1. 1 | AMD EPYC™ FAMILY OF PROCESSORS | PERFORMANCE WORLD RECORDS World records counted as of October 28, 2020. See AMD.com/worldrecords for details
AMD.com/worldrecords
2. 2 | AMD EPYC™ FAMILY OF PROCESSORS | PERFORMANCE WORLD RECORDS World records as of October 28, 2020. See AMD.com/worldrecords for details
31
10
** INCLUDES EPYC FAMILY OF PROCESSORS TIES WHICH ARE NOT COUNTED IN THE AMD WORLD RECORDS TOTAL
NUMBER
3. 3 | AMD EPYC™ FAMILY OF PROCESSORS | PERFORMANCE WORLD RECORDS
ONE SOCKET WORLD RECORDS
Right-Size Systems to
Workloads without Compromise
Memory Bandwidth Memory CapacityI/0
Expansion
TWO SOCKET WORLD RECORDS
Best in Class High-Performance,
Balanced Architecture
More Cores More Memory Bandwidth More I/O
** INCLUDES EPYC FAMILY OF PROCESSORS TIES WHICH ARE NOT COUNTED IN THE AMD WORLD RECORDS
TOTAL NUMBER
WORLD RECORDS AS OF OCTOBER 28, 2020. SEE AMD.COM/WORLDRECORDS FOR DETAILS.
93 87
4. 4 | AMD EPYC™ FAMILY OF PROCESSORS | PERFORMANCE WORLD RECORDS
• Decision Support – make quick time-to-decisions at a low per query cost
• Transwarp ArgoDB running TPC Benchmark™ DS @ 10TB SF
• SQL Server® database running TPC Benchmark™ H @ 1,000GB and 3,000GB SF
(non-clustered)
• Exasol® analytics database running TPC Benchmark™ H @ 1,000GB, 3,000GB, and
10,000GB SF (clustered)
• Big Data System – sort large amounts of data quickly
• Cloudera® Enterprise Hadoop Sort running TPC Express Benchmark™ HSv2
• MapReduce 2 @1TB, 3TB, 10TB, and 30TB SF
• APACHE SPARK® @1TB, 3TB, 10TB, and 30TB SF
• Internet of Things (IoT) – process gateway device data fast & efficiently
• Machbase™ database running TPC Express Benchmark™ IoT
World records as of October 28, 2020. See AMD.com/worldrecords for details
5. 5 | AMD EPYC™ FAMILY OF PROCESSORS | PERFORMANCE WORLD RECORDS
• IT Infrastructure – increase VM density while maintaining SLAs
• SUSE® Linux® Enterprise Server KVM running SPEC VIRT_SC® 2013
• VMware® ESXi® running VMmark® 3.1 matched pair hosts
• VMware ESXi with vSAN™ solution running VMmark® 3.1
• Dynamic relational database management – deploy cloud-like responsive
databases with confidence
• VMware® vSphere® running MySQL™ on TPC Express Benchmark™ V
World records as of October 28, 2020. See AMD.com/worldrecords for details
6. 6 | AMD EPYC™ FAMILY OF PROCESSORS | PERFORMANCE WORLD RECORDS
• Parallel Applications – rapid time-to-solutions for engineering apps
• OpenACC™ running SPEC ACCEL® workloads
• OpenCL™ running SPEC ACCEL® workloads
• OpenMP® running SPEC ACCEL® and SPEC OMP® 2012 workloads
• Medium and Large MPI applications running SPEC MPI® 2007 workloads
• SPEC CPU® 2017 running AOCC-compiled floating-point heavy apps
• Scale-up computing – handle the largest technical problems with single system-
image clusters
• ScaleMP® running SPECrate® 2017 Floating-Point applications
• Energy-efficiency – make the most of your HPC datacenter power/cooling
• AOCC-compiled applications running SPEC CPU® 2017 workloads
World records as of October 28, 2020. See AMD.com/worldrecords for details
7. 7 | AMD EPYC™ FAMILY OF PROCESSORS | PERFORMANCE WORLD RECORDS
• Enterprise Resource Management – grow your sales without growing your datacenter
• SAP® ASE running Sales and Distribution (SD) 2-tier workload
• Java®-based Business Applications – deliver performance needed for common Java®
application heavy demands
• Java SE 7 APIs running SPEC JBB® 2015 Composite, Distributed, and Multi-JVM retail
scenarios
• Energy-efficiency – improve server application performance per watt ratio
• Server-side Java application running SPEC Power® 2008 workload
• On-line transaction processing (OLTP) – support growing customer demand
• SQL Server® running TPC Benchmark™ E brokerage workload
• Goldilocks database running TPC Benchmark® C supply chain app
World records as of October 28, 2020. See AMD.com/worldrecords for details
The goal of this presentation is to capture all the industry-standard performance benchmarks based on AMD EPYC™ processors that cover a wide variety of segments, solutions, and application workloads representative of the data center market. Many of these metrics are used for initial sizing of planned deployments, provide a blueprint for solution designs, and provide insight into the cost per unit of performance.
Our partners, like AWS, ASUSTeK, Dell, GIGABYTE, H3C, HPE, Lenovo, ScaleMP, and Supermicro all are demonstrating their EPYC-based portfolio of products with leadership performance. With performance leadership, end customers can get the most out of their resource-constrained data centers to manage their operational expenses (admin, power/cooling, real estate, software licensing).
The next slide breaks down the high-level segments represented in the 170+ world records.
The four key segments AMD currently focuses on is data analytics, cloud-inspired infrastructure, HPC engineering/technical applications, and a variety of enterprise applications that are moving towards the Software Defined Infrastructure (SDI)/Hyperconverged Infrastructure (HCI) Narnia for Enterprises.
With 179 total world record results published (including one tie within EPYC family and two overall world records), the AMD EPYC processor holds 178 net records as of 10/28/2020 thanks to our OxM partners who sell the no-compromise, 1-socket and leadership 2-socket servers.
How to read the following detail slides:
Each line represents a performance publication that has one or more world record claims
The performance wins can include:
Overall world record (trophy cup) – this result is better than any other published result on this benchmark based on the benchmarking body rules for fencing i.e. companies cannot compare TPC results across different scale factors (10TB vs. 30TB)
Overall price performance world record ($/performance represented by coins) – this result is the cheapest per unit of performance than any other published result on this benchmark (NOTE: the tied records count for only one record as it is shared)
Windows (logo) – this result run on a Windows Server OS is better than any other published result on this benchmark with the given socket count
Linux (logo) – this result run on a Linux OS is better than any other published result on this benchmark with the given socket count
Socket (chart) – this result is better than any other published result this benchmark with the given socket count
One result, much like a marathon winner, can represent more than one world record (akin to best overall time, best male or female, and best-in-age group)