After a few clues this summer, IBM officially unveiled its latest generation mainframe with the z16, which includes a Telum chip with an AI accelerator. This size is suitable for performing transactions and processing orders in real time, such as detecting fraud. In addition, IBM wants to strengthen the security of critical applications around the quantum threat.
With a fairly good fourth quarter of 2021 in terms of the mainframe and storage portion, IBM has just revealed the latest version of its Z16 systems. The z16 is more than just an update to the venerable mainframe, “This latest release is a game-changer,” he asserts. Rick Lewis, Senior Vice President, IBM Systems. Compared to its predecessors, the Z range changes the processor based on the Telum chip. So the z16 brings the first on-chip AI accelerator. “Through this, companies are moving to address artificial intelligence in real time,” Elpida Tzortzatos, CTO for artificial intelligence at IBM Z, said at a press conference. According to Ross Morey, general manager of IBM Z, a whole range of new possibilities with AI will open up on the z16 system that could not be considered before.
Few technical elements were introduced during this meeting, and we have to go back to the summer of 2021 to learn more about this popular Telum chip. It is etched with a size of 7 nm, manufactured on Samsung production lines and includes 22.5 billion transistors on an area of 530 mm². The processor consists of 17 layers with a base clock of just over 5 GHz. It has 8 cores that can each run 2 threads in SMT2, built-in workload accelerators, and 32MB of what IBM calls a semi-private level 2 cache. The L2 cache is said to be semi-private because it is used to build a 256MB shared virtual L3 connection between Chip cores. Z16 is also offered as an open platform that is suitable for a mixed cloud environment. In this regard, IBM mentions the recent announcement of two cloud services, IBM Z and Cloud Modernization Stack, and Wazi as-a-service, to help customers modernize their centralized applications.
Performance on AI for fraud detection and quantum-resistant security
The entire design has been thought for real-time application processing, thus, the focus is on single-thread performance. Elpida Tzortzatos gave some figures on these renderings, “The z16 is capable of processing up to 300 billion AI inferences per day with millisecond latency.” Besides the technical aspect, CTO prefers to focus on the uses of the latest mainframe computers. We have conducted a study on the effect of z16 in terms of fraud detection. It shows that applying advanced inference models to all bank and payment card transactions can reduce fraud losses by $161 billion globally. Likewise, the Telum chip’s contribution can reduce the false positives that penalize bank transactions, “using machine learning models, the rate of false positives can be reduced by 36%,” the manager adds. Previously, AI processing could migrate to other clouds (with Model9) or be provided by zIIP chips used to migrate workloads, notably Java (many z-mainframe implementations are developed in Java). IBM does not charge software fees for zIIP capacity unless additional processing capacity is used. On the other hand, Model9 installs an agent on z systems to export data to the public cloud and perform analytics and AI, or it can make use of zIIP chips for analytics.
The other part of z16 is geared towards security and especially towards future threats, in this case those related to quantum computing that can have an impact on the strength of cryptographic algorithms. In this context, the latest generation of mainframe computers incorporate cryptographic techniques based on Euclidean networks. On the other hand, IBM has not touched upon another technology that is at its forefront, symmetric cipher. This method aims to solve the problem of the fact that the data is not encrypted all the time, especially when the application has to use it. Note that Big Blue is upgrading the Crypto Express crypto card to the 8S version (7S on the z15). Finally, to be complete, the z16 will be available from May 31, 2022.