AMD launches 256-core EPYC 9006 CPUs to fight Intel and Nvidia

AMD has launched its 6th Generation EPYC server processors with up to 256 cores per socket, positioning the lineup directly against Intel's Xeon enterprise franchise and Nvidia's Arm-based Vera CPU in the contest to power AI infrastructure. Systems built on the new chips are shipping now from leading server manufacturers and original design manufacturers, with cloud instances from Microsoft Azure and Oracle Cloud Infrastructure (OCI) to follow soon.

The release lands as enterprises race to deploy AI agents and expand AI infrastructure, a buildout that has reordered the data-center CPU market and drawn Nvidia, the dominant supplier of AI accelerators, into direct competition for general-purpose server silicon. AMD's answer is a two-front product: an x86 CPU family that must beat Intel on price-performance while also feeding GPUs and hosting the agent workloads that increasingly define enterprise computing budgets.

"With 6th Gen AMD EPYC, customers can standardize on the proven x86 foundation while choosing a workload-optimized solution to run more agents, keep accelerated infrastructure fed and power the business-critical applications they depend on," said Dan McNamara, senior vice president and general manager, Compute and Enterprise AI, AMD, in a statement.

What does the hardware offer?

The EPYC 9006 SP7 range comprises 10 processor models, with configurations spanning 64 to 256 cores and up to 512 threads per socket. The portfolio splits into two camps:

• High-density processors for workloads that require large numbers of simultaneous computing tasks, such as cloud hosting and AI agent sandboxes.
• High-frequency variants with boost speeds of up to 5.3GHz, aimed at workloads that need fast individual cores.

Across the range, the processors support 16 channels of DDR5 memory and PCIe Gen 6 connectivity, which delivers twice the bandwidth per lane compared with PCIe Gen 5.

The design philosophy matters as much as the core counts, according to Manish Rawat, semiconductor analyst at TechInsights. "Rather than simply increasing core counts, AMD aims to deliver more computing power within existing power and infrastructure limits. This makes the platform particularly relevant for cloud computing, virtualization, high-performance computing (HPC), enterprise databases, data analytics and AI infrastructure," Rawat said.

That constraint — more compute inside the same power and cooling envelope — is the practical test for data-center operators whose facilities are already near capacity. A chip that consolidates workloads without new racks or new power contracts changes procurement math in a way raw specifications do not.

How does AMD benchmark against Intel and Nvidia?

AMD is pitching the new processors against Intel's Xeon and Nvidia's Vera CPUs, and it has published internal and third-party benchmark comparisons to support the claims. The numbers favor AMD across the board:

• A server powered by the flagship 256-core EPYC 9996 delivers up to 3.7 times the performance of a system powered by Nvidia's Vera CPU.
• Against Intel's Xeon 6980P, the same chip delivers up to 2.2 times the performance and twice the performance per dollar.
• For AI agent workloads, AMD's high-density EPYC processors deliver up to 3.3 times more agents per CPU watt than Nvidia Vera, and 2.5 times more than Intel Xeon 6980P, when hosting agent sandboxes.
• AMD's high-frequency processors deliver up to 1.8 times the LLM output throughput of previous-generation EPYC processors in CPU-offloaded inference workloads.

The three competitors represent three different approaches to the same market, Rawat said. "AMD competes through high core density, memory capabilities and power efficiency, while Intel leverages its established enterprise ecosystem, platform compatibility and customer relationships. Nvidia is taking a different approach with Vera, an Arm-based CPU designed for AI and accelerated computing infrastructure. Unlike AMD and Intel, which target a broad range of general-purpose server workloads, Nvidia is positioning Vera as part of an integrated AI computing platform."

Analysts are already drawing dividing lines between the platforms for buyers. Danish Faruqui, CEO at Fab Economics, recommends AMD's high-core-count EPYC processors for cloud consolidation, VDI and AI agent workloads. He points to the high-frequency EPYC models for databases and enterprise applications where software licensing costs are a concern, since licensing often scales with core count and fewer, faster cores can cut those bills. For memory-intensive HPC and scientific workloads, he recommends EPYC 9006. He recommends Nvidia Vera for tightly integrated AI training and inference clusters, and Intel Xeon 6 for enterprises that want to maintain existing x86 infrastructure, particularly in air-cooled racks.

What should CIOs evaluate before buying?

The right choice depends on the workloads, not on the highest core count on a spec sheet. "AI agents and cloud applications may benefit from more cores, while some databases need faster individual cores. They should compare performance, electricity consumption, memory capacity, software compatibility, and cost. The best processor is the one that runs their applications most efficiently," said Pareekh Jain, CEO at EIIRTrend & Pareekh Consulting.

On price, AMD's flagship EPYC 9996 costs around $14,900, a figure broadly comparable to high-end Intel Xeon processors, Jain noted. Nvidia's Vera pricing is less transparent, he added, which complicates direct comparison for buyers weighing integrated AI platforms against general-purpose servers.

Jain advised enterprises to calculate total cost of ownership over three to five years. That calculation should include:

• Hardware acquisition costs.
• Electricity and cooling.
• Software licensing.
• Maintenance.
• Migration costs.

The consolidation argument cuts in AMD's favor in specific scenarios. The higher computing capacity of the new chips could let companies replace several older servers with fewer new ones. But expensive software licensing and migration costs could erode those savings. Enterprises running servers that are three to six years old, facing high electricity bills, or expanding their AI workloads have the strongest case for upgrading now. Companies with relatively new servers may not see enough savings to justify a replacement cycle.

The launch, the benchmarks and the Azure and OCI commitments give AMD a credible entry point into the AI-driven server refresh cycle. Whether the claims survive independent testing at customer scale will determine if the company converts silicon leadership into enterprise market share.