AMD introduces Epyc 6 processors for AI agent workloads

Advanced Micro Devices Inc. unveiled a new line of processors for server use, dubbed the Epyc 6 series, or Venice, featuring 10 CPUs divided into two groups. The first group contains chips with numerous slower cores, while the second group boasts fewer, faster cores. Both sets are tailored to enhance artificial intelligence agents' performance.

AI agents employ graphics cards for neural network computations and rely on CPUs to manage their tools and perform specific tasks within machine learning clusters. According to AMD, Epyc 6 chips can accommodate up to 3.3 times more agents per watt compared to Nvidia Corp.'s Vera CPU, outperforming similar Intel silicon. Employing CPUs with a limited number of cores for an AI agent's tools can lower software costs in certain scenarios.

Database applications, which are charged based on the number of cores utilized, can be more cost-effective when deployed on a higher-frequency core. The Epyc 6 chips' top two models possess a base frequency of 4.05 gigahertz and can reach a peak of 5.3 GHz during intensive tasks. They are equipped with 96 cores and a 384-megabyte cache.

The most compact Epyc 6 chips host 256 cores with a base speed of 2.55 GHz, yielding 512 threads that enable simultaneous processing of numerous AI agents. The series is grounded in Zen 6 core architecture, which boasts a 31% higher IPC than AMD's previous model when handling AI workloads. This performance boost is attributed to the utilization of Taiwan Semiconductor Manufacturing Co.'s latest 2-nanometer node for manufacturing.

Epyc 6 cores come in two variants: the standard Zen 6, optimized for clock speed, and the compact Zen 6c, designed for high-core-count CPUs. Zen 6 features an enhanced out-of-order processing mechanism and AI-specific optimizations such as FP16 data format compatibility and improved VNNI-INT8 support, allowing for reduced hardware footprint and faster AI calculations.

Dan McNamara, senior vice president and general manager of AMD's Compute and Enterprise AI unit, emphasized that AI is broadening the CPU's role across data center layers. With 6th Gen AMD EPYC, customers can standardize on the reliable x86 foundation while selecting workload-optimized solutions to run more agents. Epyc 6 chips are already being deployed by major data center suppliers and power Microsoft Azure instances optimized for AI inference workloads and chip design software.