On 10 November 2026 AMD announced general availability of its 6th Gen EPYC server processors (SP7 socket, “Zen 6” architecture, codename Venice), saying in its newsroom release that systems built on the new CPUs are shipping from all leading server vendors. For university HPC centres and research computing offices, the practical change is that Zen 6 server hardware can now be quoted by system vendors, while the Microsoft Azure and Oracle Cloud Infrastructure instances AMD names are not yet generally available.
What AMD announced and when
AMD previewed the Venice family at its Advancing AI event in July 2026, as Techzine reported. The 10 November release is the general-availability step for the SP7 processors, led by the flagship EPYC 9996. IT Brief UK reported the launch on 11 November, together with a 6th Gen EPYC Embedded 9006 series that is not aimed at HPC procurement.
AMD’s release describes cloud, HPC and AI infrastructure as target markets. It names Meta as a lead customer validating platforms ahead of deployment at scale. It does not name a university or national laboratory customer.
Specifications that are confirmed
AMD’s own release gives limited detail. Its footnotes describe a dual-socket EPYC 9996 configuration with 512 cores, which implies 256 cores per socket, and list 600 W as “Default CPU Power”, a term AMD says replaces its historical TDP figure. Its test configuration uses 16 DDR5-8000 RDIMMs per socket.
Techzine reports further figures for the EPYC 9996: 256 Zen 6c cores and 512 threads on TSMC 2 nm, 16 memory channels, 128 PCIe 6.0 lanes with CXL 3.1, 1,024 MB of L3 cache and a suggested retail price of $14,904. Techzine adds that standard Zen 6 models reach up to 96 cores. These come from one outlet and should be checked against AMD’s SKU list before they go into a specification.
| Item | Detail | Source |
|---|---|---|
| General availability (SP7) | 10 November 2026 | AMD newsroom |
| Systems | Shipping from all leading server vendors | AMD newsroom |
| Azure | HDv2 in preview (AI data processing, agentic AI, HPC); HXv2 listed for EDA and technical computing | AMD newsroom |
| Oracle Cloud Infrastructure | OCI Compute E7 instances will be powered by 6th Gen EPYC | AMD newsroom |
| Later parts | SP8 in the first half of 2027 and Venice-X in the second half of 2027 | Techzine |
Performance claims are AMD’s own
AMD says the EPYC 9996 delivers up to 2.2 times the performance and 2.0 times the performance per dollar of an Intel Xeon 6980P system, and up to 3.7 times the performance of an Nvidia Vera CPU-based system. Its release states that the Vera score is an AMD estimate rather than a published result, and some of its LLM and agent figures are projections. The headline claims are measured on AMD-chosen workloads, so they say little about a given simulation code, genomics pipeline or finite-element solver.
What it means for research computing
The following points are editorial interpretation, not reported facts.
- Quotes and refresh cycles. A new CPU generation is normally what refresh planning waits for. Centres with a refresh in the next budget cycle can now request Zen 6 configurations from vendors, but should expect early pricing and lead times to differ from the suggested list prices above.
- Benchmark your own codes. Procurement documents for HPC usually specify acceptance benchmarks. Ask bidders for results on representative local workloads rather than relying on vendor headline ratios.
- Power and cooling. A 600 W default CPU power figure per socket is a facilities question as much as a compute one. Check rack power, cooling and the power-management settings before comparing core counts. See our entry on sustainable HPC.
- Cloud as a bridge. Azure HDv2 is in preview and OCI E7 is announced, so burst capacity on Zen 6 is not yet something a research office can plan a grant budget around. Cloud pricing was not part of AMD’s release.
- Memory and interconnect. Core count is one variable. Memory bandwidth per core, PCIe generation and accelerator attachment often decide application performance, and those details should be confirmed in the final SKU documentation.
The launch follows other infrastructure announcements this month, including Google Cloud’s SC26 HPC and AI infrastructure and Google’s TPU 8 chips. For context on what leadership-class systems look like, see our guide to the El Capitan supercomputer.
