IBM Unveils 5nm Telum II Processor for Advanced AI Applications

IBM has recently announced the development of its cutting-edge Telum II Processor and Spyre Accelerator, designed for large language models and generative AI applications. These innovative chips are built on Samsung's advanced 5nm technology, showcasing IBM's commitment to pushing the boundaries of AI hardware.

The Telum II Processor features a scalable I/O sub-system aimed at reducing energy consumption and data center footprint. Building upon the success of the original Telum AI inference chip released in 2021, the Telum II boasts eight cores capable of running at speeds of up to 5.5GHz. Each core is equipped with a 36MB L2 cache, with a significant 40% increase in on-chip cache capacity, totaling 360MB. Additionally, the processor includes a virtual level-4 cache of 2.88GB per processor drawer, marking a 40% improvement over its predecessor.

Moreover, the Telum II integrates an AI accelerator that enables low-latency AI inferencing at 24TOPS, enhancing applications such as fraud detection in financial transactions. This represents a fourfold increase in compute capacity per chip compared to the previous generation. The processor also features an integrated AI accelerator core and a coherently attached Data Processing Unit (DPU) for efficient I/O operations.

The Spyre Accelerator, developed in collaboration with IBM Research, complements the Telum II Processor by providing enhanced AI model workloads across IBM Z mainframes. Connected via a 75-watt PCIe adapter, the Spyre chip can access up to 1TB of memory and is designed to operate within a power limit of 75W per card. With 32 compute cores supporting various datatypes, including int4, int8, fp8, and fp16, the Spyre Accelerator caters to both low-latency and high-throughput AI applications.

As the industry witnesses a shift from proof-of-concept to production for generative AI projects leveraging Large Language Models (LLMs), the demand for power-efficient, secure, and scalable solutions has become paramount. Tina Tarquinio, VP of Product Management for IBM Z and LinuxONE, emphasized the strategic importance of these advancements, stating, "Our robust, multi-generation roadmap positions us to remain ahead of the curve on technology trends, including escalating demands of AI. The Telum II Processor and Spyre Accelerator are designed to deliver high-performance, secured, and more power-efficient enterprise computing solutions."

Both the Telum II Processor and Spyre Accelerator will be manufactured by Samsung Foundry on a 5nm process, with systems utilizing these chips expected to be available in 2025. IBM's relentless pursuit of innovation in AI hardware underscores its commitment to providing state-of-the-art solutions for businesses seeking to leverage AI technologies at scale.