• United States

AMD plots its move against Intel in the data center

News Analysis
Sep 18, 20184 mins
Data CenterVirtualization

With new momentum and Intel reeling, AMD is targeting virtualization upgrades.

intel vs amd
Credit: Thinkstock

Smelling blood in the water, a revitalized AMD is preparing for a big push against Intel in the data center, hoping to win back the market share it gained and lost a decade ago.

AMD is promoting its Epyc processors, with 16 or 32 cores, as a lower TCO, higher performance option than Intel’s Xeon. It argues a single-socket 32-core server is cheaper up front and in the long run than a dual socket setup, which is Intel’s bread and butter.

“We’re not saying single socket is for everyone, but at the heart of the market is where 50 percent to 80 percent are 32 cores per server and down, and our top single socket can do it more efficiently with lower costs and licensing. But in some cases some people will want to stay at two-socket,” said Glen Keels, director of product and segment marketing for data center products at AMD.

Still lots of virtual machines

AMD cites IDC figures that state six out of 10 two-socket servers that are virtualized should be upgraded in the next 12 months. Servers, AMD noted, are staying deployed longer, going from the usual three years to up to five years. That increases the likelihood of breakage while deployed, necessitating an emergency fix.

When compared against a two-socket Intel Xeon 5118, with 12 cores per CPU, AMD claims 10 one-socket servers are still almost half the three-year TCO as 14 two-socket servers with 64 percent of the licensing costs and 2.8 times the VM density. That’s because VMware charges per socket, Keels said.

Pitching on-premises hardware may seem like an uphill battle, but AMD said its models show server sales will continue out to 2022. “That’s when we’ll cross over 50-50 on-prem/off-prem mark, but customers are still investing heavily in on-prem virtual infrastructure,” said Keels.

What they are tending to not invest in is expanding or upgrading data center facilities. For now, they are making due with the space they have and the power and cooling they have. And when they run out of space, they will move to a hosted or managed environment or public cloud, he added.

Intel price cuts

The pitch is apparently being heard. ServeTheHome, a news and reviews site for server hardware, claims it’s heard from “a number of customers” that Intel is giving significant discounts for its Xeon processors looking at Epyc and in quantities well below 1,000-unit trays, which means it’s not just to its largest customers.

“A key trigger seems to be an organization’s willingness to adopt AMD Epyc. Our advice, if you are buying even as few as 50 to 100 servers, is to get an AMD Epyc system quote from your reseller. Doing so seems to be the trigger for Intel’s discount approvals,” STH writes.

Intel has another reason to be worried. A new vulnerability has emerged called Foreshadow. It’s similar to the Spectre/Meltdown vulnerabilities in that it’s baked into the silicon and the software fix has a significant performance impact.

Intel chips use a feature called Software Guard Extensions, or SGX, which allows programs to establish so-called secure enclaves within Intel processors. These are areas of the chip, in its L1 cache, that are cordoned off to run code that the computer’s operating system can’t access or change.

(You know what’s coming next, right?)

Well, a group of researchers from five academic institutions around the world found a way to bypass SGX’s defenses. They call it Foreshadow. To defeat it, the scheduler in the L1 cache has to be turned off so one VM can’t harvest the contents of another. Doing so has a performance impact of up to 30 percent, and since Epyc doesn’t use SGX, it’s immune.

When in Rome

AMD’s next-generation Epyc processor is codenamed “Rome,” and while AMD is mum on it save for few details, such as a ship date of next year, stats are starting to leak. For starters, the 64-core Rome processor will be built (in a lot of days) on a 7nm manufacturing process. While AMD’s usual chip manufacturing partner GlobalFoundries has given up on getting to 7nm, Taiwanese manufacturing giant TSMC has made it work and will begin producing 7nm Rome processors next year.

And Rome looks like an absolute beast, if leaked benchmarks are any indication. A site called Chip Hell released a benchmark showing a prototype Rome processor performing 10 times faster than a high-end Intel Xeon. However, this is one site, and I don’t know its track record. But it looks like a doosy if true.

Andy Patrizio is a freelance journalist based in southern California who has covered the computer industry for 20 years and has built every x86 PC he’s ever owned, laptops not included.

The opinions expressed in this blog are those of the author and do not necessarily represent those of ITworld, Network World, its parent, subsidiary or affiliated companies.

More from this author