At the Hot Chips 2024 conference, IBM unveiled its latest innovation in mainframe technology—the Telum II processor and the Spyre Accelerator, designed to revolutionize the performance of IBM Z and LinuxONE mainframes. These new technologies are not only aimed at enhancing artificial intelligence (AI) workloads but are also expected to improve performance in various high-demand transactional applications, including database management, financial processing, and fraud detection. By integrating advanced AI capabilities directly into the heart of its mainframe systems, IBM is pushing the boundaries of what’s possible in enterprise computing, offering enhanced speed, resiliency, and scalability.
The release of the Telum II processor and the Spyre Accelerator marks a significant leap forward in the AI-driven capabilities of IBM’s mainframe systems, providing enterprises with the tools they need to stay competitive in an increasingly AI-focused world. As workloads grow in complexity and size, IBM’s latest innovations will serve as crucial enablers of real-time, data-intensive applications, while addressing the security and scalability concerns that often accompany such technological advancements.
The Telum II Processor: Redefining AI Integration in Mainframes
The Telum II processor represents a major leap in both processing power and AI acceleration capabilities for IBM’s next-generation mainframes. Developed using Samsung’s 5nm technology, Telum II boasts eight high-performance cores running at 5.5 GHz, and offers significant memory and cache enhancements over its predecessor. The processor also introduces a 40% increase in on-chip cache capacity, with virtual L3 and L4 caches expanded to 360MB and 2.88GB, respectively. These improvements allow for better data throughput and real-time processing, which are critical for the increasingly complex and large-scale applications being run on mainframes today.
One of the most important features of the Telum II processor is its AI acceleration capabilities, which have been significantly enhanced from the first-generation Telum chip introduced in 2021. The new AI accelerator on the Telum II chip delivers a fourfold increase in compute power, reaching an impressive 24 trillion operations per second (TOPS). However, as IBM’s Christian Jacobi and Elpida Tzortzatos emphasized, “TOPS alone don’t tell the whole story.” The true power of the Telum II processor lies in its architecture, which is specifically designed for enterprise AI workloads. This means the processor can run AI models side-by-side with traditional enterprise workloads, delivering high-throughput and low-latency inferencing for applications like fraud detection, financial settlements, and credit rating analysis.
In terms of scalability, future IBM Z systems can be equipped with up to 32 Telum II processors and 12 IO cages, each supporting 16 PCIe slots. This architecture allows for massive bandwidth, redundancy, and multi-pathing, ensuring that enterprise clients can run their most demanding AI and transactional workloads simultaneously without performance bottlenecks. The custom I/O protocols incorporated into the design further enhance availability and error checking, crucial for maintaining the “eight nines” (99.999999%) availability that IBM Z mainframes are known for.
Spyre Accelerator: Boosting AI and Machine Learning Workloads
Complementing the Telum II processor is the Spyre Accelerator, which is designed to provide additional AI compute capabilities. The Spyre Accelerator contains 1TB of memory and 32 AI accelerator cores, which share a similar architecture to the AI accelerator integrated into the Telum II processor. This makes the Spyre Accelerator a powerful addition to IBM’s AI ecosystem, particularly for enterprises looking to scale their machine learning and generative AI workloads.
One of the most exciting features of the Spyre Accelerator is its ability to cluster multiple accelerators together within the I/O subsystem of IBM Z, connected via PCIe. This clustering capability allows enterprises to scale their AI workloads dynamically, depending on the demands of the application. Whether it’s processing large language models or handling real-time data analysis, the Spyre Accelerator can offer the additional compute power needed to meet the challenge.
For AI-centric applications, such as fraud detection or compliance monitoring, the ability to scale AI processing within the mainframe environment offers clear operational advantages. According to Patrick Moorhead, founder and chief analyst at Moor Insights & Strategy, moving AI workloads from external GPU servers back onto the mainframe itself results in improved efficiency, speed, and security. In previous AI deployments, data had to be moved off the mainframe for analysis and then brought back, introducing potential delays and security vulnerabilities. With the Spyre Accelerator, enterprises can now run these critical AI tasks directly on the mainframe, ensuring faster processing times and greater data security.
AI Workloads Beyond the Scope of AI: The Flexibility of IBM Mainframes
While the Telum II processor and Spyre Accelerator are clearly optimized for AI workloads, their applications extend far beyond AI itself. As Tina Tarquinio, vice president of product management for IBM Z and LinuxONE, explained, the use cases for IBM’s next-generation mainframes span transaction-heavy applications such as database management, financial settlements, and document processing. By offloading non-AI tasks, such as transactional workloads, from the main CPU to the Spyre Accelerator, enterprises can improve the efficiency of both their AI and traditional workloads.
For example, in a hybrid cloud or distributed cloud environment, the combination of Telum II and Spyre allows enterprises to run mission-critical applications with greater efficiency. Steven Dickens, chief technology advisor at The Futurum Group, pointed out that IBM Z systems equipped with these new technologies would be “I/O beasts,” capable of handling big Oracle or MongoDB deployments more efficiently than previous mainframe models. The improved I/O bandwidth and processing power mean that enterprises can run more applications in parallel, reducing latency and increasing overall system throughput.
Moreover, the quantum-safe security features of the next-generation IBM Z systems make them particularly appealing for industries that require high levels of security and resiliency. With eight nines of availability and integrated AI-driven compliance monitoring, these systems are well-suited for sectors like banking, insurance, and healthcare, where security and uptime are non-negotiable.
The Future of Enterprise Computing: AI and Security Converge
IBM’s advancements with the Telum II processor and Spyre Accelerator underscore the company’s commitment to combining AI innovation with the security and reliability that mainframes have long been known for. In a world where enterprises are increasingly relying on AI to handle mission-critical tasks, the need for secure, efficient, and scalable AI solutions has never been greater. IBM’s new mainframe technologies address this need by allowing enterprises to keep their AI workloads within the secure confines of the mainframe, rather than moving sensitive data to less secure environments.
Christian Jacobi highlighted this point when discussing how AI can be applied directly to enterprise code within the mainframe. “Many clients run hundreds of millions of lines of code in their applications, and they are very security-sensitive about the code base,” Jacobi explained. With the Telum II processor and Spyre Accelerator, enterprises can now run AI workloads directly on the mainframe, ensuring that sensitive data and business processes remain secure while benefiting from the performance and scalability of AI-driven systems.
Additionally, IBM’s ability to cluster multiple Spyre Accelerator cards together means that enterprises can scale their AI workloads as needed, without sacrificing security or performance. This flexibility makes IBM Z systems a compelling option for enterprises looking to invest in AI-driven infrastructure that can grow alongside their business.
Conclusion
The introduction of the Telum II processor and Spyre Accelerator marks a significant milestone in the evolution of IBM’s mainframe technology. These innovations bring cutting-edge AI capabilities to the IBM Z and LinuxONE platforms, enabling enterprises to tackle increasingly complex workloads with unprecedented speed, efficiency, and security. By integrating advanced AI processing directly into the mainframe environment, IBM is helping its clients unlock new levels of productivity, scalability, and resiliency, while addressing the growing challenges of data security and compliance.
As we look ahead to the next generation of enterprise computing, it is clear that AI will play a central role in shaping the future of mission-critical applications. IBM’s commitment to delivering quantum-safe, AI-powered mainframes positions it at the forefront of this transformation, offering enterprises the tools they need to succeed in an increasingly AI-driven world.