AMD Prices 256-Core EPYC 9996 At $14,904 For Server Buyers

At $14,904 for 1,000-unit orders, the 256-core EPYC 9996 is AMD’s highest-priced CPU, according to specifications and pricing for the 6th Gen EPYC 9006 “Venice” family reported by TechRadar.

The new top-end chip combines 256 Zen 6c cores, 512 threads, 1,024MB of L3 cache and a 600W default power rating on AMD’s SP7 socket.

The published lineup covers 31 processors, making the EPYC 9996 the headline part in a broader server refresh aimed at high-density enterprise and AI infrastructure buyers.

The comparison is with AMD’s previous flagship.

TechRadar noted that the 192-core EPYC 9965 “Turin” part launched in October 2024 at $14,813.

The Venice flagship therefore adds 64 compact Zen 6c cores for a $91 higher list price, shifting the list-price comparison from about $77 per core on the older part to about $58 per core on the new one.

Benchmarks referenced by the coverage point to a significant performance upgrade for roughly a 1% increase in list price.

That framing helps explain why the new processor is positioned less like an incremental catalog part and more like a density play for buyers trying to raise throughput inside constrained data centre footprints.

That sticker comparison has limits in the server market.

Hyperscalers and OEM customers usually negotiate well below published 1,000-unit pricing, so the public figure is more guidance than a direct invoice for large deployments.

Even so, the flat pricing signal gives partners a simple way to frame the core-count jump as AMD competes for infrastructure spending tied to AI systems and dense compute clusters.

The EPYC 9996 is not the most expensive server CPU launched in recent years.

Intel’s 128-core Xeon 6980P arrived in 2024 at $17,800, leaving AMD’s new list price below that rival part despite offering twice the core count.

The comparison underscores how server CPU competition is now being measured not only by absolute price, but also by socket power, memory bandwidth, I/O and the economics of filling racks.

AMD has also changed how it labels power for this class of processor.

The company has moved away from a traditional TDP figure and uses a “default CPU power” rating intended to reflect consumption under defined workloads.

At 600W, the EPYC 9996 is a part for engineered server platforms rather than mainstream workstations or enthusiast desktops.

The platform features reinforce that divide.

PCIe 6.0 support gives the chip access to a newer generation of high-speed expansion, while MRDIMM support targets higher memory bandwidth through registered, buffered server modules.

Those technologies are relevant to current-generation data centre systems, but PCIe 6.0 is not expected to reach desktop PCs until around 2030 and MRDIMM may remain a server-only technology for the foreseeable future.

AMD’s position gives the launch extra context.

Mercury Research figures put AMD at 46% of server CPU revenue, a level that makes the company’s own previous generation one of the clearest benchmarks for the Venice flagship.

The article also points to OEM interest in Nvidia’s Vera Rubin offerings as an alternative in the same infrastructure market.

The operating question for buyers is whether the extra cores, cache, I/O and memory support justify platform power and negotiated system cost, rather than whether the public list price alone defines the upgrade.

For AMD, the list price keeps the flagship near the prior generation while using core density and server-only platform features to sharpen the case for Venice deployments.