When AMD exited the server market back in 2012, it held 10% market share. While not a large chunk of the market, the legacy of Opteron and relationships with OEMs and big datacenter customers translated into market significance. This was important, as this market significance assured broad software ecosystem support when AMD released a new generation of Opteron.
Fast forward five years, and EPYC faces a steep hill for broad support and adoption of this very solid product offering. The success of EPYC is not just tied to its performance, but the strength of its ecosystem and partners. The following diagram shows the typical route-to-market (RTM) for a CPU.
(Source: Matt Kimball, Moor Insights & Strategy)
This RTM shows many layers of abstraction between a CPU and an IT customer. To further illustrate, consider the following example:
- An IT executive at an e-tailer is looking to deploy a data analytics tool to better understand the purchase intent of site visitors.
- After doing her due diligence, she decides to further explore an analytics tool from a well-known software company.
- Upon deciding to move forward with this analytics tool, the IT executive asks the software vendor for suggestions on deploying; that is, all the hardware and supporting software to achieve best performance.
- The software company initiates an engagement with a preferred hardware partner and a channel partner to support.
- A few months later, the IT executive has helped her business make more informed decisions around marketing campaigns.
It’s all about the ecosystem
The first hurdle AMD faces is in the support from its hardware and software ecosystems. From all appearances, it looks like AMD has done a solid job of securing server platforms with hardware partners such as Hewlett Packard Enterprise and Dell, EMC.
- But what about the software ecosystem?
- Will the OSV and Hypervisor providers enable all the great security features in EPYC?
- Will there be an EPYC based reference architecture for SAP HANA?
- Or a VxBlock offering built upon an EPYC based Dell platform?
Given the performance numbers reported by AMD, it appears EPYC would be a good fit for both solutions. But AMD’s lack of market presence over the last five years will probably cause these vendors to take a “wait and see” approach.
Can I trust you again?
AMD’s exit from the server market in 2012 forced many IT organizations to make significant investments in rebuilding their datacenters on IntelXeon based platforms. And this has led to two additional barriers for AMD.
The first issue AMD faces with IT customers is one of trust. It is one thing for some of the large hyperscale customers say they will look at EPYC as a potential play in their datacenters. But for the IT administrator who resides on Main Street, things are a little more complicated. That IT executive has neither the time nor resources to embrace AMD all over again.
The second issue AMD faces is one of interoperability. It is well known in that a virtualized server environment is either based on Intel or AMD, as cross platform migrations are not supported in the major hypervisors. Given Intel’s dominance of the server market (~99%), AMD will have a longer time creating a relevant footprint in many server rooms and datacenters.
To use a baseball metaphor, AMD hit a solid double with EPYC. And because of its long hitting drought, that double seemed like a home run. It showed impressive numbers against Intel’s (soon to be replaced) Broadwell lineup. And all of the right things were said by an impressive lineup of partners. It will be interesting to see if this rhetoric translates into relevant ecosystem support and customer adoption.
And it will be interesting to see how Intel responds with Skylake.