In case you weren’t convinced by AMD’s re-entry to the server market with the initial launch of EPYC in 2017, the company today launched an what I consider an incredible follow-on product with EPYC 7002 Series (codenamed Rome). While the market was excited to see the company re-enter the market with the original EPYC CPU, there was still a little bit of a “wait and see” attitude. A couple years and about 5% market share points later, “wait and see” has turned into “hurry up and buy.”
What is Rome? What should the market expect in terms of performance and utility in the datacenter? Will AMD eat into Intel’s share? What does the path to success look like for AMD and EPYC? This article will try to examine these questions and a little more.
First – Rome was not built in a day
The EPYC 7002 Series CPU is not an incremental build on the previous generation (codenamed “Naples”). Rather, The EPYC 7002 Series is a new core design built on the industry’s first 7nm process (referencing the size of the transistors), utilizing a “hybrid multi-die design” (more on this later). What this all adds up to is a CPU that can deliver twice the core density at the same power envelope and in a slightly smaller package.
The EPYC 7002 Series CPU is industry-leading in virtually every relevant feature: cores, memory capacity, memory bandwidth, PCIe lanes (and Gen 4 support), and security features. These are improved performant cores, to boot, which means EPYC should be able to support the most demanding high-performance workloads as well as the less glamorous virtualized infrastructure.
AMD was smart to improve the technology around the entire EPYC package. Enhancements to the cache architecture, including the doubling of the L3 Cache size should make this a good platform for traditional relational database management systems (RDBMS) such as Microsoft SQL Server and PostgreSQL. Additionally, the doubling of the floating-point data path width to 256 bytes should expand high-performance computing workloads where EPYC has game.
By the way, what is a hybrid multi-die design, or, as we call it, “chiplets?” Basically, it is a design process intended to increase the compute density within a given physical footprint, by creating chip modules that connect on a separate die that contains I/O and memory controllers. With EPYC, AMD connected the chip modules via a high-speed interconnect, named Infinity Fabric. This design delivers better “per socket performance” across the board—from cores to memory, as “hops” from module to module are reduced through the I/O and memory die. What this all means is better real-world performance of the servers powering your datacenter.
What does “Rome” mean for IT Organizations?
The EPYC 7002 Series should bring a number of benefits to IT organizations. Three obvious advantages are performance gains, tighter security, and lower total cost of ownership. Let’s dig into each of these a little more:
Performance gains: It’s often the little things that make all the difference in the world, and in some ways, this is the case with “Rome.” Yes, the core counts doubled. Yes, it’s the first 7nm x86 server CPU on the market. Yes, it’s borderline embarrassingly feature-rich. Yet, it’s the things an IT admin may not notice that make “Rome” such an ideal platform, like improvements made to the core execution engine that allow Rome cores to perform quicker and with greater efficiency and precision. These enhancements are what make servers based on the EPYC 7002 Series CPU great platforms for virtually all the workloads that power the datacenter.
So, what does all of this mean for an IT Administrator?
- A lot more virtual machines per server than what they see on their current infrastructure, and those VMs are running at bare metal speeds.
- Line of business applications running on traditional three-tiered architectures that can process orders faster and handle inquires quicker.
- Big data and analytics environments that drive real-time insights.
Tighter security: When AMD launched its initial EPYC CPU (“Naples”), security was a big focus of the company’s messaging. AMD designed EPYC to secure servers and the data that resides on them from initial boot to end of life. The CPU seemed to be immune to many of the exploits that plagued its competition.
There was still more to EPYC than hardening of the architecture. AMD also announced a comprehensive security strategy.
- AMD Secure Processor ensures servers boot to a clean and untampered environment.
- AMD Secure Memory Encryption (SME)physically encrypts server memory, preventing scraping and physical attacks.
- AMD Secure Encrypted Virtualization (SEV)encrypts and isolates virtual machines, providing protection of VMs against attacks, other exploited virtual machines, and even from rogue administrators.
While AMD Secure Processor and SME worked seamlessly, SEV required some ecosystem support. So, while it was an important feature when “Naples” launched, its value could not be realized as the ISV community had not yet enabled it. Fast forward, and the EPYC 7002 Series brings the full realization of this feature. The Linux community embraced support for SEV and we can expect to see VMware enable its capabilities as well.
Total cost of ownership (TCO): Any and every IT Administrator will say that TCO carries very little value since IT solutions providers can manipulate numbers to show value. Isn’t it incredible that every vendor in the market can save IT organizations money and free resources? I believe IT organizations that invest in servers based on the EPYC 7002 Series really can achieve savings—and noticeable savings at that.
IT organizations deploying two-socket servers based on the EPYC 7002 Series CPU can realize levels of VM density that are almost too good to be true: 128 cores/256 threads feeding 4TB of RAM across 16 memory channels. This is a beast of a virtualization platform. Capital expenditures should drop considerably due to consolidation, and operational costs should drop due to lower power and cooling as well as reduced management overhead.
IT organizations that are a little more forward-thinking can realize even greater savings by deploying single-socket servers based on EPYC. These servers can deliver VM density and performance levels that exceed most of two-socket servers populating the datacenter today. For that matter, socket-based licensing costs are cut in half. Why more IT organizations aren’t exploring this option is baffling, given the significance of software licensing on IT budgets.
I have a hard time buying into specific claims of TCO savings—seriously, how can a company promise its technology will deliver a 17.4% TCO savings? Still, I do believe the savings an organization can realize with EPYC are obvious and significant.
Will AMD capture significant market share?
In a word, yes. With the initial launch of EPYC, AMD found a lot of success with hyperscalers. Cloud providers attended the EPYC launch and spoke of plans to adopt and every following quarter seemed to include the announcement of a new cloud customer. Meanwhile, the initial EPYC CPU was used for enterprise IT to prove out AMD as a viable alternative in the datacenter.
Expect to see AMD capture significant market share with the EPYC 7002 Series as enterprise and mid-market IT organizations begin to adopt in large scale. All of this, while cloud providers continue to increase the footprint of EPYC in datacenters around the globe.
What does success look like?
When AMD launched EPYC, the company heavily focused on the “low hanging fruit” in the market—cloud providers that could realize the benefit of EPYC and deliver much needed revenue (and share) to AMD.
If AMD is wise in its go-to-market (GTM) approach, Rome will mark the expansion of AMD’s addressable market in the datacenter. Based on conversations I’ve had with IT executives, the trepidation of deploying EPYC in enterprise datacenter and downstream should be gone. Enthusiasm is high. The market is ready for AMD.
AMD has quietly been making investments in building out its infrastructure to drive market reach directly and through its OEM and channel partners. I expect these investments to continue and increase as the company starts to see success in the mainstream market.
AMD should also see greater success in the HPC market. The company found success in HPC workloads that were memory bound. Its increased integer and floating point should lead to a much broader set of applications. Keep an eye on this market as AMD will aggressively chase (and score) design wins.
The ultimate measure of success is market share. I will refrain from making any specific predictions around share gains, but I do believe former Intel CEO Brian Krzanich’s warning that Intel needed to prevent AMD from gaining 15-20% market share in the datacenter could prove to be prophetic.
Where is AMD challenged?
Rome is a great architecture and the EPYC 7002 Series CPU should see success. However, AMD is fighting against a competitor that has a lot of gravity in the server market. A few things AMD needs to account for are:
- Strengthening the indirect sales force: Great technologies don’t mean anything if they aren’t used. AMD now has a presence in every major OEM. Its challenge is to now turn those OEM sales channels into Rome advocates. Further, the company must continue to invest the IT solutions vendors (e.g. CDW, Connection, SHI) to make servers based on Rome a part of every conversation. Winning those transactional sales accounts for a lot of volume, which translates into much-needed market share.
- Standing up the ecosystem: A great CPU architecture is the cornerstone to the datacenter. However, the CPU requires adjacent technologies to become a complete solution. AMD is competing with a company that has a deep portfolio of those adjacent technologies, from application accelerators to persistent memory. Therefore, it’s very important for AMD to demonstrate similar capabilities through its ecosystem partners. Done right, this can turn from a weakness to a strength, as it will show an openness to its market approach.
- Aligning to the cutting edge: Machine learning and artificial intelligence (ML and AI respectively) are examples of emerging technologies where AMD’s competition has put a lot of emphasis. These technologies also happen to be top of mind for most enterprise IT executives, even if not widely deployed. AMD would be wise to demonstrate the affinity between Rome and such technologies. It will help counter arguments at the CIO level around AMD’s ability to support the needs of enterprise IT fully.
Contrary to how this article reads, I’m not an AMD fanboy. When the company launched “Naples,” I gave an assessment of the company’s prospects that was more sober than those of an overenthusiastic market. Not because of the product or the company, but because of the conservative nature of IT organizations. Additionally, AMD needed to rebuild trust with the market—from OEM partners to the ISV ecosystem to the channel partners that support those conservative IT organizations.
Over the last two years or so, the company has established itself as not just a viable alternative in the market, but a smart choice. Rome signifies more than a great product. It signals AMD’s return to the enterprise datacenter.