Everything looked normal. An audience of company executives and employees, journalists, and analysts staring at the brightly lit stage as the chief executive officer strode from one side to the other boasting of a chip’s performance and power efficiency, the advanced technologies that went into making it, and the marked advantages it held over the best the competition had to offer.
But this was not an Intel CPU announcement, and it was Lisa Su, chief executive officer at AMD, that was doing the walking and the talking, not Intel’s Pat Gelsinger, as she introduced the “Genoa” Epyc 9004 server processors.
The Genoa chips have up to 96 cores, 192 threads, 384 MB of L3 cache, and up to 6 TB of main memory per socket, and are etched with a mix of 5 nanometer (the cores) and 6 nanometer (the I/O and memory die) processes from Taiwan Semiconductor Manufacturing Co. The chips sport DDR5 memory, PCI-Express 5.0 peripheral controllers, which in turn support the CXL 1.1+ protocol for memory extension.
(You can get our deep dive into the Genoa lineup here.)
All of this not only exceeds what AMD has been able to do for the past year with its Gen 3 Epyc chips, but far outdistances what Intel is able to do with its Xeon SPs, something that Su and other executives were quick to point out throughout the presentation without ever uttering Intel’s name.
“It’s the highest performance datacenter processor, it’s the most efficient, and we’re delivering significantly better performance-per-watt than our competition,” Su said. “What that means for enterprises and for cloud datacenters it translates into lower CapEx, lower OpEx, and lower total cost of ownership.”
The Genoa portfolio comes five years after AMD launched the first generation built on AMD’s new Zen microarchitecture. The company had been lost in the wilderness for almost a decade after fielding some pretty good Opteron processors in the early to middle 2000s. After that, Intel was by far the dominant player in the datacenter chip space.
AMD spent years developing the Zen cores and debuted Epyc server chips and the Ryzen PC chips based on them with hopes of chipping away at Intel’s massive market share. The company had to prove to the industry over multiple generations that not only could it deliver the performance and power efficiency needed for enterprises, cloud providers, and the HPC space, but also that it could deliver this on time and with a roadmap that laid out the future.
Under Su’s leadership, the company has done just that, aided in part by corresponding stumbles by Intel in everything from shrinking the manufacturing process to rolling out new chips on time. It’s oft-delayed “Sapphire Rapids” next generation of Xeon SP processors is now is not expected until in January.
With all that, AMD now owns more than a quarter of server CPU shipments.
Meanwhile, the new Epyc 9004 – based on the Zen 4 architecture – comes after recent releases of other AMD chips, including the Ryzen 7000 client processors, RDNA3 GPUs based on its Radeon 7900 series, Zen 4C aimed at the cloud, and XDNA adaptive SoCs. Today was the datacenter’s turn.
“We’ve said the datacenter represents our largest growth opportunity and the number-one strategic priority for our company,” Su said. “When you think about the modern datacenter, customers need the highest performance compute engines actually across the board. In addition to our leadership Epyc processors, we also offer a very full portfolio, including our Instinct GPU accelerators built for our HPC and AI, our leadership FPGAs and adaptive SoCs through our acquisition of Xilinx and our leadership DPUs from our acquisition of Pensando. If you look at all of this, this is really the broadest datacenter portfolio in the industry.”
That extends into HPC as well, another sector that Intel for a long time ruled. AMD’s Epyc processors are in five of top ten fastest supercomputers (as listed on the Top500 list), including Frontier, the first exascale computer built with Hewlett Packard Enterprise’s hardware (acquired via its $1.3 billion acquisition of Cray in 2019). It also is in eight of the 10 most power-efficient supercomputers, with Frontier topping that list, she said.
For those cloud environments, Su noted that the dual-socket Gen 3 Epyc processor was 40 percent faster than Intel’s best Xeon SP Platinum processor and that Genoa increases that performance advantage to three times better.
“What that means is the core density advantage of 4th Gen Epyc allows our cloud service providers to support more than double the number of instances for servers,” she said. “What that means for our end customers is that each of them gets to experience these instances with much higher performance.”
Su also noted tests run on the latest Epyc chips for measuring the power efficiency of the performance.
“For cloud datacenters, efficiency has become equally important as pure performance,” she said. “Our 4th Gen Epyc delivers 2.6 times the energy efficiency of the competition. What this means is that choosing AMD Epyc CPUs in the cloud translates to lower energy usage and lower energy costs. A number of our cloud providers are actually telling us that power is becoming a significant limit for what they can install on their datacenters, so we believe that this is really differentiators for us.
During the event, AMD also was bolstered by support by the likes of HPE, Oracle Cloud, Google Cloud, and Microsoft Azure, all of whom laid out plans for bringing the Genoa Epyc chip to their environments, both for on-prem systems (from HPE) as well as new cloud instances to take advantage of the higher performance and efficiency.
Clay Magouyrk, executive vice president of Oracle Cloud Infrastructure, told Su that the conversations he has had with organizations has evolved in recent years.
“If I go back to where we were five or six years ago, a lot of the conversations I was having with customers was kind of around, ‘Why would I choose AMD?’” Magouyrk said. “That conversation has really shifted to where now the conversation starts with, ‘Why not AMD?’ And we’re seeing that across the board. Whether it’s peak performance workloads or general-purpose workloads, people are choosing AMD in general.”
He pointed to Red Bull Powertrains, which manufactures power units for the Red Bull Formula 1 racing team and which is building its own power plant to meet upcoming regulations. And Red Bull Powertrains is doing the work on Oracle Cloud Infrastructure – using Genoa processors already installed.