TL;DR
- AMD EPYC processors deliver 1.6x performance per watt and better performance per dollar compared to the current generation of Intel Xeon processors
- AMD EPYC™ 9005 server CPUs provide more cores and support for high frequencies and faster DRAM than the previous generation
- AMD EPYC™ CPUs have exceptional core density with up to 192 cores (9005 series)
- AMD EPYC™ server CPUs also have AMD Infinity Guard features that work at the silicon level
Building an AI-ready data center is a long and complex process with many parts.
There is no one piece of critical technology at the heart of it; in fact, there are many processes happening in different places, with their own unique technical requirements. To train AI, you need to first collect, process, and store data, usually across a multitude of systems. You also need specialized hardware like GPUs to train on the data.
When we get to the point of real-time insights, which typically need low-latency and high efficiency, the CPU is an important host node.AI requires some special features. Here, we will highlight the best features that AMD EPYC CPUs have to offer…
What is AMD EPYC?
AMD EPYC is the AMD line of server and data center processors. 6th Gen AMD EPYC™ server CPUs, also known as AMD EPYC™ 9006 series server CPUs, are the most recent generation of the EPYC™ family.
Industry-leading performance and value
AMD is a name synonymous with power; 40% of the top ten fastest supercomputers in the world, as measured by TOP500 in June 2026 are powered by AMD technology, and 40% of the top ten most energy efficient also use its hardware, according to the June 2026 Green500 list.
AMD EPYC processors are also high-value, delivering better performance per dollar than Intel Xeon processors with up to 64 cores at an affordable price.
Consolidate space, increase power
AI-ready data centers need compute infrastructure optimized for data ingestion and preprocessing, not just training itself. This is because training data must be efficiently structured, validated, and organized before training can begin. As such, server-grade processors are essential for streamlining these processes.
AI-optimized processors typically have lower power-to-computation ratios and can maintain the high compute demands of AI model training without sacrificing performance or drawing excessive amounts of energy. With the right software and orchestration layer, the AI data center can dynamically balance workloads to help ensure workloads run on the most efficient compute nodes, while energy-efficient interconnects can reduce wasted compute cycles.
AMD EPYC™ 9005 server CPUs can accelerate your AI projects while freeing up space with their potential to consolidate data centers. All while they perform fast AI inference and host GPU-accelerated systems efficiently. EPYC CPUs excel for small to medium AI inference models, and help maximize GPU accelerator performance by handling data preparation and post-processing tasks efficiently as the host processors in GPU-accelerated systems.
The AMD EPYC™ 9005 server CPUs are also ideal for businesses seeking to consolidate and modernize IT infrastructure. Take, for example, 2P servers using AMD EPYC™ 9965 CPUs that deliver 1.6 times the performance per watt compared to Intel Xeon 6980P CPUs.
To achieve approximately 391,000 units of integer performance, you can replace 100 old 2P Xeon 8280 CPU-based servers with 13 new AMD EPYC™ 9655 CPU-based servers or 87% fewer servers that use an estimated 68% less power. To achieve the same performance level, you would need 16 2P Intel Xeon 6980P CPU-based servers.
Higher cores, higher frequencies
Perhaps the biggest advantage of AMD EPYC CPUs is their excellent core density.
Many inference workloads are suited to AI-optimized CPUs. AI model training consumes significant amounts of energy, with larger models requiring considerably more than small, single-purpose models. Also, AI training involves sustained, resource-hungry workloads, unlike the short bursts of computation typically used for inference.
Two 192-core EPYC™ 9965 Server CPUs can achieve up to 1.9 times the inference throughput when running extreme gradient boosting with the Higgs boson data set compared to servers with two 128-core Intel Xeon 6980P CPUs.
5th Gen EPYC CPUs’ high-core-count architectures, combined with high memory bandwidth and caching, make them a big asset for AI workloads. Additionally, support for parallel data ingestion helps data centers to process multiple streams of incoming data, simultaneously, helping reduce bottlenecks in AI workflows.
AMD Infinity Guard
Another key feature of AMD EPYC processors is the built-in layers of security. AMD Infinity Guard works from silicon-level protections through virtualization and confidential computing environments. It also includes memory encryption features that help protect data in use across applications and virtual machines..
Infinity Guard is a security suite built from the ground up, and one that continues to evolve. 5th Gen AMD EPYC processors come with advanced features, like secure Encrypted Virtualization (AMD SEV). This is an Infinity Guard technology that helps protect data in use through encrypted, confidential virtual machines, with no code changes in order to deploy. SEV is built on industry standards and supported by the industry’s most mature confidential computing ecosystem. What’s more, solutions powered by AMD SEV are available for deployment from major cloud providers, such as Linux and Windows operating systems, virtualization platforms, and OEM hardware.
For more information on what AMD EPYC CPUs can offer your business