AMD announced a double whammy at its Next Horizon event in San Francisco: first by announcing EPYC processor-based instances were now available on AWS, and second by undercutting Intel Xeon-based instances by 10%.

“Our goal with EPYC is to reach the largest number of users possible, and the cloud is the tip of that sphere,” said AMD president and CEO, Dr Lisa Su. “To do that, we must partner with the number one cloud service provider in the world.”

That partner is of course Amazon, with the company’s vice president of computing services, Matt Garman, joining Dr Su on stage at Next Horizon.

“When we think about our compute platforms, there are a couple of things that are important to our customers,” said Garman. “The very first thing is security, reliability and performance, and AMD is great on all of those fronts.

“The second thing that many customers come to AWS for is choice,” he added. “We want to be the compute platform for the world, we want to support every single workload out there, and for that our customers need a broad range of capability.”

But this was all a warm-up to AMD’s big advantage over Intel: value for money. “But the final thing is frankly cost. I talk to a lot of customers, and never have I talked to a customer who wasn’t interested in lower prices.”

“I feel the same way,” quipped Dr Su.

In practice, that means AWS customers in a number of regions, including Europe, can switch to AMD-based R5 sand M5 instances via the AWS Management Console or AWS Command Line Interface.

Amazon says T3 instances will be available “in a couple of weeks”.

“One of the cool things about the cloud is that you can go and get them today,” said Garman. “It’s really easy. It’s a simple API call to launch an instance. You can use the exact same instances you use today, you can use the exact same scripts.

“And one of the great things for our customers is that simply by making that change and switching, they can instantly save 10% on their compute costs.”

AMD president and CEO, Dr Lisa Su, unveiled the company’s next-generation EPYC server processor in San Francisco.

Dubbed “Rome”, it brings 7nm manufacturing to the server CPU market for the first time, and is the harbinger of future 7nm Ryzen chips.

At the same time, AMD launched the Radeon Instinct MI60 and MI50 graphics accelerators, which also sport 7nm-based chips and use AMD’s latest Vega architecture. 

“We’re taking the power of 7nm tech, we’re taking that advantage of doubling that bandwidth and we’re bringing tremendous performance,” said Dr Su at the launch event. “We’re thinking not just about the CPU but about how the CPU connects to the rest of the components.”

This, emphasised Dr Su, was just as important as the processor itself. “What Rome does is brings total system capability when you put [the CPU, GPU and interconnects] together.”

The new Rome EPYC processors are built on top of AMD’s “Zen 2” architecture, with the processor’s core complex containing eight core dies, and each of those core dies has eight Zen 2 cores. That translates to 64 cores on the top-end version of Rome chips.

For the moment, however, AMD is keeping quiet on processor clock speeds.

What should this mean in practice? AMD claims Rome will deliver an unprecedented 3x speed increase compared to the previous generation processors in certain applications. “I’ve been in the semiconductor industry for a long time,” said Dr Su, “you don’t get 3x.”

The company promises an even bigger boost for floating point performance, with speeds four times that of the current EPYC processors.

While demos at launches should always be treated with scepticism, AMD was happy to pit a top-of-the-range dual socket Xeon against a single-socket Rome processor and a current dual-socket EPYC system.

Putting all three systems through the C-Ray benchmark side-by-side, the Rome-based system completed the test in 27.7 seconds compared to 30.5 seconds for Intel’s machine and 28.4 seconds for the dual-socket EPYC system.

AMD is already shipping the Radeon Instinct MI60 graphics accelerators to data centre customers, but would not be drawn on price.

The company also claims that Rome has started sampling with customers but has not announced a price or availability other than saying “next year”.