AMD this week officially introduced its first new server processor in five years and promised that the new line would have the market potential, and the product roadmap, that its older generation of chips lacked.
AMD's new EPYC line of x86-compatible processors, expected to be available in June, offer stronger performance than Intel's latest Broadwell-based processors and the ability to bring significant total cost of ownership savings, said Scott Aylor, corporate vice president and general manager for enterprise solutions at the Sunnyvale, Calif.-based company.
The EPYC processors signal a new phase for AMD's data center ambitions. "The last time AMD brought a server processor to market was five years ago," Aylor said. "At the time we were in a declining trend in our competitiveness. However, four years ago, we got a new CEO and CTO who made the painful decision to stop producing processors based on that technology."
AMD focused on the development of a new core design it called Zen, and that concentration has paid off, Aylor said. The Zen architecture offers a 40-percent improvement in the number of instructions it can handle per clock cycle over previous AMD architectures.
The Zen architecture is base on which both the AMD Ryzen commercial desktop processor, introduced earlier this year, and the EPYC server processor is built, Aylor said.
"We've seen a 50-percent improvement in desktop processors with Ryzen, and with EPYC we're on par or better with that improvement, with the potential to exceed the performance of the highest-end Intel x86 processors," he said.
The Zen architecture gives EPYC eight memory controllers per processor to provide the balance needed to manage increasingly-large data sets. It also provides 128 I/O lanes per processor vs. the 40 lanes typical of server processors, Aylor said.
"This lets server makers connect more storage, GPUs, and NICs (network interface cards) directly to the processor," he said.
All told, the innovations combine to provide performance that AMD this week demonstrated to be 1.4-times that of the highest-performing Intel Broadwell processors, Aylor said.
"For us to go from not having a large share in the server market to performance leadership is quite an accomplishment," Aylor said.
An Intel spokesperson told CRN via email, "We take all competition seriously, but it would be premature to comment on claims as these processors are not yet available." According to published reports, Intel is expected to release the latest version of its Skylake processors later this year.
Given EPYC's advertised performance, one area where AMD hopes to see initial success is in the entry-level server market. Aylor said that single-socket and dual-socket servers now account for over 90 percent of the servers sold, with the vast majority being dual-socket models.
"Dual-socket servers are often the only option for performance and features," he said. "With EPYC, we have enough performance and I/O lanes to corner the bottom half of the server market with single-socket solutions. That's a TCO (total cost of ownership) advantage of over 20 percent, with no compromise in performance or features. And in this market, a 20-percent TOC advantage is very significant."
One way AMD hopes to bring server vendors back to the AMD processor fold is to assure them the company has a solid roadmap for the future, Aylor said.
While he was unable to provide much detail yet, he said that the "Naples" version of EPYC, which features 14-nanometer production technology, will be followed by "Rome with the "Zen 2" architecture and 7-nanometer technology and "Zen 3" with higher-density 7-nanometer technology through 2020.
Before it stepped away from the server market, AMD had introduced a number of innovations that eventually became common in processors, including the first 64-bit architecture, multi-core processing, and integrated memory controllers, he said.
Custom system builders are looking carefully at AMD's re-entry into the server market that it left despite technology breakthroughs in the past.
AMD's Ryzen desktop processor has proven to offer very high performance, and word from OEMs like server maker Supermicro is that the EPYC is already performing very well, said Andrew Kretzer, director of sales and marketing at Bold Data Technology, a Fremont, Calif.-based custom system builder.
Performance, however, is not everything given how difficult it is to introduce a new server platform to customers, Kretzer told CRN.
"Enterprise people tend to be entrenched with current technology," Kretzer said. "Switching them to something new can be hard. But we could see some of the larger data center companies – Amazon, Google, Facebook – with their own server initiatives adopt EPYC if the price, performance, and efficiency is there. As a system integrator, we'll try, but it will be a tough row to how to win back AMD mindshare."
Another custom system builder told CRN on condition of anonymity that there is a good chance AMD can gain market share.
"Unlike with the Opteron, I'm seeing more ODMs and hardware manufacturers planning servers around EPYC," the system builder said. "Many of these companies have at least two servers in development using EPYC, and some have more. I'm seeing server manufacturers embrace EPYC more than they did Opteron."
One advantage AMD could have over Intel is price, as Intel's top-of-the-line Xeon processors have a street price of up to $13,000 to $14,000 each, the system builder said. Intel declined to confirm or deny such a price.
However, AMD will have one stumbling block in its processor roll-out. "The name is weird," the system builder said. "Who the hell can pronounce it?"
AMD is not yet disclosing which server partners are working with EPYC, and expects to provide more information when it releases its EPYC processors in late June, Aylor said. He said that AMD has already shipped over 5,000 EPYC processors to partners and developers, and that there are about 30 single-socket and dual-socket server platforms in development.