AMD’s Andrew Feldman shares why working with Intel was challenging and why AMD wasn’t a loser in a recent major SeaMicro deal in an interview with VR-Zone.
When the microserver category first arrived on the market, there were those who doubted it would go anywhere. Upstart SeaMicro, founded by now-AMD server manager Andrew Feldman, encountered resistance from one of the dominant players in the processor business when he approached it some time ago to buy chips.
“Intel told us we couldn’t use Atoms in servers, and nobody wanted it. They wouldn’t work with us,” he told VR-Zone at AMD’s recent #APU13 conference. “Two years later they’re saying its 15 percent of the market. We sort of invented the category of microserver.”
Microservers, what SeaMicro specializes in, aren’t like traditional servers as they are designed to handle high-volume tasks that require little compute power or enterprise scenarios that don’t require a full scale server. They can be deployed to supplement existing rack servers for load balancing, or used as a cheap alternative in places like developing markets. Microservers are usually the domain of low-powered chips like Atom or ARM, not the mighty Xeon.
Though Intel initially didn’t think it could happen, SeaMicro eventually went on to offer Atom parts in its servers. But this wasn’t easy. Even after convincing Intel an Atom-based microserver was possible, SeaMicro had to contend with the fact the original Atom wasn’t quite good enough for them.
“Intel’s parts weren’t getting the job done. The original Atom is a 32-bit part with two gigs of DRAM. [The chip itself] was a weak part that didn’t do its job. It was supposed to win the iPad and it lost. It wasn’t good enough,” Feldman said. “We worked with them, and we worked with them, and we basically told them what should be in what is now Avoton and what oughta be in Centerton.”
In early 2012 both Intel and AMD took interest in SeaMicro and launched bids for the company. AMD eventually had the winning bid with $334 million in cash and stock options, placing Feldman as the new head of AMD’s server business unit.
AMD was in a challenging place now: one of its subsidiaries made money selling servers with Xeons and Atoms. Granted, AMD would profit from every server sold, but regardless SeaMicro was still selling chips from one of AMD’s competitors.
This relationship led to a report in October from ExtremeTech that a deal Verizon had struck with SeaMicro to provide it with servers for its new cloud platform was actually a “massive win” for Intel.
“The vast majority of the CPUs (more than three quarters) AMD is using for this implementation with Verizon are Intel Xeon E3s using the SeaMicro fabric they acquired. So in essence AMD is now becoming a microserver OEM for Intel and is still largely selling Intel parts in microservers for their customers,” ExtremeTech’s Sebastian Anthony reported.
Certainly, if true, this would be embarrassing for AMD that its own Opteron platform was passed over in favor of the Xeon E3.
But Feldman says this isn’t exactly the case.
“ExtremeTech is wrong. I didn’t speak to ExtremeTech. They didn’t ask me,” he said.
Feldman explains that before SeaMicro was acquired by AMD, and before Verizon’s cloud deal had been announced, it had sold Verizon Xeons. But things changed after the acquisition and once Opteron-based systems came online.
“After the acquisition, and once we built an Opteron-based system, all forward-going orders have been on Opterons,” Feldman said.
For the Verizon cloud deal, Feldman said that Intel simply couldn’t meet Verizon’s requirements.
“They didn’t have a part that met the requirement, is the truth,” he said. “The requirement was 64 gigabytes of DRAM on a single socket part and Intel didn’t have a part. We had a part.”
“It was a big loss for Intel.”