IBM and Arm have unveiled a processor design intended to bring Arm-native Linux software directly onto future IBM Z and LinuxONE systems.
Well, this is different. At the Hot Chips conference, IBM and Arm revealed they’re working on an Arm chip in which each core can natively execute IBM Z or LinuxONE instructions and ARM instructions. IBM says the approach will allow Arm-native Linux environments to operate alongside z/OS and existing Linux on IBM Z mainframes.
Arm chips on a mainframe? Who would have thought it?
The proposed processor family is the first disclosed hardware milestone from IBM and Arm’s recent agreement to work together. IBM is positioning the design as a way to combine its high-reliability transaction-processing systems with the much larger Arm-targeted Linux, cloud and AI software ecosystems.
IBM and Arm’s future chip is a 2nm, 11-core mainframe processor that runs above 5.7 GHz and is designed to execute both Arm and IBM instruction sets natively on every core. However, we only have these bare facts. IBM hasn’t yet disclosed availability, power, cache capacity, memory bandwidth, or an exact microarchitecture.
The design also includes AI inference accelerators aimed at in-transaction fraud detection, a dedicated on-chip data processing unit for I/O acceleration, and a large cache hierarchy for enterprise workloads.
These chips will enable Arm and IBM workloads to run concurrently on the same processor cores. That is a big distinction from heterogeneous systems that use separate accelerators, Arm, or general-purpose CPU complexes. Instead of just pasting Arm chips into a mainframe, the two tech giants will do the hard work of creating a new chip combining both architectures. The new design will make both instruction-set environments native capabilities of every core.
According to Christian Jacobi, IBM Systems Development CTO, in a VentureBeat interview, “On this chip are 11 cores, and each core can dynamically switch back and forth between Arm software mode and traditional Z software mode. That enables us to run the mission-critical enterprise software right next to, on the same chip, the much broader software ecosystem of Arm applications.”
The chip will flip between the two software modes using Linux’s built-in KVM virtual machine manager. You might think this would cause slower functionality, but Jacobi claims that’s not the case. The “switch takes about the nanosecond scale. Because you’re running for many milliseconds in the virtual image, this switching overhead sort of amortizes to zero — pretty much no impact at all.”
For mainframe customers, the selling point for this new generation of chips is access to Arm-native applications without isolating those workloads on a separate server estate. The company said Arm workloads will inherit platform capabilities including hardware-level fault detection and recovery, encryption, secure key management, I/O acceleration and AI acceleration. Put it all together, and we should see not a mainframe AI inference engine, but an Arm and AI-enabled mainframe for its traditional business and financial transactions workloads.
You won’t be buying this next-generation mainframe anytime soon. IBM hasn’t announced a system availability date, product name, pricing, or a manufacturing partner.
IBM described the chip as a future IBM Z and LinuxONE processor, rather than an immediate successor product. IBM’s current z17 generation shipped only in 2025, and IBM usually releases a new generation of chips every three years. I have to think, however, that such a radical new chip will take longer than that to appear. My best guess is you’ll see mainframes with these new CPUs released in 2030.
