Strategy

Configuration Efficiency

Where adding more cores, memory, or storage to a Power configuration stops producing proportional economic value, and how to spot the sweet spot before you over-configure.

Every Power model has a configuration zone where cost per unit of usable capacity is lowest. Configure past it, and cost climbs faster than usable throughput -- often because you have crossed an IBM i software tier boundary or exceeded a usable-memory cap the extra hardware cannot actually help you use.

Three places diminishing returns show up

Core count vs. software tierOne extra core can push a partition from P10 into P20 pricing -- a step-function cost jump, not a gradual one.
Memory vs. usable capEntry models often cap usable IBM i memory well below their physical memory ceiling; extra DIMMs beyond that cap only help non-IBM i partitions.
Storage vs. workload needNVMe capacity scales cost close to linearly, so over-provisioning storage rarely pays for itself the way over-provisioning cores or memory tier does.

How to find your sweet spot

Start from your actual workload's required CPW or rPerf, not the largest configuration a vendor quotes by default. Size to the smallest core count and lowest software tier that comfortably clears your requirement with headroom for growth, then check whether the next tier up is close enough to be worth pre-buying versus revisiting at the next hardware refresh.

This is the same logic behind Workload Value's sizing guidance -- read that page for profile-specific recommendations (small IBM i, mid-range IBM i, AIX, mixed).

Related reading