The battle for data center dominance between Intel and AMD is about to get a little hotter.
AMD made the first move Monday with the release of its third-generation Epyc “Milan” server processors, which AMD officials said will offer higher-performing and more efficient alternatives to Intel in key segments such as enterprise, high-performance computing (HPC) and the cloud.
The launch of the Epyc 7003 family of x86-based 7-nanometer processors, which range from eight to 64 cores and are based on AMD’s Zen 3 microarchitecture, comes just weeks before Intel is expected to roll out its “Ice Lake” portfolio, the latest of its 10nm Xeon Scalable processors.
Further reading: Biden Moves to Bolster Chip Industry, But It will Take Time, Money
“Our customers always want more,” AMD President and CEO Lisa Su said during a virtual event introducing the Milan family. “They want more performance, greater scalability and new layers of security. We thought about all of this when creating our next-generation server processors.”
Intel is still the dominant player in the server processor market. According to Mercury Research, in the fourth quarter 2020 Intel held 92.9 percent of the market to AMD’s 7.1 percent. However, that was down from the 95.5 percent Intel had during the same period in 2019, when AMD’s share was 4.5 percent.
New AMD microarchitecture
After more than a decade of floundering with its Opteron server chip offerings, AMD under Su embarked on a mission to create a new microarchitecture. The Zen microarchitecture has since become the foundation for both AMD’s Ryzen client chips and its Epyc portfolio. The first Epyc chips launched in 2017, followed by the second-generation processors in 2019. The Ryzen chips already are using the redesigned Zen 3 architecture, which is aimed at improving performance, latency and security and is now available in the latest family of Epyc server processors.
Among the key innovations in Zen 3 is a change in the layout of the processor’s cores, according to Mark Papermaster, AMD’s executive vice president and CTO.
“The difference is in the core complexes,” Papermaster said. “Zen 2 was organized with two four-core attached complexes that were tightly integrated on the core die. Zen 3 transitions to a single unified eight cores on the die. Every core has direct access to the 32MB of free cache rather than 16MB in Zen 2. Every core can now communicate directly to the cache without traversing across the die, reducing latency.”
The result is server chips that can offer a performance increase of up to 19 percent more instructions per clock over the previous generation of Epyc. Rob Enderle, principal analyst with The Enderle Group, told InternetNews that AMD is making the right moves with Zen and the Epyc chips.
“As the challenging vendor, they need to outperform their competition,“ Enderle said. “On paper, they more than do that, but in straight performance and security features. It is an impressive offering that showcases their focus and dedication, particularly when it comes to cloud workloads.“
For HPC users, Epyc offers more throughput capabilities, enabling more work-like simulations to be done during a given time period and the ability to leverage bigger data sets or more complex computer models.
Forrest Norrod, senior vice president and general manager of AMD’s Datacenter and Embedded Solutions Business Group, also noted that the new Milan chips are being used in developing two pre-exascale systems and will be used in Frontier, the supercomputer being built with Cray systems from Hewlett-Packard Enterprise at the Oak Ridge National Laboratory that will be the country’s first exascale system when it goes online later this year.
“These systems will tackle problems that have hitherto been beyond our reach,” Norrod said. “AMD is leading the journey to exascale with our third-generation Epyc CPUs and [Radeon] Instinct compute-optimized GPUs. We are helping companies and organizations around the world build pre-exascale systems that are laying the foundation for the exascale era.”
The Epyc 7003 chips also deliver high core density and advanced security features – including Secure Encrypted Virtualization-Secure Nested Paging (SEV-SNP) for protecting against malicious hypervisor attacks – which are key for cloud providers, a key customer demographic considering the growing impact hyperscalers and cloud providers have on the data center hardware market as demand for cloud services grow.
Booming cloud and hyperscale market
Analysts with Synergy Research Group said that in the fourth quarter, global enterprise spending on cloud infrastructure services grew by 35 percent year-over-year, to more than $37 billion, and that in the first three quarters of 2020, hyperscale operator capital expenditures surged to $99 billion, a 16 percent jump from 2019.
For enterprises themselves, the processors provide performance and value.
Dan McNamara, senior vice president and general manager of AMD’s server business, said enterprises are being squeezed by demands to run more demanding workloads, to improve operational efficiencies and be adaptable. He added that “on top of all these challenges, there are ever-increasing security threats that need to be defended against. With this environment in mind, we developed Epyc to help enterprises overcome these challenges. [Epcy] is the industry’s highest-performance server CPU, and that translates to lower total cost of ownership, helping our customers do more with less.”
With the launch of the Epyc 7003 chips, AMD also noted that almost a dozen system OEMs and cloud providers are on board. Cloud providers like Google, Microsoft, Amazon Web Services, Oracle and Tencent said they are releasing or plan to roll out new instances leveraging the processors. In addition, systems makers like HPE, Dell, Lenovo and Cisco also outlined plans for using the Milan chips in new or updated systems.
ARM and IBM Power also challenge Intel
How the launch of the latest AMD and Intel server chips impacts the market remains to be seen. The data center space is getting a lot of attention. IBM is continuing to innovate around its Power architecture, and chip makers leveraging ARM designs are beginning to see some traction with server OEMs. Nvidia could ramp that up more if it is allowed to follow through on its bid to buy ARM for $40 billion.
However, taking market share away from Intel won’t be easy. The company has struggled at times over the past few years in its manufacturing efforts, which has led to product delays – particularly as it pushes to move beyond a 10nm manufacturing process and into 7nm and then 5nm – but it still brings in a lot of money and has deep innovation resources.
That said, the new Epyc chips will help AMD improve its market share. Having the latest editions of both Epyc and Xeon Scalable chips coming to market could make for interesting competition, Enderle said.
“The timing couldn’t be better, hitting just as the world seems to be ramping up again as it recovers from the pandemic,” he said. “This timing could allow AMD to gain more share than otherwise would have been the case thanks to the drag the pandemic placed on on-premises sales. Intel is getting its game back, though, and Ice Lake looks strong as well. This could be a fight for the record books once Intel gets their latest to market.”
Supply constraints could influence market share too, Enderle noted. “Intel is still recovering, but both of these efforts are solid offerings, and permanent share changes may have more to do with availability, which has been iffy, than any other factor.”