Stacking Up Power11 Entry Server Performance To Older Iron

October 6, 2025 Timothy Prickett Morgan

Here we are heading into the fall and if you have budget set aside for a new system or an upgrade, now is the time to start thinking about spending it. For customers buying mainstream P10-class and P20-class systems, the Power11 machines that have been out for a few months now and are the machines you will need to consider. And if you have older Power8 or Power9 iron, you might even be considering an upgrade to a Power10.

With all of this in mind, we have put together the mother of all performance tables to compare the current and prior three generations of entry servers – what IBM sometimes call the “scale out” servers, meaning that customers buy lots of them and run distributed computing workloads on them instead of buying a big iron, huge NUMA memory “scale up” systems. Most IBM i shops do not write distributed applications that run across clusters of single-socket or two-socket servers, as is the norm with the hyperscalers and cloud builders. They have one box running production workloads and maybe one out of five times have a second machine on site as a hot backup or they dump copies of their applications and databases out to a Power Systems machine running in the cloud.

This performance comparison table spans Power8, Power9, Power10, and Power11 machinery, and is missing the Power S1112 machine in the P05 software tier that we expect early in 2026. (It is a bit of a mystery why IBM didn’t put it into the field with the other machines.) The data is based on the IBM Power Performance Report, which we just got our hands on even though it was technically out there for a while. The reason we did not see it is because for the longest time, IBM called its generational performance report the IBM Power Systems Performance Capabilities Reference (that link is to the Power8 machines and IBM i 7.3 launched in 2017) and we kept looking for an updated document of that title. Anyway, you can see it here.

The latest performance report has updated performance metrics that take into account firmware patches for the Spectre and Meltdown speculative execution mitigations, which burn a few points of performance when they run on newer iron to keep checking that no malicious code is exploiting these vulnerabilities.

We have filled in the gaps in the report to give you the performance of each system with its maximum number of cores based on the processor features available on each make and model in the entry systems lineup for the past four generations. It was a pain in the neck to create this table, and to make estimates for full system performance at those times when IBM did not test a full system. You can’t just take the performance of one core and apply it across a two-socket machine. In some cases, IBM’s report inexplicably shows the same performance on a bunch of systems using processors with different clock speeds, which is not possible. (Or, rather, it should not be the case.) We marked one column in gray so it pops out to your eyes. Take a look at the table below:

Click to enlarge this table.

In the case of the Power S1022 above, IBM is testing the performance of four-core partitions on the Power S1022 and shows them all delivering 106,300 CPWs across four cores even though their base clock speeds for the features shown are running at 2.45 GHz, 2.9 GHz, and 3.1 GHz.

In the many cases where we have to estimate the performance of the system with the full core counts, they are shown in bold red italics.

The obvious thing to realize is that the Power S1122 and Power S1124 entry systems have a pretty substantial performance increase given the modest increase in the clock speeds of the Power11 processors compared to the Power10 chips, but more importantly, sport a substantial increase in the core counts – roughly 25 percent more cores, like for like models. This is entirely due to an increase in yield from the 7 nanometer processes that IBM uses from its foundry partner, Samsung.

The other thing to note is that the Power10 machines, which are still available, are no slouches. And in terms of performance, you can probably get a good deal on these as well. We are working on price/performance comparisons with the Power11 machines compared to the prior three generations, which is tough because pricing information for these machines is hard to come by. But we are working on it. Stay tuned. This performance table is a good place to start as you consider your options.

This Issue Sponsored By

Table of Contents

Content archive

Recent Posts

Subscribe

Pages

Search