AMD is launching three new SKUs today based on the 7nm second-generation Epyc “Rome” platform, targeting commercial HPC workloads as well as hyperconverged infrastructure and database applications. Dell EMC, HPE, Lenovo, Supermicro and Nutanix have announced support for the new CPUs, and IBM Cloud is the first public cloud partner.
The three processors introduced today are the eight-core Epyc 7F32, the 16-core 7F52 and the 24-core 7F72, featuring base clocks up to 3.7 GHz (up to 3.9 GHz with boost) within a TDP range of 180 to 240 watts, and a list price between $2,100 and $2,450 (with a commitment of one-thousand units). Unlike previously announced 7002-series SKUs, which were differentiated for single-socket and dual-socket platforms, these SKUs will do double-duty. (Technically-speaking, every Epyc processor can be made dual- or single-socket.)
The 7F-series are high frequency SKUs that offer the highest per core performance of the Epyc family, said Dan McNamara, senior vice president and general manager, server business unit, AMD, in a pre-briefing held for media last week. He stressed the importance of the L3 cache, up to 16 megabytes per core, for databases and commercial HPC applications, notably electronic design automation (EDA). In particular, the 16-core part is “a sweet spot for some of these EDA workloads that really require a higher cache per core,” said McNamara, adding that AMD fine-tuned boost and cache for that market and for other enterprise HPC workloads, including R&D simulation and computational fluid dynamics, as well as database environments.
While AMD’s top-bin parts (such as the 7742, the 7702 and the goosed 7H12 SKU) offer up to 64-cores to satisfy cutting-edge performance needs, the 7F-series family is cost-performance oriented. “Looking at the commercial HPC target, it’s clearly a couple of steps below [what we’re doing for] El Capitan and Frontier,” said McNamara, referring to the two planned exascale machines at Lawrence Livermore National Lab and Oak Ridge National Lab, respectively. “However,” continued McNamara, “what’s really important [on the commercial side] is not only to hit that performance point, but to also hit a cost point where the core-based licensing is attainable for our partners and our customers.”
AMD touts the new 7F-series as having the highest per core performance of any x86 server CPU. The claim is based on SPECrate 2017_fp testing of the 8-core Epyc 7F32 part, which achieved 12.75 base result per core, besting all other SPEC publications as of today’s launch date, according to AMD (see link to SPEC results).
In HPC benchmarking tests conducted by AMD, two 16-core 7F52 chips offered 94 percent higher per core performance for a computational fluid dynamics application and 33 percent higher performance per core for a finite element analysis workload, compared with two Intel Xeon Gold 16-core 6242 CPUs. That performance lead falls to 54 percent and 18 percent, respectively, comparing 7F72 Epyc processors with Xeon Platinum 8268 CPUs – see slide below. For electronic design automation using Synopsys software, a single-socket Epyc 7F72 (24-core) server provided 31 percent higher performance compared with a dual-socket Xeon Gold 6246-based server (24 total cores), with AMD further claiming a 60 percent lower CPU acquisition cost with its hardware ($2,450 versus $6,572).
Moving to the enterprise database space, AMD’s Epyc 7F52 processor offers 17 percent higher SQL server performance compared to the competition on a per-core basis, according to AMD, and for hyperconverged infrastructure, the 7F72 processor achieved a 47 percent higher VMmark 3.1 score over a competitive product, using vSAN as the storage tier in a 4-node cluster.
Launch partner Supermicro announced a new blade server — the H12 A+ Superblade — based on the 2nd gen Epyc 1P family of CPUs, including the new 7F-series SKUs, targeting the volume enterprise datacenter. It is the industry’s first blade server platform to implement Epyc processors, according to the partners, and features 20 hot-pluggable single-socket nodes in 8U, PCIe gen4 support, integrated 25 Gigabit Ethernet with optional 100 Gigabit EDR InfiniBand support (200G HDR is planned). Additionally, Supermicro A+ platforms — the Ultra, GPU, WIO, Twin and Mainstream systems — can be outfitted with the new Epyc 7F-series processors starting today.
HPE continues to welcome AMD CPUs into its portfolio, and will be offering all three Epyc 7Fx2 processors on a number of platforms, including the recently announced HPE Apollo 2000 Gen10 Plus system, the ProLiant DL385 Gen10 Plus server and ProLiant DX servers.
System maker Lenovo will offer the AMD 7Fx2 processors on its single-socket ThinkSystem SR635 and SR655 platforms, targeting data analytics, software defined storage and remote work capabilities — areas where per core performance is critical, noted the company.
Another launch partner, Nutanix, together with Hewlett Packard Enterprise, announced support for Nutanix HCI software on select AMD Epyc based HPE ProLiant servers by May.
Dell, VMWare, and Microsoft have also announced support for the Epyc 7F-series processors.
The first public cloud partner is IBM Cloud, which will offer access to the AMD Epyc 7F72 processors in its bare metal offering that just two weeks ago introduced AMD Epyc 7642 processors. “Now, IBM Cloud provides access to another high core-count dual socket bare metal server with high clock speed frequency, giving our clients more optimized platform choices for compute-intense workloads such as analytics, commercial HPC and EDA,” said Satinder Sethi, general manager, IBM Cloud Infrastructure Services.
AMD, who works closely with all the major cloud providers, said it is on track to support 150 public cloud instances by year end.
The Epyc 7F-series processors are available today; system availability varies by OEM partner.