At Hot Chips 2024, IBM announced its next-generation mainframe technology, featuring the Telum II processor and Spyre AI accelerator. Both products will significantly improve AI capabilities on IBM Z systems, addressing the growing demand for efficient, secure, and scalable enterprise AI solutions.
The Telum II processor, set to power future IBM Z and LinuxONE platforms, boasts impressive specs. It features eight high-performance cores running at 5.5GHz, with a 40% increase in on-chip cache capacity. The processor includes an integrated AI accelerator, delivering four times the compute capacity of its predecessor for low-latency, high-throughput AI inferencing.
A key addition to Telum II is the new Data Processing Unit (DPU) for I/O acceleration. This on-chip feature is designed to improve data handling efficiency, supporting a 50% increase in I/O density. The DPU aims to simplify system operations and boost performance for data-intensive applications and AI workloads.
Complementing Telum II is the Spyre accelerator, IBM's answer to the growing demand for large language model (LLM) capabilities in mainframe environments. Each Spyre chip houses 32 compute cores (supporting int4, int8, fp8, and fp16 datatypes) for AI applications. The accelerator is designed to work in tandem with Telum II, enabling what IBM calls "ensemble methods of AI" - combining traditional machine learning models with LLMs for potentially more accurate and robust results.
For IBM’s mainframe customers, this means being able to run AI-powered applications securely and efficiently, directly on the Z platform. Telum II can handle large-scale AI workloads and data-intensive business needs, while the Spyre accelerator handles complex AI models and generative AI use cases—all without compromising the security and reliability of the IBM Z environment.
Both the Telum II processor and Spyre accelerator will be manufactured using Samsung Foundry's 5nm process technology. IBM expects these products to be available to clients in 2025.