Now that AMD has confirmed its purchase of low-power server maker SeaMicro, I think its next move will be an announcement around licensing the ARM architecture. That’s right: Intel’s long-abused x86 underdog should do a deal with Intel’s newest rival— the company that provides the architecture for chips inside your cell phone.
With the SeaMicro purchase, AMD gets a company that has fundamentally redesigned the server to make it dense, low-power and able to handle a variety of different types of CPUs. SeaMicro’s CEO Andrew Feldman has told me on many occasions that the company’s box is flexible enough to handle a variety of different CPUs. Its real value is in the specialized chip that manages the communications between the hundreds of processors inside the SeaMicro server — its so-called fabric. And that fabric works with any chip, as Feldman and AMD executives said over and over again on the conference call discussing the deal.
By the end of the year, AMD plans to offer SeaMicro’s boxes using AMD’s Opteron chips as opposed to the Intel silicon currently in the boxes. But because AMD doesn’t want to hack off its server-making customers at Dell, HP, etc., it will likely offer its customers access to SeaMicro’s “technology building blocks,” which can be interpreted to access SeaMicro’s fabric.
However, the far more exciting story is how the SeaMicro buy plays into AMD’s heterogeneous computing strategy. AMD calls this its “ambidextrous strategy.” It clearly sees that the future in the cloud and webscale computing world is in providing the right kind of compute for the workload, a task that companies such as Calxeda, Tilera and SeaMicro were all attempting to do in varying ways. If AMD is really serious about this ambidextrous effort, it doesn’t make sense for it to focus solely on x86 chips. Lisa Su, SVP and general manager of Global Business at AMD, agrees. When asked about a potential ARM license in an interview, she said, “We’re not going to be religious on architecture.”
And since Marvell, Calxeda and others are trying out ARM chips in servers at the behest of customers pursuing better efficiency (and SeaMicro can adapt its fabric to that architecture), AMD should quit dithering and call up ARM to get a license.
Pat Moorhead, a former AMD executive and now president at Moor Insights & Strategy, said the SeaMicro deal is a “good move for AMD and gives them instantaneous access to the microserver market which at its worst would be 20-10 percent of the market.” Far more exciting from his perspective is that it underscores and gives AMD the technical pieces to build out its ambidextrous computing strategy. A strategy that should involve an ARM license.
AMD actually had an ARM license when it purchased ATI, but it sold those businesses to Qualcomm and Broadcom. In fact, the Adreno graphics core inside the Snapdragon mobile application processor is former AMD technology by way of its ATI acquisition. If we dig further back into the previous decade, AMD also had a MIPs-based line of processors. So AMD is no stranger to the non-x86 chip world.
And apparently other ARM-based server vendors would welcome the addition of AMD to the ARM server business. Barry Evans, the CEO of Calxeda emailed the following statement:
We are seeing accelerating market momentum for ARM-based solutions in the data center where an entirely new class of applications such as Big Data, Analytics, cloud and web demand drastic improvements in power efficiency. AMD’s announcement demonstrates that they realize that this new epoch requires a different approach.
So unless AMD wants to buy Calxeda (which is located in Austin, Texas, where AMD has a huge presence) maybe it should call up Sir Warren East and talk licensing terms. Plus, if AMD does license the ARM core, there’s no reason it has to limit its application to servers. Perhaps it can develop processors for the mobile market, where all the action is — a market it has failed to develop for in the last five years.