Stacking Up Entry IBM i-Power11 Systems Against Windows X86 Platforms

December 1, 2025 Timothy Prickett Morgan

Having spent the past several weeks going over the hardware and software costs and performance of systems in the IBM i P10 and P20 software tiers. We had an admittedly small sample, but one that we could get Big Blue’s pricing information for, which is hard to come by. This week we wanted to see how those entry P10-class and P20-class machines stack up against modern X86 servers based on Intel’s latest Xeon 6 systems running a Windows Server stack that was functionally equivalent to the IBM i stack.

So we did a little spreadsheet work on your behalf.

Our comparisons include base hardware systems with 128 GB of main memory and two 1.6 TB NVM-Express flash cards, plus rudimentary 25 Gb/sec Ethernet networking to link to the outside world. We put redundant, energy-efficient power supplies in the box, and then loaded them up with an operating system, a database, and a compiler stack as well as any freebies, and costed it all out with three years of software licenses. The IBM i machines are subscription only for three years and the Windows Server machines have perpetual licenses, end user client access licenses, and Software Assurance tech support for three years.

We wanted to get as much of a like-for-like comparison as we can, and decided that the so-called Performance variations of the Xeon 6 processor – the ones called “Granite Rapids” that use full-bore cores and not ones crimped so more of them could be crammed into a socket for loads that don’t need complex math or lots of cache – were the most like the Power10 and Power11 processors in the current IBM Power Systems line.

It is hard to get a Xeon 6 processor to scale down to the eight cores we used in our comparisons of Power10-based Power S1014, Power S1022s, and Power S1024 machines and Power11-based Power S1122 and Power S1124 machines. But there are a few options that have as few as eight or sixteen cores, and we went to the Dell site and ginned up the configurations using them.

To come up with relevant performance metrics, we grabbed some performance information that IBM supplied way back in September 2022 for its comparisons between Power10 entry and midrange platforms and chose the SAP Sales & Distribution benchmark, IBM’s Commercial Performance Workload (CPW) and Relative Performance (rPerf) benchmarks for Power Systems, and relative performance metrics for the Xeon line from Intel developed by us over at The Next Platform to come up with SAP SAPS and CPW throughput rates for the machines shown. We have added in the Power E1050 and Dell PowerEdge R840 results that IBM gave out here when pitting Power10 against the “Cascade Lake” Xeon SP and “Ice Lake” Xeon 4 processors and then filled in the blanks, which we show in bold red italics in the table below:

Click to enlarge

In the PowerEdge machines that we ginned up, some had 16 cores and some had eight, but we are only pricing the hardware with eight cores activate (which are just showing the base configurations). Taking out half the cores does not change the price that much in the X86 world, where cores are cheap but software is not. The PowerEdge R470 is a 1U rack server with a single socket and the PowerEdge R770 is a 2U server with two sockets, and these machines are half the height of the 2U Power S1122 and the 4U Power S1124. IBM is not trying for maximum compute density, but offering expandability and therefore longevity for its Power11 (and prior) platforms.

In the table, we show the cost per SAP SAPS, the cost per CPW, and the cost per user. The smaller configurations have 10 users and the fatter machines have 250 users.

The Power11 machines shown in the table have IBM i and its integrated database plus the compiler tools priced out, and the Windows machines have Windows Server 2024 Datacenter Edition, SQL Server 2022 Standard Edition, and Visual Studio Enterprise on them. Software Assurance costs 25 percent of the perpetual licensing fee for each year you want support. And this factor, more than any other, puts IBM i within spitting distance of the Windows Server stack running atop fairly small Xeon 6 servers.

To be specific, on the Power S1122, the cost per user is 18.4 percent larger over three years than for the PowerEdge R470, and the Power S1124 costs 25 percent more than the PowerEdge R770 server. If you want to look at throughput comparisons, the Power S1124 and the PowerEdge R770 have the same cost per CPW, and the PowerEdge R470 actually costs 34 percent or 42 percent more per unit of CPW work than the Power S1122. Ands the

How about that!

We did not expect that when we sat down to do this math.

Merry Christmas, go get a Power11 upgrade now.

This Issue Sponsored By

Table of Contents

Content archive

Recent Posts

Subscribe

Pages

Search