CPW and rPerf help you compare processor capacity. They do not size the whole Power11 system.
The useful move is to turn your current workload into a benchmark target, add room for growth, and then test that answer against the operating system, partition, storage, I/O, and resilience plan. That is how a benchmark becomes a better quote instead of a bigger number.
CPW for IBM i workloads
Commercial Processing Workload (CPW) is IBM's relative capacity measure for IBM i commercial workloads. A higher CPW value gives you more benchmark capacity, but the comparison still needs the same workload type, IBM i tier, and logical partition limit.
S1122 shows why that last point matters. Midland lists S1122 in P10 and gives most processor builds a maximum IBM i LPAR size of 4 cores or 118,200 CPW, with an exception for ERGR processors. Buying more hardware cores does not automatically create a larger IBM i partition.
rPerf for AIX and Linux workloads
Relative Performance (rPerf) is the matching comparison measure for AIX and Linux. The current performance tables publish rPerf by processor configuration and simultaneous multithreading mode, so the number needs the same context on both sides of the comparison.
AIX and Linux sizing still needs observed CPU use, peak windows, memory pressure, storage latency, and I/O demand. rPerf tells you whether the processor comparison is moving in the right direction. It does not tell you whether the rest of the system can keep up.
A worked S1122 and S1124 sizing example
Suppose an IBM i system is rated at 320,000 CPW and its highest sustained window uses 75 percent of that capacity. The team expects 25 percent growth during the next planning period.
320,000 times 75 percent.
240,000 times 1.25.
Below the example target.
Above the example target.
The arithmetic points to S1124 in this example. It does not finish the design. The next check is whether the peak is CPU-bound, whether one IBM i partition needs the capacity, and whether memory, storage, I/O, high availability, and batch time support the same choice.
What to bring into a Power11 sizing call
A reseller can work faster when the benchmark target arrives beside the workload facts that created it.
- Current model, processor feature, active cores, and published CPW or rPerf
- Observed CPU use during the busiest sustained window
- IBM i, AIX, Linux, and VIOS partition map
- Current memory use and paging behavior
- Storage capacity, latency, throughput, and adapter demand
- Growth, high availability, recovery, and service-life assumptions
That is the difference between asking for a Power11 box and asking for a configuration that can do the work.
The worked example explains the method. It is not a sizing recommendation. Use current Collection Services or other performance records and have the final configuration checked against the complete workload and partition plan.