AMD is targeting Intel's hyperconverged infrastructure clout in the enterprise with new high-frequency EPYC Rome processors that deliver better per-core performance.
The Santa Clara, Calif.-based company announced the new EPYC 7Fx2 processors on Tuesday, expanding AMD's second-generation, 7-nanometer EPYC lineup with three SKUs under 32 cores that feature boost frequencies of up to 3.9GHz and L3 caches reaching 256MB. Previously, the highest boost frequency achieved by an EPYC Rome processor was 3.4GHz.
The three new processors — the 24-core EPYC 7F72, the 16-core EPYC 7F52 and eight-core EPYC 7F32 — are being supported by server platforms from Dell EMC, Hewlett Packard Enterprise and Supermicro as well as 48 new base metal instances from IBM Cloud.
AMD's new EPYC processors are arriving only a couple months after rival Intel revealed its new lineup of Cascade Lake Refresh processors, which push higher core counts, caches and frequencies over the company's initial set of second-generation Xeon Scalable processors.
Dan McNamara, a former top Intel executive who now leads AMD's server business, made it clear in a briefing with journalists that AMD is targeting Intel's expanded Xeon Scalable lineup with its EPYC 7Fx2 processors, which he said were designed for three target markets: hyperconverged infrastructure, high-performance computing and relational database applications.
"We believe that the same value [proposition] that is allowing us to win at the cloud will help us to win in the enterprise with HCI," said McNamara, whose title is senior vice president and general manager of AMD's Server Business Unit, referring to the more than 150 EPYC-based cloud instances from Amazon Web Services, Microsoft Azure and others that will be live by the end of the year.
At the top of the stack is the EPYC 7F72, which has a list price of US$2,450 and comes with 24 cores, a 3.2GHz base clock frequency, a boost clock of up to 3.7GHz, a 192MB L3 cache and a 240W thermal design power. The 16-core EPYC 7F52, on the other hand, is more expensive at US$3,100 but comes with higher base and boost clock speeds of 3.5GHz and 3.9 GHz, respectively, as well as a higher L3 cache at 256MB, while also sporting a 240W TDP.
The last SKU is the eight-core EPYC 7F32, which runs for US$2,100 and comes with a higher base clock speed at 3.7GHz while sporting the same boost clock as the 16-core option and a lower L3 cache of 128 MB as well as a lower TDP of 180W.
McNamara said the new processors provide better performance and performance-per-CPU dollar over comparable processors in Intel's second-generation Xeon Scalable lineup, even though Intel's latest Cascade Lake Refresh processors feature higher frequencies than even the new EPYC CPUs.
To make his case, McNamara showed the three EPYC 7Fx2 processors achieving higher scores on a per-core basis in the SPECrate 2017_int_base integer performance benchmark against comparable Intel Xeon processors, which include the recently launched Xeon Gold 6248R and high-frequency Xeon Gold 6250. The EPYC processors also had higher scores than the Intel Xeon processors when considered on a performance-per-CPU dollar basis.
"What's really important there for us is not only to hit that performance point, but to also hit a cost point where the core-based licensing is attainable for our for our partners and our customers," he said.
To highlight the new processors' prowess in HCI workloads, McNamara said a four-rack Dell PowerEdge R6525 system running on the new 24-core EPYC 7F72 beat Intel's world record for the dual-socket, four-node VMmark 3 vSAN benchmark test by 47 percent.
"This really brings us into the heart of the enterprise in a big way with a lot of value," he said. "And I think that the value we're bringing is going to be challenging to not look at for even folks who maybe hadn't looked at it in the first launch of Rome."
Among the vendors supporting AMD's new EPYC 7Fx2 processors are Dell EMC, which will feature them in its PowerEdge R6525 server, and Hewlett Packard Enterprise, which will use them in upcoming ProLiant DX servers that will come with Nutanix HCI software as part of a partnership formed last year. Supermicro also plans to feature the processors in its new SuperBlade system, which is the vendor's first blade server to used AMD EPYC.
By the end of 2020, more than 140 server platforms will support AMD EPYC processors, McNamara said, which shows the progress the chipmaker has made since the first-generation lineup launched in 2017.
"We really feel like not only are the products coming to life, in across these different segments, but the ecosystem is really starting to grow heavily for us," he said.
Dominic Daninger, vice president of engineering at Nor-Tech, a Burnsville, Minn.-based HPC system integrator that partners with AMD and Intel, told CRN that with the new high-frequency EPYC processors, AMD is starting to close the gap with Intel, which has long held a frequency advantage.
"AMD has typically been lower frequency than Intel. One of the claims that Intel [has] made, and it was generally true, that on a per-core basis, Intel processors got more work done," he said. "But maybe [AMD is] closing the gap here, and we're getting closer at least to core-to-core comparability."
Higher frequencies in processors means improved economics in both HCI and HPC applications because both largely use per-core software licensing, according to Daninger, which means that there is no additional cost for running the applications at higher speeds.
"You run that same number of cores at a higher frequency, you get more work done and you can turn around and use that work to get more iterations done or higher fidelity in all of the models and still keep your license costs in check," he said.