AMD looks to continue momentum vs Intel with new 3rd Gen EPYC ‘Milan’

AMD’s 2nd generation EPYC chips, Rome, made AMD’s big push in the data centre, and now with more third-party server vendors on board than ever before, they are looking to extend that edge with Milan.

It wasn’t THAT long ago that Intel had basically run AMD and its Opteron server chips out of the data centre. AMD’s comeback began with the first chip of its EPYC family, Naples in 2017, which featured AMD’s new Zen architecture, which had the ability to process many more instructions per clock cycle than Opteron. While the exact metrics depended on what you chose to measure, it’s safe to say that Naples got AMD back in the data centre conversation. Rome, the second generation EPYC chip from 2019, changed the numbers for the first time to be clearly in AMD’s favor. Now with the 3rd generation chip, Milan, AMD is looking to extend that lead, with enhancements which include 19% more instructions per clock.

“Four years ago, [AMD CEO] Lisa Su laid out a vision for the data centre,” said Dan McNamara. SVP and GM of the AMD Server Business Unit, who came to AMD from Intel in January 2020. “This is the latest step to drive continuous performance and TCO benefits, with more focus this time on the ecosystem, and its value-add.

“Naples, in June 2017, marked our re-entry into the market,” McNamara said. “Rome, in August 2019, was a giant leap forward. It has been a very nice run for Rome. Now our third generation EPYC, Milan, delivers yet another improvement over Rome, as well as the next generation of security enhancements.”

McNamara highlighted the greater focus being played in the role of strategic partners in this launch compared to the past.

“We have a partner portal with over 20 solutions,” he said. “We are honing in with solution sets on top of us, so customers can accelerate the value we are bringing. We feel we have a well-rounded story here.”

“Our goal is to deliver the best data centre performance,” said Ram Peddibhotla. Corporate VP for EPYC Product Management. “The second generation EPYC was the transformational product, that set the bar for everything else in the data centre. Today, with Milan, we extend that leadership even further. Our two key messages are first time, with AMD being a metronome for the industry, and performance, where we are consistently driving the envelope higher.”

Peddibhotla said that the Milan chip is 70% faster than the competition. Other core metrics include increasing transactional database processing by up to 19%, and improving Hadoop big data analytic sorts by up to 60% with 61% better price to performance than the competition.

Peddibhotla also emphasized that unlike Intel, which will require new server platforms with their upcoming processors, no such upgrade is needed with AMD.

“Our platforms are fully compatible, and just take a BIOS upgrade,” he said. “That’s what enabled us to accelerate to Milan faster than with Rome.”

The Milan chips also offer increased flexibility in memory options.

“The second generation EPYCs had two options,” said Noah Beck. SOC Silicon Design Fellow at AMD. “You could populate all 8 memory channels, for mamimum performance, or 4 channels, for cost optimization. What’s new is a 6 channel option that sits in between. It gives more flexibility in optimizing the cost of memory configurations.”

The security enhancements include a new feature within AMD Infinity Guard, Secure Encrypted Virtualization-Secure Nested Paging (SEV-SNP). SEV-SNP expands the existing SEV features on EPYC processors.

“Secure Nested Paging protects against attacks by an untrusted hypervisor, by adding interrupt restriction to SEV-E5 to prevent a malicious hypervisor injecting into a guest,” said Mike Clark, Lead Zen Architect at AMD.

Eighteen vendors and their server models are now listed on AMD’s Launch Partner Page. They are: Atos; AWS; Cisco, Dell Technologies; Equinix; Gigabyte; Google Cloud; HPE; IBM Cloud; Lenovo; Microsoft Azure; Nutanix; Oracle; QCT; Supermicro; Tencent Cloud; VMware and Xilinx. AMD says 100  new OEM platforms are expected by the end of 2021.