IBM z17 Mainframe's Telum II Processor Unveiled at Hot Chips

On August 26, 2024, at the prestigious Hot Chips 2024 conference in Palo Alto, IBM officially unveiled the Telum II processor, powering its next-generation z17 mainframe. Far from a routine hardware update, this marked a significant leap forward—one that redefined what mainframes could offer in the era of artificial intelligence and hybrid workloads. The Telum II chip is an architectural breakthrough. Built on a 5nm process by Samsung, it features eight high-performance 5.5 GHz cores and a massive 360MB of on-chip cache, ensuring blazing-fast single-threaded performance. For I/O-heavy workloads, IBM added a coherent Data Processing Unit (DPU) that accelerates data movement between compute and memory. But the true highlight was the enhanced on-chip AI accelerator, capable of 24 TOPS (trillions of operations per second)—quadrupling the AI throughput of the previous Telum chip. This AI muscle enables real-time inferencing within enterprise transactions. IBM demonstrated scenarios such as fraud detection happening as a credit card transaction is authorized, without the need to offload to an external AI engine. Complementing this is the new Spyre Accelerator, a PCIe-based AI module introduced alongside Telum II, which allows the z17 system to run complex ensemble models, making the platform ideal for hybrid and generative AI tasks. The moment wasn’t just about raw power—it was about IBM reaffirming the role of the mainframe as a secure, always-on foundation for mission-critical AI. Designed for sectors like banking, healthcare, and government, the z17 with Telum II combines extreme reliability (99.999999% uptime), robust security, and AI-native capabilities into a single enterprise platform. In a world increasingly dominated by cloud-native buzzwords, IBM’s Hot Chips 2024 unveiling was a reminder: the mainframe isn’t just surviving—it’s evolving into the brain of the AI-powered enterprise.