AMD Making Strides with EPYC Server Platform

on December 4, 2017
Reading Time: 3 minutes

We all know that a competitive market is one that is healthy. Multiple options for your PC, your smartphone, and yes, your server, mean that every party involved needs to be more aggressive in development to outdo the competitor. For many long years, that type of environment did not exist in the server space and Intel was able to dominate the field with the Xeon processor family and almost no pressure from outside companies.

AMD announced its EPYC processor family this past summer and though it always takes time for adoption and ramp of a new enterprise-class technology, there has been more angst to see retail-ready releases from this launch than any other. Many questioned AMD’s ability to re-enter the server market with Intel’s 99%+ market share and strong grip on the hardware and software ecosystem. Any noise or promotion that might come from partners would be welcome and required for the community at large to have confidence in AMD’s claims.

HPE Claims Performance Records with EPYC

Last month we got one of our first, and likely most important to date. HPE not only announced its second family of servers that would be integrating AMD EPYC processors but did so with a press release touting record breaking performance and impressive claims across the board. For those that want the details: the HPE ProLiant DL385 Gen10 system with dual-socket EPYC processors broke records for two-socket systems on SPECrate2017_fp_base and SPECfp_rate2006. In short, the server showed impressive scaling in floating point performance by combining a pair of 32-core EPYC processors in this industry standard benchmark.

Additionally, HPE claimed that this platform would offer “up to 50% lower cost per virtual machine” compared to other dual-socket servers, thanks in part to the 4TB of addressable system memory and 128 lanes of available PCI Express provided by the AMD CPUs.

This is just one server family, and just one OEM (a big one), but this marks another milestone in AMD’s march back to relevancy in the server space. AMD CEO Lisa Su cautioned me recently that this would be a slow and arduous process, even if AMD was excited about the rate of adoptions it was seeing. It tells me that AMD is doing the right things, working with the right people, and has the right mindset and aggressive stance to make waves in the enterprise space once again.

Intel is Paying Attention

For its part, Intel is taking notice. Though it has a 99% market share in the server and data center space, Intel went to press last week with internal testing that compares its own Xeon Scalable processor family against the AMD EPYC platform. It runs through a myriad of benchmarks and concludes that Intel still offers an advantage in performance per core and many of the workloads and benchmarks that server professionals look to for guidance. Intel also questioned the performance consistency of AMD EPYC processors because of the complications surrounding its multi-die approach to core scaling (as opposed to a single, monolithic die that Intel utilizes).

It’s unlikely that any of the results that Intel presented to the media are “wrong” but the importance of the effort in my mind, is that Intel felt pressured enough to address this in a public fashion. All companies do competitive analysis on systems and hardware but rarely is that data presented in such a fashion to essentially “call out” that competing company and the coverage by media and analysts. It means that Intel sees a threat and is taking it seriously – something it hasn’t done in this space for nearly a decade.

AMD was upfront during its launch of EPYC that it would do very well in specific areas of the enterprise and datacenter space, but in others it would be behind what Intel can offer. That still seems like an accurate assessment though Intel is doing some of the heavy lifting to indicate where those “other” areas are. I still view EPYC as competitive in enough areas to retain its original value proposition and it appears that partners like Supermicro and HPE agree.

As we move forward, the future in the server space is brighter for customers thanks to a competitive landscape. Intel executives and financial bottom lines won’t appreciate any drop from the near-100% market share, obviously, but for the rest of us, seeing a healthy and active AMD in this space is a critical piece of the story of improvement and scalability for the datacenter. AMD should continue to see customer adoption and a resulting improvement in the financial status of the enterprise business unit