A Deep Dive Into That Power S1112 Entry Power11 Server

August 3, 2026 Timothy Prickett Morgan

The entry “Bonnell+” Power S1112 server based on the Power11 processor has been shipping for a little more than a week. While we still have not been able to get our hands on updated Power10 entry server pricing as well as the pricing on the new Power S1112, we can tell you a bunch more about the new machine even though IBM has not yet added the Power S1112 to the entry Power11 Redbook.

The feeds and speeds of the Power S1112 were detailed on July 22 in an IBM i Guided Tour, called Next-Generation Entry Infrastructure for IBM i: AI-Ready Performance for Core and Edge Workloads, which was presented by Fiona Tang, product manager for Power scale-out systems, and Todd Boyd, senior technical staff member and principal product manager for Power. You can see replays of the presentation at this link at COMMON.

To recap, here is the Power11 system lineup:

Most of the Power11 systems rolled out last July with entry, midrange, and high-end boxes spanning the IBM i P10 through P30 software tiers (the P40 and P50 tiers have long since been merged into the P30 tier) and was completed just a week ago with the delivery of the Power S1112, which has a skinny version with four cores that is in the P05 tier and a fatter version (still only ten out of sixteen cores though) in the P10 tier.

We did a lot of the feeds and speeds in our initial take on the Power S1112, but it is good to get the spec sheets from IBM as we have been used to for many decades now. IBM is sharing this data with users through COMMON, which is great, but there used to be a lot more detailed analysis and “compared to what” data in the older presentations that we used to brief business partners.

In any event, here are the primary specs for the Power S1112 in an east to digest page, including images of the doubled sledded rack Power S1112 and the single sledded tower Power S1112:

The thing that jumped out to my eye immediately was that the four PCI-Express slots on the system are not hot pluggable, which means they can be removed and replaced without having to shut down the Power S1112 server. This was not in any of the spec sheets that we saw for the “Bonnell” Power S1012 server launched a few years back, but as it turns out, the Power S1012 also did not have hot plugging for its PCI-Express slots. Which stands to reason because both the Power10 Bonell and Power 11 Bonnell+ machines are based on essentially the same system board.

This may come as a bit of a surprise to prospective Power S1112 buyers, particularly if they had not been shopping for a Power S1012 over the past several years. This may seem odd considering that the Power entry machines in the P05 and P10 class, including the Power S812, the Power S914, and the Power S1014 did have hot swapping on the PCI-Express. Obviously, the processor/memory cards are not hot swappable on any Power Systems iron – where would the operating system run? – but power supplies, cooling fans, and NVM-Express flash storage modules are hot swappable across the Power8, Power9, Power10, and Power11 lineups.

The other interesting thing, if you happen to want to run Red Hat Enterprise Linux on some of the cores on the Power S1112, is that it supports the running of the open source KVM hypervisor atop IBM’s PowerVM hypervisor for Power processors.

Here’s a deeper technical dive into the specs:

I presume that BC means “base cores” in the processor description. The new stuff with the Bonnell+ machine is shown in blue, which just goes to show you how little has changed in the underlying iron.

There are some changes, of course, which you can see in the comparative table below:

First of all, the Power11 processor is running faster and so is its DDR5 main memory to deliver high bandwidth, with a 286 watt maximum power envelope versus the 240 watts of the Power10 processor in the Power S1012, which topped out at eight active cores.

The Power S1112 also uses the same differential DIMM (DDIMM) memory based on IBM’s OpenCAPI Memory Interface, also known as OMI, which uses the same “BlueLink” I/O SerDes to drive memory that are used for other I/O and NUMA links. The Power S1112 has double the memory of the Power S1012, at 512 GB maximum across four slots, twice as much capacity as the Power S1012 had. Moreover, because the DDR5-based DDIMM memory runs at 4 GHz in the Power S1112 server, it delivers 2.5X higher memory bandwidth than the Power S1012 server, which used standard DDR4 DIMM memory with the OMI buffer on the memory planar card. That extra memory bandwidth comes in handy for a lot of workloads, such as AI inference, which is more sensitive to memory bandwidth than it is for memory capacity.

You will also note in this chart that there is no longer a single-core P05 machine option. You get four cores, like it or not. It is not clear that all four cores have to be running IBM i for those who use this platform, but I certainly hope not unless IBM is going to yield some discounts on IBM i licenses for the Bonnell+ machine. The following chart says that you can order a four-core machine and deactivate one, two, or three of the cores if you don’t need all of that IBM i capacity. No matter what, an IBM i logical partition is restricted to a maximum of 64 GB of main memory no matter how much you install in the Power S1112 box.

This chart obviously shows something else that is really important, and it is something that we have been arguing for IBM to do for many years now. And that is to make use of all of that latent processing capacity on the unused cores on the Power processor when those one to four cores are activated for IBM i.

As you can see, now you can activate up to six cores on the Power S1112 server and address up to 448 GB of additional main memory to run the Virtual I/O Server (VIOS), which is a baby AIX subsystem that provides virtual driver support for IBM i, Linux, and AIX operating systems. The important thing is that the NVM-Express flash storage can also be shared across all of the partitions on the Power S1112 box. I can envision having an MCP Server or an AI inference stack running in these shadow partitions, and this is precisely what IBM wants to encourage.

IBM listened.

I would add that I asked for this shadow partition to be able to run a hot backup of the IBM i operating system in an HA configuration. This would not prevent hardware errors from taking down a machine, but it would allow companies to recover from software crashes as well as human error. So, maybe just one special case for a shadow IBM i instance, Big Blue. Think about it, especially considering that most of the system downtime that IBM i shops encounter comes from human error or software bugs, not hardware failures. With memory and flash prices skyrocketing and the cost of a base Power S1112 machine quite possibly being 2X more expensive than an entry Power S1012 configuration from two years ago, something has to give. And this costs IBM nothing and builds goodwill.

This Issue Sponsored By

Table of Contents

Content archive

Recent Posts

Subscribe

Pages

Search