Partners: AMD EPYC Milan Widens Gap Over Intel, Builds Trust

The improved capabilities of AMD’s new EPYC processors are creating more competitive pressure for Intel while proving that the chipmaker can be a reliable server vendor, partners tell CRN. ‘The news just keeps hitting that AMD is executing well and so that’s resulting in increased customer confidence,’ one partner says.


Partners said AMD’s new third-generation EPYC server processors are widening the chipmaker’s performance gap over Intel’s Xeon Scalable chips and building trust among customers in the company’s ability to execute on product plans.

The new 7-nanometer processors, code-named Milan, may not be as game-changing, partners said, as the previous generation—which doubled EPYC’s maximum core counts to 64 and introduced industry-first PCIe 4.0 support—but they still provide a meaningful boost in performance, efficiency and features that will keep the pressure on server processor juggernaut Intel.

[Related: Intel CEO Pat Gelsinger To Discuss Comeback Plan In Live Webcast]

Sponsored post

“If you ask AMD, they’ll say it’s monumental to them because they’re so much further ahead than Intel,” said Alexey Stolyar, CTO of International Computer Concepts, a Northbrook, Ill.-based server integrator. “From the perspective of technology, it’s similar technology to [second-generation EPYC] Rome, but it’s a lot more efficient and better. But the gap [between AMD and Intel], I think, is the big picture.”

While Intel has suffered from delays of its 10nm processors over the past several years while recently delaying its 7nm products by six months, AMD has largely stayed true to its product road map since it launched its first processors based on the Zen architecture in 2017, according to Stolyar. That’s a major change from AMD’s previous attempts in the data center.

“One of their big messages is: ‘We weren’t very good about staying true to our road map [in the past]; it was all over the place. Well, now we have a simple, concise, aggressive road map, and we’re staying true to it every single time,’ so it’s really building that confidence with customers,” he said.

An Intel spokesperson said the company’s chief strength in the data center market is providing “overall system value.”

“Customers’ buying decisions are based on the ability to deliver overall system value. That is where we are focused and why customers continue to choose Intel,” the spokesperson said. “Our unmatched portfolio of hardware and software solutions deliver leading workload acceleration for critical applications, while reducing deployment complexities and total cost of ownership.”

Stolyar said one improvement with the Milan processors that is opening up more opportunities with International Computer Concepts’ financial customers is the redesigned core cache complex, which has doubled the L3 cache per chiplet to 32 MB and the number of cores sharing the L3 cache to a maximum of eight cores from the previous generation. This allows the cores to communicate with each other faster, making the processors better for low-latency applications, which was previously an area of weakness for AMD.

“The big difference is that applications that need eight cores now can all be within one [core cache complex], and that is really big because in the financial world, eight [cores] is that magic number to start off with,” said Stolyar, who added that the larger cache also helps.

Stolyar, whose company saw AMD-based servers reach roughly half of system sales last year, said he believes the improvements with Milan will allow the chipmaker to continue to increase market share.

“I think they’re going to continue to dominate. They’re going to continue to win more and more business,” he said. “Milan opens up more doors for them with this architecture, so they’re going to win more opportunities with it for sure.”

But if AMD’s supply issues impact the company’s ability to ship Milan processors, it could dampen the chipmaker’s momentum, Stolyar added.

“If they have a great product but can‘t ship it, it’s going to be hard,” he said.

Marc Fertik, vice president of technology solutions at Elk Grove Village, Ill.-based Ace Computers, agreed that the new Milan processors increase the advantage AMD has over Intel while proving that the chipmaker can be trusted to execute on its road map. He added that Milan is receiving wider support in the hardware ecosystem than previous EPYC generations, with AMD expecting more than 100 server platforms to support the new processors this year.

“There’s definitely more support for AMD in the industry than there has been,” he said. “They’ve done a good job at growing their hardware partners.”

Supermicro, in particular, is delivering a broader set of server products supporting Milan than previous EPYC generations, according to Fertik.

“They’re bringing out some newer products that have not previously been available to take advantage, so they’re giving more flexibility in terms of network and GPU support,” he said.

With Milan delivering the same high core counts as the previous EPYC generation, Fertik said, AMD continues to make a good argument for customers to switch to EPYC-based single-socket servers, which can provide more cores than a dual-socket server running Intel’s second-generation Xeon Scalable CPUs. What also sells the single-socket proposition is Milan’s 128 lanes of PCIe 4.0 connectivity.

“A lot of people are willing to look at single-socket, which means that you can use a very high-wattage CPU with a lot of I/O support,” said Fertik, who added that Supermicro is offering a lot of options for single-socket EPYC servers.

One area where the Milan processors fall short, according to Fertik, is their lack of a virtual RAID component, which means customers need to spend extra for a RAID controller if they want to set up multiple NVMe drives on a single array. Intel, on the other hand, has a virtual RAID capability with its Xeon Scalable processors, which makes them more attractive for such a use case, he added.

“This is why if somebody just wants a plain server with a couple of NVMe SSDs, high speed, in a RAID 1 mirror, they’re going to stick with Intel because it’s cheap,” Fertik said. “But I don’t think AMD is playing a cheap game. I think they’re playing the feature game.”

Dominic Daninger, vice president of engineering at Nor-Tech, a Burnsville, Minn.-based HPC system integrator, said one feature in Milan that could potentially play well with midmarket customers is the new six memory channel option that isn’t available on previous EPYC generations. Previously, EPYC processors only offered four- or eight-channel memory configurations, but that changed with the new Milan processors, which can help save on memory costs thanks to the new six-channel option.

“It gives them a little more versatility for certain workloads,” he said.

Daninger said AMD has made gains in single-threaded performance with Milan, which is an “area where Intel has beat them previously.”

Like Stolyar and Fertik, Daninger credited AMD for sticking to its product road map.

“It gives people the confidence that AMD is hitting their road map, and it looks like they’re going to be around for a while,” he said.

Eliot Eshelman, vice president of strategic accounts and HPC initiatives at Microway, a Plymouth, Mass., HPC system integrator, said while his company hasn’t done exhaustive testing of the new Milan processors, early test results seem to confirm AMD’s performance claims.

“We have run enough benchmarks already to see some of what AMD is promising,” he said.

Whereas AMD already had an advantage for HPC applications that require a lot of parallel computing because of EPYC’s high core counts, the new Milan chips are giving the chipmaker new advances in lightly threaded HPC applications like engineering simulation programs, according to Eshelman.

“Intel has had that advantage of maximum single-core performance. But with Milan, AMD is really turning that around, and we’re starting to see those results,” he said.

Eshelman said the promising results for Milan’s single-threaded performance were based on tests with a high-core-count model and not AMD’s new frequency-optimized processors, which are designed to deliver higher performance for single-threaded applications.

“But even the high-core-count model is outperforming some of Intel’s frequency-optimized models, so it’s a good sign,” he said. “But we have to get our hands on the full set of SKUs to fully see that borne out.”

The challenge AMD will continue to face, according to Eshelman, is that some customers would still rather work with Intel because they have had a much longer relationship with the company.

“It’s easy to be comfortable with Intel,” he said.

However, Eshelman added, AMD is gaining more favor among customers because of its solid track record over the last few years.

“Anecdotally, there’s definitely more comfort,” he said. “The news just keeps hitting that AMD is executing well, and so that’s resulting in increased customer confidence.”

William Wu, vice president of hardware products at Penguin Computing, a Fremont, Calif.-based HPC system builder, said one thing partners will need to help customers determine is whether they should go with AMD’s new EPYC Milan processors or wait for Intel’s forthcoming third-generation Xeon Scalable processors, code-named Ice Lake, for new servers.

Milan will be a much easier sell for customers who are already using AMD’s previous-generation EPYC processors, code-named Rome, according to Wu, in part because Milan chips are compatible with Rome server platforms but also because of their familiarity with the processor line.

For customers who still rely on Intel-based systems, however, it will be a harder decision, Wu said.

“Our existing customer base that does more Intel with [second-generation Xeon Scalable], it does put a little bit of pressure [on them], either from their business, their market [or] their industry to adopt cutting-edge technology,” he said.

The real “game-changer,” according to Wu, will be when Intel releases Sapphire Rapids, its next generation of Xeon Scalable after Ice Lake.

The bigger questions facing partners now are whether Intel will be able to deliver a meaningful response to AMD’s Milan chips with Ice Lake and whether the chipmaker can get its execution mojo back—something that new Intel CEO Pat Gelsinger has vowed to restore.

“When EPYC first came out, [AMD] didn’t have much of a road map to speak of, but now [that] they have a road map and a cadence developed, it has helped a lot with confidence,” said an executive at a system integrator, who asked to not be identified. “Actually, now I think people are losing confidence in Intel—for the first time in a very long time.”