At the Hot Chips conference, IBM unveiled the first-ever dual-architecture mainframe processor. This processor features 11 cores built on a 2nm process, operating at a base frequency of 5.7 GHz, according to Forbes. It runs both IBM's mainframe instruction set and Arm's instruction set on the same silicon, marking a significant technical achievement for enterprise systems in 2026.
IBM's mainframes have historically been defined by their proprietary Z architecture. However, the new processor natively integrates Arm's instruction set, a strategic embrace of a broader ecosystem. This directly challenges the traditional proprietary nature of mainframe systems.
Companies will likely see a new era of hybrid enterprise systems. These offer unparalleled flexibility and performance for demanding AI and data workloads. Mainframe modernization could accelerate, repositioning them as a premier platform for secure, high-performance hybrid cloud workloads.
Dual ISA and AI: IBM's Technical Advance
- IBM's next-generation AI processor is the first to natively support dual ISA execution within the same core, executing either z/Architecture or AArch64 instructions and switching between them dynamically, according to Tom's Hardware.
- The processor, presumably for z18 mainframes, features 11 cores built on a 2nm process, operating at a base frequency of 5.7 GHz, Tom's Hardware reports.
- IBM's next-gen AI accelerator includes 16 cores with optimizations for newer AI data formats like FP4/MXFP4 and 96 GB of HBM3e memory, offering up to 4TB/s bandwidth, according to Tom's Hardware.
Tom's Hardware describes an 11-core dual-architecture processor and a 16-core 'next-gen AI accelerator'. The 11-core count refers to the main CPU, while the 16-core count applies to a distinct, integrated AI acceleration component within the system. This clarifies the overall architecture, indicating a specialized division of labor for optimal performance.
