Recently, during a larger briefing on AMD in the Enterprise, I gave an AMD EPYC™ processor update to press and analysts on our progress since we launched last June. It was satisfying to pause for a few minutes, take stock, and reflect on the amazing momentum we’ve created in less than a year. A special guest joined me half way through with some exciting news that shows we’re just getting started on the journey in bringing value to customers through the innovations on EPYC™ processors. Below are some of the highlights from the session.
I started by giving a quick update on our ramp to date. Bottom line, a month away from our first birthday, we’ve been aggressive and purposeful in ramping the EPYC™ processor business, helping the industry rethink the server and the datacenter.
- We’ve built a revolutionary product line eagerly adopted by world class customers, ecosystem partners, and their customers. In fact, we now have 14 server systems partners and over 50 server platforms introduced and ramping.
- We’ve focused on rethinking the technology and economics of the server to help our customers win in the era of the software designed data center. The results are in: for virtualized environments, EPYC™ based systems can deliver up to 28% lower total cost of ownership than similarly configured Intel Xeon servers.
- Equally important, we’ve created superior performance, value, and advanced security features in emerging, rapidly growing use cases. Here’s some of the compelling examples I walked through. I’ve put some links at the bottom of this in case you want to get your ‘geek on’ and go deep:
One of the amazing customer examples I walked through was, Hivelocity. A dedicated server and cloud hosting provider, they’ve already taken the leap in deploying what we’ve coined the “industry’s first no-compromise single server” approach. It’s paid off, according to their COO Steve Eschweiler: “Our AMD EPYC™ processor-powered Tyan servers have truly blown away our price/performance expectations”. In fact, with a single AMD EPYC processor the Hivelocity client is experiencing read/write speeds six times faster than that of two Xeon processors.
Quite an accomplishment. We’re proud, and thankful… But we’re just getting started.
I was then joined by a special guest Lee Caswell, VP of Products, Storage and Availability at VMware. Lee joined me to cover the next high-growth use case we’re rethinking: Hyperconverged Infrastructure (HCI for short). That same ‘no compromise’ single socket approach applied by Hivelocity has the potential of putting this hypergrowth market into warp speed.
For those that aren’t familiar with HCI, it is fundamental to the Software Defined Datacenter. HCI uses software-defined storage through the same server resources used to run Virtual Machines, eliminating legacy storage systems, converging all elements into a single, easily managed, pool. VMware vSAN™ runs on industry-standard x86 servers and components that help lower TCO by up to 50% versus traditional storage. Lee helped me walk through the market, the challenges, and the opportunities through the lens of the HCI industry leader, VMware vSAN. Some takeaways from Lee’s presentation:
- HCI is a $5B market growing approximately 30% year over year. And, VMware vSAN is the undisputed software leader according to IDC’s latest report on April 3 of this year.
- Customers are adopting HCI not only for the prospect of massively reducing total cost of ownership, but also to accelerate the transformation to software defined data centers (SDDC) and move towards more flexible cloud consumption models.
- To meet customer requirements for HCI, configuration and testing are rigorous and deep. This goes well beyond just a standard Hardware Compatibility List (HCL). Many customers prefer to consume HCI as an appliance like Dell EMC’s VxRail as a result.
- But, many customers prefer to build their own HCI systems. This is where VMware’s proven vSAN ReadyNode™ program has tapped into an unmet need. These servers, precisely configured and rigorously tested, help ensure that customers can be confident in their ability to quickly deploy, scale, and operate HCI clusters. In fact, Lee mentioned that vSAN ReadyNodes actually make up the majority of vSAN implementations.
- Finally, most vSAN ReadyNodes to date have had to be beefy two socket platforms that are ripe for optimization and new thinking for storage centric environments. That’s where VMware, Dell, and AMD EPYC are really excited about the world’s first EPYC™ processor-based VMware certified vSAN ReadyNode.
The AMD EPYC™ processor-powered Dell PowerEdge 7415 recently launched and received its vSAN ReadyNode certification along with its big and little brother the PowerEdge 7425 and 6415. As a “no compromise” single socket server, it was designed with the prospect of delivering leading core density and I/O capability for storage heavy HCI use cases. Well, once again, the results are in. And they’re EPYC™. Lee and I revealed recent third-party test results showing we’ve more than delivered on that potential. In fact, versus currently shipping Xeon based systems, Dell shows the EPYC processor-powered PowerEdge 7415 delivering up to 20% lower TCO and slashing licensing costs by up to half. This is EPYC.
Thank you to Lee, his team at VMware, and the Dell EMC team for the partnership.
To close, I’d like to reiterate that we’re only getting started, expect to see more partners, more platforms, more performance, more value, more security in the coming months as we ramp to our first birthday and beyond. This is going to be EPYC™.
28% Lower TCO based on 3-year Virtual Machine Cost: 1. Data received from HPE TCO Calculator. Configure your own analysis at https://www.hpe.com/us/en/solutions/tco-calculators.html. 2. Based on an estimate of 1.5 virtual machines per core, the greater number of core in AMD EPYC processors exceeds that available in Intel Xeon Scalable processors 3. Based on comparison of top published SPECrate2017_fp_base scores for HPE ProLiant DL380 Gen10 and DL385 Gen10. AMD EPYC results available at www.spec.org as of 4/29/2018. 4. Source of pricing Configure your own TCO analysis https://roianalyst.alinean.com/ent_02/AutoLogin.do?d=898755097515045746. HPE ProLiant DL380 Gen10 config: 1 x 6130 (16 Total Cores), 64GB Memory, 8 SFF Chassis, 2x800W PS, P408i-a; HPE ProLiant DL385 Gen10 config: 1 x 7451 (24 Total Cores). 64GB Memory, 8 SFF Chassis, 2x800W PS, P408i-a
25% Fewer Servers for In Memory Analytics: Tests commissioned by AMD and run by Principled Technologies, Inc. running HiBench Spark, processing a 220 GB k-means dataset. AMD cluster of: 3 x Grandstands (chassis est. $600 ea.), each with (2) AMD EPYC 7601 SOC (AMD 1ku pricing $4200 ea), RAM: 32 x 16GB sticks ($169 ea. Samsung M393A2K40BB1-CRC per newegg.com) of DDR4 (512GB of RAM per server, 1536GB per cluster), Disk: 24 x Samsung MZ-7LM120E 120GB SSDs ($113 ea. per CDW), Network: 1 x Mellanox ConnectX-4 Lx ($332.77 per newegg.com); (for total cluster cost of approx. $52,449), completed the test in 213.6 seconds; versus Intel cluster of: 4 x Lenovo servers (chassis est. $600 ea.), each with (2) Intel Xeon E5-2699 v4 ($4115 ea. per ark.intel.com), C612 chipset ($54 per ark.intel. com), RAM: 24 x 16GB sticks ($169 ea. Samsung M393A2K40BB1-CRC per newegg.com) of DDR4 (384GB of RAM per server, 1536GB per cluster), Disk: 24 x Samsung MZ-7LM120E 120GB SSDs ($113 ea. per CDW), Network: 1 x Mellanox ConnectX-4 Lx ($332.77 per newegg.com), HBA: 1 x LSI LOGIC 9305-16i HBA card ($399.99 per newegg.com); (for total cluster cost of approx. $65,628), completed in 243.1 seconds. Each cluster was connected via a Mellanox SN2700 model switch, and ran Red Hat Enterprise Linux 7.3, Hortonworks Data Platform (HDP) 2.6, and OpenJDK 1.8. NAP-23 Testing by Principled Technologies and not verified by AMD. This compares acquisition cost for a 4-node cluster of Intel Xeon E5 2699 v4 processor-based servers compared to a 3-node cluster of AMD EPYC 7601 processor-powered servers. NAP-23
50% More Throughput for NoSQL Database: AMD EPYC 7601 based server performance comparison to Intel Xeon 2699 v4 based server in Principled Technologies white paper at http://www.principledtechnologies.com/AMD/EPYC_ Cassandra_competitive_1217.pdf
77% More Performance for Weather Research Forecasting: AMD internal testing on WRF v3.8.1 (GCC 5.4.0 compiler with Basic Nesting (WRF option) and MPI 3.2 (dmpar compile flag); CONUS 12KM dataset; WRFIO_NCD_LARGE_FILE_SUPPORT=1). Tests conducted on AMD “Grandstand” reference system with 2 x EYPC 7601 SoCs, 16 x 32GB DDR4-2400MHz, 3 x 350GB SATA SSDs, Ubuntu 16.04.2 completed the test in an average of 65 seconds; versus HPE DL380 server with 2 x E5-2699A v4 processors, 16 x 16GB DDR4-2400MHz, 3 x 1TB NVMe and 3 x 1TB SATA SSDs, Ubuntu 16.04.4 completed the test in an average of 115 seconds. NAP-92
133% Better Performance/Dollar for Supercomputing: Based on SPECfp®_rate2006 scores published on www.spec.org as of October 25, 2017. 2 x EPYC 7601 CPU ($4,200 per processor at AMD 1ku pricing) in Sugon A620-G30, Ubuntu 17.04, x86 Open64 v22.214.171.124 Compiler Suite, 512 GB PC4-2666V-R memory, running at 2400 1 x 1TB SATA 7200RPM has a peak score of 1850 (base score 1670); versus 2P Xeon Platinum 8180 ($10,009 per processor per ark.intel.com)-based Huawei 2288H V5 system with SUSE Linux Enterprise Server 12 SP2, ICC 17.0.0.098, 384GB PC4-2666V-R memory, 1x1200GB SAS 10000RPM score of 1890 (base score 1850). SPEC and SPECfp are registered trademarks of the Standard Performance Evaluation Corporation. See www.spec.org for more information. NAP-47
26% Faster Simulations for Computational Fluid Dynamics: Based on Dell internal testing using the ANSYs Fluent benchmark test in November 2017, comparing to a similarly configured Dell PowerEdge R7425 with a traditional processor. Actual performance will vary based on configuration, usage and manufacturing variability. World Record SPECfp(R)_rate2006: Result available at https://www.spec.org/cpu2006/results/res2017q4/cpu2006-20171114-50603.html. 2 x EPYC 7601 CPU in HPE ProLiant DL385 Gen10, SUSE Linux Enterprise Server 12 SP3, x86 Open64 v126.96.36.199 Compiler Suite, 1 TB (16 x 64GB 4Rx4 PC4-2666V-L) memory, 1 x 300 GB 15k RPM SAS. SPEC and SPECfp are registered trademarks of the Standard Performance Evaluation Corporation. See www.spec.org for more information.
World Record SPECrate2017_fp_peak as of May 21, 2018: Result available at https://www.spec.org/cpu2017/results/res2018q2/cpu2017-20180319-04087.html. Based on SPEC CPU 2017 scores published on April 19, 2018. 2 x EPYC 7601 CPU in Supermicro A+ Server 4023S-TRT, SUSE Linux Enterprise Server 12 SP3 (x86_64) kernel 4.4.114-94.11-default, AOCC v1.0.0 compiler, 1 TB (16 x 64GB 4Rx4 PC4-2666V-L) memory, 1 x 500 GB SATA III 7200RPM has a peak score of 279 (base score 267). NAP-96
Up to 50% lower TCO VMware vSAN 6.7 Datasheet at https://www.VMware.com/content/dam/digitalmarketing/VMware/en/pdf/products/vsan/VMware-vsan-datasheet.pdf
Up to 20% Lower TCO and up to 50% Lower Licensing Costs: Demartek test report: Dell EMC PowerEdge R7415 AMD EPYC VMware vSAN Mixed Workloads Performance – April 2018 www.demartek.com/Demartek_Dell_EMC_PowerEdge_R7415_vSAN_Mixed_Workloads_Evaluation_2018-04.html