IBM's z17: New mainframe generation also for AI workloads

IBM is focusing entirely on AI capabilities in its new generation of mainframes. The recently announced z17 system has been equipped with many AI features for the hardware, software and operating system. The basis is its new Telum II processor, in which a second-generation on-chip AI accelerator is integrated. This means a higher frequency, which according to IBM should enable over 450 billion inferences per day. That would be 50 percent more than its predecessor, the z16.

There are plans to add the Spyre accelerator as a PCIe card to the Telum II processor in the fourth quarter. This should provide better support for multi-model AI methods in particular. The Spyre Accelerator was specially developed to bring generative AI functions to the mainframe and make better use of all company-wide data.

New AI operating system

According to IBM, the z17 was designed to be fully integrated into hybrid environments. The system therefore supports many open standards and tools so that everything can be closely connected. In particular, this includes the AI operating system z/OS 3.2, which will be released in the third quarter. It supports hardware-accelerated AI functions, modern data access methods, NoSQL databases and hybrid cloud data processing.

There are also new security features: The functions of IBM subsidiary HashiCorp, which are designed to better manage confidential business data, are now also available on the Z series. These complement IBM Vault, a module that uses identity-based security approaches to authorize all access. There is also Threat Detection for z/OS to detect potentially malicious anomalies that could be the result of a cyberattack.

The new Z Operations Unit also belongs in this category. It brings together key performance indicators and logs from multiple sources in OpenTelemetry format. This should shorten the time it takes to detect anomalies.

Storage for better data performance

IBM also refers to the latest DS8000 storage generation, which provides consistent and optimized data performance for critical workloads, as well as a modular architecture. On the application side, the new z17 is intended to improve the efficiency of developers through the use of AI assistants and AI agents. The watsonx Code Assistant for Z will be used for this purpose.

The system will be generally available on June 18, 2025. There are currently said to be more than 250 AI solutions on IBM mainframes, ranging from credit risk and chatbots to image analysis in healthcare and curbing retail crime.

(kbe)