Power7-to-Power11 Sizing Guide
Translate an old Power7 workload into modern Power11 (or Power10) capacity so you don't over-buy or under-buy the replacement -- most Power7 workloads need far less modern hardware than the core count suggests.
The single most common sizing mistake in a Power7 migration is treating it as a core-for-core replacement. It isn't, and the gap is large: per-core performance has grown dramatically across four generations, so a modern Power11 system needs a fraction of the cores to match -- and usually comfortably exceed -- a legacy Power7 system's actual throughput.
Why core count is the wrong starting metric
IBM Power performance is measured in CPW (Commercial Processing Workload, used for IBM i sizing) or rPerf (relative performance, used more broadly). A single Power11 core delivers dramatically more CPW/rPerf than a single Power7 core -- the exact multiple varies by specific model and clock speed on both ends, but it is not a small difference. An entry Power 730 Express core rated in the low thousands of CPW is not remotely comparable to a current Power11 core. Sizing on core count alone will lead you to either wildly over-buy (assuming you need a like-for-like core count) or, less commonly, under-buy if you undersize based on a rough guess.
A practical sizing approach
1. Establish your current utilization, not your installed capacity
Pull actual CPU utilization data from your Power7 system over a representative period (including peak, not just average) -- most legacy systems are running well under their rated capacity.
2. Convert to CPW/rPerf, not cores
Work in the performance-rating unit your workload was originally sized in (typically CPW for IBM i), not raw core count, when comparing against a target Power11 or Power10 configuration.
3. Size for your actual growth horizon
Don't size purely for today's utilization -- add realistic headroom for the years you expect the new system to run before its own next refresh.
4. Get a vendor-run sizing study for anything non-trivial
For production enterprise workloads, an IBM or reseller-run capacity planning study using your actual performance data is worth the cost versus a rough estimate.
Why entry and small mid-range Power11/Power10 configurations are usually enough
Because per-core performance has grown so substantially across the Power7-to-Power11 span, the great majority of workloads that ran comfortably on an entry or mid-tier Power7 system (710 through 760) will run comfortably on an entry-tier Power11 or Power10 system with a small fraction of the core count, not an equivalent or larger one. Reserve mid-range and enterprise-tier new hardware for workloads that were themselves running on Power7's largest systems (770, 780, 795) at meaningful utilization.
See the Power generation graph for the underlying cores, memory, and performance data behind every generation from Power7 through Power11, and our migration comparison for how specific Power7 model tiers map to their current-generation equivalents.