10 Sep AMD EPYC Is Poised For Big Gains, But Not For The Reasons You Think
AMD garnered a lot of attention with the launch of its EPYC 7002 Series processor a couple of weeks ago, and with good reason. It’s a highly performant part with a feature set that differentiates the product from the rest of market. The press has been very positive in its coverage of EPYC. The strength of EPYC lies not only in benchmarks, but in the strength of its ecosystem support and the real-world benefits associated with its single socket designs.
This single socket thing is real
The evolution of the server has been interesting to watch over the last few years. It wasn’t long ago when four-way servers (servers with four CPUs) were the backbone of IT infrastructure. When multi-core CPUs hit the market in the mid-2000s, virtualization allowed two socket servers to become the new norm in the datacenter.
AMD was on to something with the introduction of the first EPYC CPU (codenamed “Naples”) back in 2017 and its “no compromise single socket server” tagline. But the market wasn’t quite ready to embrace this. Perhaps this was due to server vendors fearing it could lead to lower average sales prices and software vendors worried their “per socket’ licensing could be negatively impacted. My personal belief is that the biggest reason is due to the conservative nature of the IT community. Enterprise IT administrators are a conservative bunch when it comes to technology adoption. The misguided belief that a single socket server would impact performance and reliability was a major barrier to adoption (though the reality is dual socket servers offer no level of “failover” that would increase reliability).
Fast forward a couple of years, and I believe “Rome” represents the next inflection point in server evolution-where single socket platforms begin to supplant dual socket as the standard for IT infrastructure. The capabilities of these single socket platforms are greater than what the market sees in competing two socket platforms, and the cost savings are undeniable.
TCO calculations can be fuzzy sometimes, but here’s some simple math to consider -“Rome” can reduce licensing costs by up to 75%, while increasingperformance. Forget all of the other operational costs such as power, management, and whatnot – IT organizations can reduce one of their biggest expenses in half and realize immediate savings.
The OEMs have bought in
The success of a CPU in the server marketplace is hard. While it’s the cornerstone of the datacenter, there are many layers of abstraction between the CPU and the applications that run our organizations. For starters, server vendors (OEMs) have to design systems that are optimized for that CPU.
The launch of “Rome” showed that OEMs have fully bought into the AMD vision. The company doubled its number of platforms in the market (compared to “Naples”-based systems).
- Hewlett Packard Enterprise (HPE) tripled its number of AMD-based SKUs, dropping Rome into its existing DL385 and DL325 Gen 10 platforms. In its original launch of AMD-based platforms, HPE aggressively positioned the DL385 and DL325 to apparent success. I expect HPE will double down on its aggressiveness as the performance gains in “Rome” make it a stronger fit across the enterprise – from virtualized infrastructure to database to high performance computing.
- While Dell Technologies didn’t announce any platforms at AMD’s launch event, CTO Robert Hormouth was very bullish on the Dell-AMD relationship and the joint-innovation taking place. Expect to see the results of that joint innovation in the not-too-distant future.
- Lenovo announced two new platforms based on “Rome.” Interestingly, both server platforms (SR635 and the SR655) are single socket platforms. This says a lot about how Lenovo sees the single socket market shaping up.
- The SR635 is a 1 rack unit single socket (1U/1P) server that the company messages as ideal for virtualization and hybrid IT. With support for up to 16 NVMe drives and 3 single wide GPUs, Lenovo hopes to target workloads such as online transaction processing (OLTP), analytics, and software defined storage. The company believes this server has affinity in grid computing, high frequency analytics (FinServ), electronic medical records (EMR) and electronic health records (EHR) analysis for healthcare. Additionally, Lenovo believes it is well-suited for capacity planning and supply chain optimization for telco and manufacturing.
- The SR655 is Lenovo’s 2U single socket server based on EPYC. Virtual Desktop Infrastructure (VDI) and other workloads that benefit from a richness of storage and acceleration (such as GPUs) are good candidates for the SR655.
Lenovo’s adoption of the EPYC 7002 Series processor is perhaps the most interesting play I see from the OEMs. Because Lenovo forwent building platforms on the original EPYC processor, I expected to see the company approach support a little more cautiously – much like its competition did with the launch of Naples. Rather, it appears the company fully embraced AMD’s “no compromise single socket” strategy. Given the Lenovo’s strength in the channel over the last few quarters, I expect it will find good momentum with its SR635 and SR655 platforms. The SR635 in particular has the potential to be a disruptive channel play if priced appropriately.
It will also be interesting to watch the innovations in supercomputing between Lenovo and AMD. EPYC is a server CPU that should shine in a number of HPC workloads. Lenovo has a relentless focus on HPC and maintains the lead on the Supercomputing 500 list (with 173).
While this article covers some of the major North American OEMs, it is important to note that AMD has found success across the hardware ecosystem and around the globe. For a server to be fully optimized for modern workloads, partnering with hardware technology providers is critical. Accelerators in the form of GPUs, ASICs, and FPGAs are required to enable the workloads that drive the modern datacenter. AMD’s work at shoring up its hardware ecosystem is evident in the chart above. What does this mean for an IT administrator? Servers based on EPYC should be able to power the most demanding workloads because of the depth of the EPYC hardware ecosystem.
Let’s not forget the software ecosystem…
CPUs and server technology mean nothing if the software supported by these platforms either doesn’t run or runs better on the competition. A vibrant ISV ecosystem is perhaps the most critical factor in the success (or failure) of a server platform. During the launch of Naples, I noted that the lack of ISV support hampered the success of the CPU in enterprise IT. Over the course of the last 24 months, the company quietly started to build an ecosystem for its CPU and is now a viable player for enterprise IT.
Note that the partnerships AMD fostered span the ecosystem, accounting for the applications and workloads that modern datacenters look for to gain a competitive edge. If there is an area where I believe AMD has most demonstrated its enterprise IT “worthiness,” it would be here. After all, it’s in the sphere of workload and application support that IT administrators and executives ultimately operate.
The one area of potential weakness for EPYC is in the area of artificial intelligence and machine learning (AI/ML). Intel INTC +0% invests a lot of time and resources into enabling this marketplace and has both an unrivaled product and IP portfolio. While this market may seem a bit like the wild west, AMD would be wise to build a stronger response to Intel and the score of silicon and hardware startups that are beginning to emerge.
Look at the little pictures when it comes to EPYC
It’s easy for AMD to wow the market with its new EPYC processor. Impressive specs give the company a decided advantage over a competitor that has had very little competition for the better part of a decade.
However, what’s most impressive about AMD’s accomplishments in the server space has less to do with EPYC’s almost gaudy features. Rather, it’s the company’s approach to foundational market building. Strategic partnerships, ecosystem development, and the ability to demonstrate real and tangible value to enterprise IT are all key to AMD’s successful disruption of the server market.
We will continue to track the progress of EPYC over the coming quarters, and how Intel responds to the new threat.