The Pulse
AMD makes 6th-gen EPYC server processors generally available
AMD's 6th-generation EPYC server processors are generally available, with systems shipping from server vendors. AMD outlined cloud availability, performance comparisons and rack-scale options for AI workloads.
AI.info Team · Written with AI · fact-checked ·
Processors reach server market
AMD announced general availability of its sixth-generation EPYC server processors, based on its new Zen 6 architecture. Systems using the processors are shipping from server vendors, while cloud instances from Microsoft Azure and OCI are expected to be generally available soon.
The chips are designed for workloads including agentic AI, enterprise services, cloud computing and high-performance computing. The processors extend five generations of EPYC products and are intended to run AI agents, accelerated infrastructure and business applications.
AMD also says its fifth-generation EPYC processors became the fastest-ramping and highest-volume EPYC generation in its history. AMD describes the new lineup as a range of workload-optimized server CPUs, rather than a single processor configuration for every data-center task.
AMD’s performance claims
For general-purpose computing, AMD calls the EPYC 9996 the world’s best processor and says its system delivers up to 3.7 times the performance of an Nvidia Vera CPU-based system. AMD also reports up to 2.2 times the performance and 2.0 times the performance per dollar of an Intel Xeon 6980P system.
Those figures are AMD’s comparisons. AMD says the Vera comparison uses its internal estimate, which is based on engineering projections and subject to change.
For AI infrastructure, AMD describes three processor roles: general-purpose services, GPU host nodes and agent sandboxes. High-frequency host-node processors can reach up to 5.3 GHz and, by AMD’s projection, deliver up to 1.8 times the LLM output throughput of the previous EPYC generation.
AMD estimates that its high-density processors can provide up to about 3.3 times the agents per CPU watt of Nvidia Vera and about 2.5 times that of Intel Xeon 6980P. AMD says agent-count estimates use available CPU threads as a proxy, and that actual capacity and throughput will vary with workload and system configuration.
Cloud and rack deployments
AMD outlined three Helion rack options: Liquid for large-scale agentic AI deployments, Air for adding capacity in existing air-cooled data centers, and Air Inference for combining CPU execution with local GPU inference. Some Helion rack configurations reach up to 73,728 threads in one rack, AMD says.
AMD says Meta is validating EPYC-based platforms as the companies prepare for deployment at scale. Microsoft Azure has Azure HDv2 in preview for AI data processing, agentic AI and high-performance computing, AMD says. Azure HXv2 is for electronic design automation and technical computing, while Azure ND MI455X v7 is based on AMD Helios for production-scale AI inference, according to AMD.
Oracle Cloud Infrastructure’s E7 compute instances will use the new processors for enterprise applications and CPU services around AI models, according to AMD.