Skip to main content
AMD unveils 6th-generation EPYC as AI competition heats up in server CPUs - 디지털투데이

AMD unveils 6th-generation EPYC as AI competition heats up in server CPUs - 디지털투데이

AMD Processors
12 lis 2026, 00:50Hardware News
Oryginał
1 min read

AMD has unveiled its 6th-generation EPYC 9006 server processor with up to 256 cores, intensifying competition in the AI data center market against rivals Nvidia and Intel.

On Nov. 10, AMD officially launched the SP7 platform, the first product family of its 6th-generation EPYC processor previously known by the codename Venice. The new processors are based on the Zen 6 architecture and support up to 256 cores and 512 threads, applying next-generation connectivity technologies including DDR5 and high-bandwidth MRDIMM memory, as well as PCIe 6.0 and CXL 3.1.

Performance Claims

AMD emphasized that server CPUs play an expanding role as AI spreads, requiring high core density and memory bandwidth for multiple simultaneous tasks. The company claimed that a dual-CPU server equipped with the EPYC 9996 recorded up to 3.7 times higher integer computing performance than a system based on Nvidia Vera CPUs and 2.2 times higher performance than an Intel Xeon 6980P system.

The high-clock products operate at up to 5.3 GHz and can increase large language model (LLM) output throughput by up to 1.8 times from the previous generation.

AMD

Market Expansion

Along with the new processors, AMD highlighted Helion, a rack-scale AI infrastructure design that supports companies in building large-scale AI agent execution environments. Servers based on the 6th-generation EPYC have begun shipping through major manufacturers, with Microsoft Azure providing a preview of related cloud services and Oracle Cloud Infrastructure set to roll out services applying the new products.

AMD plans to continue expanding product lines tailored to uses such as AI host nodes, enterprise servers and high-performance computing. The company noted that performance figures are based on internal measurements and can vary depending on server configuration and workload type.

Z.ai
Article written by AI
Z.ai GLM-4.5-FlashFree
Free AI articles150 of 150
No limit after signing in with Telegram