By Hamza Faisal
AMD has launched its 6th Gen EPYC server processors, codenamed Venice, bringing its new Zen 6 architecture to data centres with a flagship chip packing 256 cores per socket. The company announced general availability on Tuesday, November 10, pitching the lineup squarely at the AI agent workloads that now dominate server demand.
The Venice family comprises 10 processors split into two groups: high-density chips built for maximum core counts, and high-frequency parts tuned for speed. The flagship EPYC 9996 offers 256 cores and 512 threads per socket, a boost clock of up to 4.1 GHz, 1,024 MB of L3 cache and a 600 W power rating. The fastest frequency-optimised models carry 96 cores with a 4.05 GHz base clock, a 5.3 GHz boost and 384 MB of cache. The platform supports 16 DDR5 memory channels and 96 PCIe 6.0 lanes, and the compute dies are built on TSMC's 2-nanometre-class N2P process.
AMD's pitch: more agents per watt than Nvidia's Vera
AMD's headline claim is efficiency: the company says the EPYC 9996 can run up to 3.3 times as many AI agents per watt as Nvidia's Vera CPU, and deliver up to 3.7 times the performance. It also claims roughly double the performance of Intel's Xeon 6980P, and twice the performance per dollar, alongside a 27 percent uplift in instructions per clock over its own Zen 5 generation. All of those figures are AMD's own, and none has been independently verified.
The first independent numbers are more mixed. Tom's Hardware testing found a dual-socket EPYC 9996 system scored up to 93 percent higher than a dual-socket EPYC 9965 on an estimated SPEC CPU 2026 integer-rate run. But in a database test published the same day, the 256-core flagship ran slower than the previous generation's 192-core EPYC 9965 — a reminder that more cores do not win every workload. Notably, AMD's own chip architects have said Venice was not designed specifically for agentic AI, even as the company's marketing leans hard into the agent era.
Servers shipping now, cloud instances to follow
Servers with the new chips are already shipping from major data centre suppliers, including Quanta Cloud Technology, Lenovo, Hyve Solutions and Cornelis Networks. Cloud availability is close behind: Microsoft Azure is previewing HDv2 instances powered by the chips for AI data processing and high-performance computing, and Oracle Cloud Infrastructure plans Compute E7 instances for enterprise workloads, though neither cloud vendor has given a firm date.
AMD did not announce pricing with the launch. A price list published in September puts the range at $700 to $14,904 for the EPYC 9996 in 1,000-unit quantities. The new chips use the SP7 socket, which is not backward compatible with the SP5 socket used by the previous Turin generation, so existing server boards cannot be upgraded in place.
What comes next
Venice is the first step in a wider Zen 6 rollout. AMD plans a smaller SP8 platform spanning 8 to 128 cores in the first half of 2027, followed by Venice-X variants with 3D V-Cache later in 2027 and low-power Verano chips with LPDDR5X memory in the second half of the year. Consumer Ryzen processors based on Zen 6 are expected next year but have not been announced.
The launch lands on the same day Nvidia's Vera CPU received its first full independent reviews, setting up a direct contest for the CPUs that sit beside AI accelerators in modern data centres. AMD unveiled the Venice family in July; with general availability now live, the benchmarks — and the orders — will decide whether its agent-era bet pays off.



