Last IBM announced a new entry-level server to its S-line first introduced in July 2022. The S1012’s features include:
- Either rack mounted or tower configuration (like the S1014);
- 2U height when rack mounted (like the S1022s and S1022), but only half-wide;
- A single-socket machine (like the S1014), with a maximum of 8;
- A maximum memory of 256GB – much smaller than the 1TB maximum of the S1014;
- 4 NMVe slots giving a maximum of 6.4TB internal data storage and
- AI acceleration built-in (as with all power10-based servers).
As with the rest of the range, users can run IBM i (7.4 or 7.5), AIX or Linux. These are the only servers on the market to include on-chip AI acceleration, while almost all the others are based on Microsoft Azure or Windows and GPUs from Nvidia. As a RISC-based server supplier Oracle’s Sparc could follow IBM’s approach, although its generative AI approach is more cloud-based and its Sparc’s architecture – arguably less open. I believe most users will build their language models for generative AI in their own data centers, avoiding the need to trust others to secure their data (see my Figure below). With this machine you can process your private data at the edge as well.
The main use case for the Power S1012 is to run transactional AI at the edge close to where data is being generated outside of a main data center. IBM is targeting Small and Medium Businesses (SMBs) and/or remote offices with the new machine: typically in Retail (appropriate for the tower/desk-side configurations), Manufacturing, Logistics or Retail industry sectors.
IBM’s reference customers include Equitus, which is working to deploy the new servers for federal intelligence and military applications in the field. Processing data and reacting to it at the edge side-steps the delays, security risks and potential network failures of shifting data back and forward to the center, which may be in another country or continent.
Of course users can either buy or lease the machine; they can also purchase PowerVS services from IBM, although I expect IBM will be delivering these from its larger Power servers.
As the lowest model in the range IBM doesn’t need to offer its Dynamic Capacity on Demand options offered on its larger models from the S1022s upwards (see my Figure above for product specs). However it does come with the other extensive technical features, such as the transparent memory encryption, enhanced isolation and trusted boot to prevent side-channel cyber attacks alongside its unique integrated AI acceleration on each processor core.
While new machines can also be used as a data backup for cloud-based applications, as High Availability (HA) servers and/or for application modernization, for me its most important innovation is in bringing inferencing, analytics and Machine Learning (ML) to the edge, where it will help stretch the use of generative AI far beyond today’s successful and growing centralized, cloud-based approaches (such as ChatGPT).