory availability seems worth it. It's always the workload type. For the mixed environments (some with a heavy constant load while the other have only some occasional spikes) the increase of RAM per node is the most important part what allowed us to actually decrease the node count. The whole racks with multiple switches was replaced by a single rack with a modest amount of servers and a single stacked switch.