cancel
Showing results for 
Search instead for 
Did you mean: 

AMD EPYC: Your No-Compromises Path to an AI-Ready Datacenter Today

Robert_Hormuth
1 0 979

Introduction

Data centers are the backbone of the modern global economy, providing myriad services for a variety of applications and users. However, data centers also consume a lot of energy, between 10 and 50 times the energy per square foot of floor space as a typical office building[i]. In total, the International Energy Agency estimates that data centers alone account for between 1-1.5% of global energy consumption[ii]--that is more than many countries each consume in most years.

As foundational data center workloads continue and AI becomes a fundamental part the data center, the energy efficiency of data center hardware becomes a critical element for reducing operational costs, minimizing environmental impact, and enabling business growth.   The quest to reduce data center energy consumption had led to some suggestions for new architectural approaches backed by seemingly attractive claims. 

In this blog post, we will explore how the unique approach from AMD achieves best-in-class energy efficiency with neither compromise in performance nor requiring an expensive architectural transition to alternative architectures such as ARM or RISC-V. We will focus on the advantages delivered by AMD EPYC CPU-based servers while providing insights into the realities behind the claims of alternative-architecture systems touting energy efficiency that might be more challenging to realize in a real data center and business environment.

AMD EPYC: x86 architecture plus energy efficiency leadership

 

AMD implements x86 architecture in an extremely efficient Zen4 core—and an even more optimized but fully compatible Zen4c core, leveraging the best semiconductor process nodes. AMD has led the charge in chiplet architectures and implemented various features in silicon to minimize and enable fine-grained control over power consumption without compromising on performance. The numbers speak for themselves with the 4th generation EPYC-based server delivering more than 300 world-record performance results across a massive variety of workloads and deployment scenarios. While it has been clear for some time now that Intel has struggled to maintain competitiveness on performance and energy efficiency for the last several years, AMD has continued to drive x86 performance and efficiency advancements. AMD has also been clearly demonstrating AMD EPYC-based platform efficiency benefits compared to the Arm-based alternatives from Ampere, Nvidia or others with esoteric architectures.  Let’s look at some data-driven headlines:

Of course, AMD does an enormous amount of benchmarking to help us verify delivery of the type of performance and efficiency our customers expect. One of my favorites is located here, showing our published SPECpower_ssj® 2008 results compared to published results for Intel® Xeon® Platinum 8490H and Ampere® Altra® Max M128-30. The AMD EPYC 9754-based systems show between ~2.5X and 2.7X better ssl_ops/watt advantage versus the Ampere solution.  And versus the Intel Xeon, the EPYC efficiency advantage is ~2.1 and 2.2X.

The AMD-based systems I’m citing here are systems you can buy today, not futures, not projections, and are published, verifiable results.  These are systems that leading business customers are deploying easily in on-prem, public or hybrid cloud options.  You can order these today from system vendors you know and trust.  

Compatibility: The Perfect Complement of Efficiency and Performance

 

The ability to count on AMD as a critical element of your time-tested IT supply chain cannot be discounted. AMD performance and efficiency means that data center operators can continue to leverage the existing x86-based infrastructure and software, without requiring major changes or investments. x86 architecture is compatible with most of the operating systems, virtualization platforms, and software frameworks that are used in today’s data centers, such as Linux, Windows, VMware, Kubernetes, Apache, and TensorFlow. AMD compatibility across x86 architecture can also support multiple generations of processors, allowing data center operators to upgrade their systems with minimal disruption and cost.

Standard AMD EPYC-based servers support continuous data center innovation and transformation.  Providing a flexible and open platform for developing and deploying new applications and services. x86 architecture can also support the latest (as well as legacy) standards and protocols, such as PCIe, Ethernet, USB, and NVMe, which can enable data center operators to utilize the peripherals, accessories, and connectivity they need for evolving business requirements.

Conclusion: AMD EPYC for No-Compromise Data Center Efficiency

 

AMD enables a leading server platform for data center computing, offering high performance, energy efficiency, and compatibility for a variety of data center workloads. The no-compromise AMD approach enables data center operators to leverage their existing infrastructure and software and facilitate data center innovation and transformation on-prem or in the cloud and from traditional enterprise workloads to the most demanding AI workloads. Despite the promise of disruptors seeking to revolutionize server architectures, AMD EPYC-based servers and cloud instances from all leading trusted suppliers is a proven approach with no compromises on either performance or efficiency. As such AMD is the best choice for data center operators who want to reduce their operational costs and environmental impact, while improving their service quality and customer satisfaction.

 

[i] Source: U.S. Department of Energy: https://www.energy.gov/eere/buildings/data-centers-and-servers

[ii] Source: IEA: https://www.iea.org/energy-system/buildings/data-centres-and-data-transmission-networks

About the Author
Robert Hormuth is Corporate Vice President, Architecture and Strategy of the Datacenter Solutions Group (DSG) at AMD. Robert has 36 years in the computer industry, joining AMD in 2020 after 13 years with Dell where he was CTO of the Server Business unit, 8 years with Intel and 11 years at National Instruments. At AMD Robert is charged with creating the long-term vision & strategy for DSG by identifying the technical requirements/implications to the DSG portfolio. Robert has a B.S. in Electrical and Computer Engineering from The University of Texas at Austin and currently holds 47 patents.