IBM Power server targets AI workloads at the edge

IBM has filled out the low end of its Power server portfolio with a 2U rack-mounted server designed for running AI inferencing workloads in remote office or edge locations outside of corporate data centers.

The 1-socket, half-wide, Power10 processor-based system promises a threefold performance increase per core compared to the Power S812 is basically replaces, IBM stated. Running IBM AIX, IBM i, Linux, or VIOS operating systems, the S1012 supports in-core AI inferencing and machine learning with a Matrix Math Accelerator (MMA) feature.

MMA is a feature of Power 10-based servers that handles matrix multiplication operations in hardware, rather than relying solely on software routines; it offers four-times better performance per core for matrix multiplication kernels at the same frequency, according to IBM. Each Power S1012 includes four MMAs per core to support AI inferencing.

Also featured in the new servers is 256 GB of system memory distributed across four DDR4 Industry Standard Dual In-Line Memory Modules (ISDIMM) slots.

“By deploying Power S1012 at the edge, clients can run AI inferencing at the point of data, thus eliminating data transfers,” wrote Steve Sibley, vice president of IBM’s Power product management group, in a blog about the new servers. 

Sibley cited analysis from Gartner related to AI workloads at the edge: As more organizations embrace AI to further drive business value, Gartner finds that clients in industries such as retail, manufacturing, healthcare, and transportation are deploying workloads at the edge to capitalize on data where it originates.

“By placing data, data management capabilities and analytic workloads at optimal points, ranging all the way to endpoint devices, enterprises can enable more real-time use cases. In addition, the flexibility to move data management workloads up and down the continuum from centralized data centers or from the cloud-to-edge devices will enable greater optimization of resources,” Gartner notes in its March 2024 Market Guide for Edge Computing.

Securing those workloads is also a key feature of the S1012. “To ensure insights remain a competitive advantage and don’t fall into the wrong hands, transparent memory encryption with Power10 secures data in and out of AI models running locally addressing data leaks,” Sibley stated.

The servers feature advanced remote management capabilities to let organizations efficiently manage and monitor their IT environments remotely, to enhance responsiveness and minimize downtime, Sibley stated. “High-availability features such as redundant hardware and failover mechanisms can help ensure continuous operation, all within a compact physical footprint,” Sibley stated.

The S1012 is aimed at small-to-medium users and offers a lower entry point for customers to get into the IBM Power lineup that includes the high-end 240 core, 64Tb Power E1080, and the E1050, which is aimed at memory-intensive workloads and includes up to 48 cores and 16TB memory. 

The IBM Power S1012 will be generally available in a 2U and a tower model from IBM and certified business partners on June 14, 2024.