AMD/SeaMicro - Microserver Strategy
SeaMicro was recently acquired by AMD and has a unique position in the industry.
AMD does not have a dedicated microserver processor at this point, but has bought itself a leading vendor with a few key patents in the microserver market that could show the path how this market could evolve over time. Traditionally, neither AMD nor Intel has done well with acquisitions and products that were targeted to reach through their customer chain and offer products to the end-user directly.
With Sea Micro, AMD is competing with potential customers of microserver processors.
The scenario is even more interesting since SeaMicro founder Andrew Feldman, today corporate vice president and general manager of AMD’s Data Center Server Solutions, told us that there is no intention to drop Intel processors from the SeaMicro product line anytime soon. As a vendor, he told us, he would be required to offer what his customers are asking for, even if it means that he has to build rival processors into his servers. At this time, customers are asking for Intel processors (Sea Micro recently launched an Opteron-based server as well).
Feldman has been rather successful in the modern microserver field and even says that Intel told him that he would not be able to build servers based on Atom processors. He ran with this anyway, but did not believe that the simple integration of an Atom processor into a server would work out well in the long term.
To make his company stand out, he invested in research and development and came up with three distinct technologies - I/O virtualization that allowed Seamicro to remove components from a motherboard; Turn-it-Off that allows Seamicro to switch off unused CPU components, as well as a compute fabric to link its custom motherboards via low-power connections - to create the Sea Micro microserver value proposition.
It is this innovation that Feldman believes will help Sea Micro compete with much larger vendors down the road; Credit card-sized motherboards carrying small processors are integrated and directly connected an extremely dense space, while traditional implementations require slower Ethernet connections, more space and more power.
His illustration of the opportunity for microservers resembles driving the right car at the right time. "If you have to move apples, you can buy large trucks when gas is cheap," he said.
But in some scenarios, he explained, small cars may be more efficient moving smaller numbers of apples since those smaller cars will use less gas. Sure, those "smaller cars" could also be achieved by slicing a Xeon processor into smaller portions via virtualization, but Feldman believes that this approach is inefficient in applications where microservers can excel due to their greater density."
We can't do CAD-CAM and fluid dynamics calculations, because microservers are not built for applications that take advantage of huge single thread performance." Instead, Feldman sees microservers do well in the delivery of cloud and web applications. In this specific area, a reduction of power and space consumption can deliver meaningful business value.
For AMD, Sea Micro's greatest benefit is the experience in integrating processor technologies into a cohesive architecture, which provides AMD with a considerable advantage over its rivals, should AMD be working on a processor dedicated to the use in microservers. Integration and customer exposure is something that Intel has not done - and has proven to be extremely valuable to score customer wins.
However, Feldman relies on Intel's Atom, Xeon and AMD's Opteron processor for now and said that he is keeping an eye out for the upcoming 64-bit ARM processor: "We have not seen it, but it could be an interesting product for throughput workloads." He considers the ARM architecture as a potential threat for x86 chips as "hundreds of ARM licensees" could provide customized parts Intel cannot deliver, while he expects Intel to keep its edge in manufacturing technology.
The strength of his parent, AMD, lies in its ability as a fabless company "to innovate on different dimensions, and on the chip as well."
Feldman agrees that there is room for different types of server processors, other than the current Xeons and Opterons. "We need to think about the server processor differently," he said. "Historically, a server CPU has been a desktop CPU on steroids, but we need to think carefully about what people are doing and what they need in their data centers. There will be fewer servers in the future that are dependent on floating point performance."