IBM revamps the venerable mainframe to run generative AI workloads and agents

IBM Corp. is bringing the mainframe into the modern age with the latest version of its venerable hardware powered by a new processor that supports generative artificial intelligence and AI agent workloads.

The company lifted the lid on the new IBM z17 mainframe today, saying it’s powered by its most advanced Telum II processor, which gives it capabilities that exceed standard transactional AI.

Slated to launch in June, the z17 boasts enormous processing power and can evaluate millions of transactions as they occur in real time, enabling financial services providers to create newer and more efficient applications to counter fraud and perform analytics. The IBM z17 also accelerates AI inference by 50% over its predecessor, the IBM z16.

Other capabilities include multimodel support, including IBM’s family of Granite large language models. There are also new data protection features, which were introduced based on feedback from more than 100 mainframe customers, IBM said. All told, it supports more than 250 different AI applications, including mitigating loan risk, managing chatbots, analyzing medical imagery, preventing retail crime and more besides.

Enduring popularity

The venerable mainframe computer is a specialized, on-premises server that’s designed to handle certain kinds of applications, such as data analytics, enterprise resource planning and large transaction processing, with the utmost reliability.

While most organizations have moved at least some of their information technology operations to the cloud these days, mainframes remain popular thanks to advances in their processing power and, more importantly, their high degree of availability. This is why many companies still consider them to be essential for applications where it’s critical that downtime is avoided.

According to IBM, around two-thirds of the Fortune 500 and 45 of the world’s 50 biggest banks are still using mainframes, along with eight of the top 10 insurance firms, seven of the top 10 global retailers and eight of the top 10 telecommunications firms.

IBM manufactures the vast majority of mainframes in use around the world today, though some are also made by the Japanese firm Fujitsu Ltd.

Engineered for AI

The continued widespread adoption of mainframes explains why the launch of IBM z17 is considered to be a very big deal, and it’s why IBM went to great lengths to enhance its AI capabilities.

The company increased anticipation ahead of the launch when it announced the Telum II chip (pictured below) last August, saying it delivers a 40% improvement in cache performance thanks to the inclusion of an I/O Acceleration data processing unit, plus an integrated AI accelerator, known as the IBM Spyre.

Telum II is based on eight high-performance cores that run at 5.5 gigahertz, with 36 megabytes of memory per core. That amounts to an increase of 40% in on-chip cache capacity, for a total of 360 megabytes.

The Spyre AI accelerator is built to handle low-latency and high-throughput in-transaction AI inference operations, which makes the z17 more suitable for applications such as real-time fraud detection in financial transactions. It also paves the way for generative AI to run on mainframes that utilize retrieval-augmented generation techniques to tap into the data held within.

In addition, the chip allows the IBM z17 mainframe to support AI agents such as the IBM watsonx Code Assistant for Z and IBM watsonx Assistant for Z. Customers will be able to use the AI Toolkit for IBM Z and LinuxOne to create their own AI agents, too.

IBM Z and LinuxONE General Manager Ross Mauri said it has become clear to most companies that AI systems are only as good as the infrastructure they run on, hence the need for such powerful processing capabilities.

“With z17, we’re bringing AI to the core of the enterprise with the software, processing power, and storage to make AI operational quickly,” he insisted. “Only 1% of the enterprise data is part of large language models. With z17, businesses can put their untapped enterprise data to work with AI in a secured, cost-effective way.”

Enhanced software

Other components include a specialized operating system for AI called z/OS 3.2, which will be released in the third quarter, after the z17 mainframe is launched. The software has been revamped to power hardware-accelerated AI across the entire system and provide AI-powered operational insights to streamline system management. Moreover, it supports simple data access methods, NoSQL databases and hybrid cloud processing, enabling z mainframes to work in concert with cloud servers and access even more data.

Then there’s IBM Z Operations Unite, which is a new tool that allows operational data to be combined in the OpenTelemetry format, simplifying information technology management operations. In more general terms, it will help teams to detect anomalies faster, reduce alert investigation durations by up to nine-times and speed up time to resolution, IBM said, while isolating the impact of any problems.

On the security side, new features include a secrets management tool from Hashicorp, which IBM acquired for $6.4 billion earlier this year. It’s called IBM Vault, and it relies on identity-based security to authenticate and authorize access to secrets, certificates, security keys and other sensitive information, in both the mainframe and also hybrid cloud environments. In this way, it offers companies a single solution to securely access and manage all of their most sensitive assets.

As always, IBM is offering comprehensive support to enterprises to help them manage their mainframe operations. These services, delivered by IBM Technology Lifecycle Services, can help customers to maintain peak performance and minimize risk and disruptions for mission-critical applications. Customers will also get access to IBM Agent Assist, an automated customer service bot that can help them to streamline incident resolutions and get their mainframes back to work faster.

Image: SiliconANGLE/Meta AI