The best databases in the world perform better on AMD EPYC processors.

Oracle Exadata

They at AMD think that their EPYC processors are the finest data center CPUs available. Their clients concur. EPYC processors have become widely used, as shown by the fact that they run the Frontier system at Oak Ridge National Laboratory, the most powerful supercomputer in the world, and that they are used in over 800 public cloud instances across all major cloud service providers. Businesses are using more and more EPYC CPUs to power their on-premise infrastructures across all business verticals.

Indeed, AMD EPYC now holds over 300 world records for efficiency and performance, both on-site and in the cloud. However, did you realize that data management and analytics account for more than 50 of those records? This indicates that AMD EPYC is the best option available for processing the most important and fundamental business tasks in the globe.

The two databases that big businesses use the most often are Oracle and Microsoft SQL Server. AMD’s 4th Gen EPYC CPUs provide impressive performance whether you’re using Microsoft SQL Server or the Oracle Exadata Database Platform. First, let’s examine the Oracle Exadata platform to understand how AMD EPYC has contributed to the advancement of database computing.

The Database Platform for Oracle Exadata

Managing workloads for Oracle Exadata databases is a major endeavor. Exadata offers real-time vector search capabilities, analytics, and fast transaction processing. Tasks that are vital to the mission may be handled by the Exadata system. These workloads allow for the instantaneous completion of online financial transactions while instantly detecting fraudulent activity. These database workloads help carriers maintain their networks and keep up with the increasing needs of the worldwide 5G network. They also keep the internet transactions flowing, particularly during the peak shopping seasons of Cyber Monday and Single’s Day.

Oracle developed the Oracle Exadata Database Platform to provide performance, scalability, availability, and security while supporting these workloads. This platform, which includes hardware, software, and an Oracle database, is simple to set up.

Oracle made a significant shift in 2023 after depending on Intel CPUs for years to power the Oracle Exadata Database Platform. Oracle decided to employ the 4th generation AMD EPYC CPU (previously code-named “Genoa”) in the most recent version X10M Exadata system in order to achieve better performance and energy efficiency.

Exadata Oracle

Think about the impact that EPYC CPUs have: Compared to the preceding Exadata X9M-2 system, the Exadata X10M Database Platform features three times more database server cores per socket from 32 to 96 cores thanks to AMD EPYC CPUs. More EPYC CPU cores simply translate into more database transactions and quicker analytics when combined with Oracle’s optimized software.

Meanwhile, the dual-socket X10M Database server can now tackle tasks that previously needed pricey and power-hungry 8-socket servers thanks to its remarkable core count. An organization may recover precious space in the data center and save energy by using fewer, smaller servers to do the same tasks.

X10M Database server

In order to ensure that database performance grows linearly with a core count of up to 192 per Database Server (2 X 96 core CPUs), AMD and Oracle worked together to enhance the performance of the new Exadata X10M server. Exadata system software, on the other hand, can encrypt and decrypt data quicker than other components can get it to the chip thanks to the 4th generation AMD EPYC CPU. Furthermore, the Exadata system software was tailored to take full use of AMD EPYC’s reliability, availability, and serviceability (RAS) features. One example of this is Platform First Error Handling, which improves the uptime of workloads related to key databases.

Microsoft SQL Server

Let’s now examine Microsoft SQL Server and see how AMD EPYC leads the industry in database transaction processing speed. Businesses often utilize the TPC (Transaction Processing Performance Council) benchmarks to assess processor performance as well as the overall performance of server systems. To guarantee the validity of the benchmark test findings, TPC.org has an audited library of benchmarks. Let’s examine two widely used TPC benchmarks to assess database performance: TPC-E and TPC-H.

Decision support systems that analyze vast amounts of data, carry out intricate queries, and respond to important business inquiries are evaluated using the TPC-H benchmark. According to this test, the best performance the most queries per hour is achieved by a non-clustered system running Microsoft SQL Server 2022 on a 64-core 4th Gen AMD EPYC CPU. Performance on the AMD-based system is 14% better than that of a system powered by the most recent 5th Gen Intel Xeon 8592+. To put it another way, the firm can evaluate data considerably more quickly and get business outcomes since there are 14% more inquiries answered in an hour.

The findings shown above pertain to a database with a 10TB size. Furthermore, the TPC-H findings show that AMD EPYC powered systems also provide the best non clustered performance for 3 TB (EPYCWR-869) and 1 TB databases (EPYCWR-865).

In the meanwhile, online transaction processing (OLTP) is measured by the TPC-E benchmark. There are twelve simultaneous transactions of varying kinds and complexity involved. Either online or time- or price-based criteria are used to initiate the transactions.

Once again, while running Microsoft SQL Server 2022, the AMD-based system achieves the greatest performance for a non-clustered server on the TPC-E benchmark. When compared to a system using a 5th generation Intel Xeon CPU, the system built around the 4th generation AMD EPYC processor performs 7% better.

In summary

AMD has shown that its EPYC CPUs provide the greatest performance available on the market when used with Microsoft SQL Server and Oracle Exadata databases. It is not worth taking shortcuts when it comes to organizing and evaluating data that is essential to the continuous operation of your organization.

AMD continues to set the standard in the server industry with the upcoming release of 5th Gen EPYC CPUs. For the highest database workloads, AMD’s unrivaled 4th Gen EPYC CPUs are the obvious option for the finest database performance in the interim.

EPYC processors

AMD fundamentally altered the rules when it returned to the server market in 2017 with the EPYC chip. Record-breaking performance, robust ecosystem support, and platforms tailored for contemporary workflows allowed EPYC to seize market share fast. AMD EPYC began the year with a meagre 2% of the market, but according to estimates, it now commands more than 30% of the market. All of the main OEMs, including Dell, HPE, Cisco, Lenovo, and Supermicro, offer EPYC CPUs on a variety of platforms.

Best EPYC Processor

Given AMD EPYC’s extensive presence in the public cloud and enterprise server markets, along with its numerous performance and efficiency world records, it is evident that EPYC processors is more than capable of supporting Red Hat OpenShift, the container orchestration platform. EPYC is the finest option for enabling application modernization since it forms the basis of contemporary enterprise architecture and state-of-the-art cloud functionalities. Making EPYC processors argument and demonstrating why AMD EPYC should be taken into consideration for an OpenShift implementation at Red Hat Summit was a compelling opportunity.

Gaining market share while delivering top-notch results

Over the course of four generations, EPYC’s performance has raised the standard. The fastest data centre  CPU in the world is the AMD EPYC 4th Generation. For general purpose applications (SP5-175A), the 128-core EPYC provides 73% better performance at 1.53 times the performance per projected system watt than the 64-core Intel Xeon Platinum 8592+.

In addition, EPYC provides the leadership inference performance needed to manage the increasing ubiquity of  AI. For example, utilising the industry standard end-to-end AI benchmark TPCx-AI SF30, an Intel Xeon Platinum 8592+ (SP5-051A) server has almost 1.5 times the aggregate throughput compared to an AMD EPYC 9654 powered server.

A comprehensive array of data centres and cloud presence

You may be certain that the infrastructure you’re now employing is either AMD-ready or currently operates on AMD while you work to maximise the performance of your applications.

Red Hat OpenShift-certified servers are the best-selling and most suitable for the OpenShift market among all the main providers. Take a time to look through the Red Hat partner catalogue, if you’re intrigued, to see just how many AMD-powered choices are compatible with OpenShift.

On the cloud front, OpenShift certified AMD-powered instances are available on AWS and Microsoft Azure. For instance, the EPYC-powered EC2 instances on AWS are T3a, C5a, C5ad, C6a, M5a, M5ad, M6a, M7a, R5a, and R6a.

Supplying the energy for future tasks

The benefit AMD’s rising prominence in the server market offers enterprises is the assurance that their EPYC infrastructure will perform optimally whether workloads are executed on-site or in the cloud. This is made even more clear by the fact that an increasing number of businesses are looking to jump to the cloud when performance counts, such during Black Friday sales in the retail industry.

Modern applications increasingly incorporate or produce  AI elements for rich user benefits in addition to native scalability flexibility. Another benefit of using AMD EPYC CPUs is their shown ability to provide quick large language model inference responsiveness. A crucial factor in any AI implementation is the latency of LLM inference. At Red Hat Summit, AMD seized the chance to demonstrate exactly that.

AMD performed Llama 2-7B-Chat-HF at bf16 precision​over Red Hat OpenShift on Red Hat Enterprise Linux CoreOS in order to showcase the performance of the 4th Gen AMD EPYC. AMD showcased the potential of EPYC on several distinct use cases, one of which was a chatbot for customer service. The Time to First Token in this instance was 219 milliseconds, easily satisfying the patience of a human user who probably anticipates a response in under a second.

The maximum performance needed by the majority of English readers is approximately 6.5 tokens per second, or 5 English words per second, but the throughput of tokens was 8 tokens per second. The model’s performance can readily produce words quicker than a fast reader can usually keep up, as evidenced by the 127 millisecond latency per token.

Meeting developers, partners, and customers at conferences like Red Hat Summit is always a pleasure, as is getting to hear directly from customers. AMD has worked hard to demonstrate that it provides infrastructure that is more than competitive for the development and deployment of contemporary applications. EPYC processors, EPYC-based commercial servers, and the Red Hat Enterprise Linux and OpenShift ecosystem surrounding them are reliable resources for OpenShift developers.

It was wonderful to interact with the community at the Summit, and it’s always positive to highlight AMD’s partnerships with industry titans like Red Hat. EPYC processors will return this autumn with an update, coinciding with Kubecon.

Red Hat OpenShift’s extensive use of AMD EPYC-based servers is evidence of their potent blend of affordability, effectiveness, and performance. As technology advances, they might expect a number of fascinating breakthroughs in this field:

Improved Efficiency and Performance

EPYC processors of the upcoming generation

AMD is renowned for its quick innovation cycle. It’s expected that upcoming EPYC processors would offer even more cores, faster clock rates, and cutting-edge capabilities like  AI acceleration. Better performance will result from these developments for demanding OpenShift workloads.

Better hardware-software integration

AMD, Red Hat, and hardware partners working together more closely will produce more refined optimizations that will maximize the potential of EPYC-based systems for OpenShift. This entails optimizing virtualization capabilities, I/O performance, and memory subsystems.

Increased Support for Workloads

Acceleration of AI and machine learning

EPYC-based servers equipped with dedicated AI accelerators will proliferate as AI and ML become more widespread. As a result, OpenShift environments will be better equipped to manage challenging AI workloads.

Data analytics and high-performance computing (HPC)

EPYC’s robust performance profile makes it appropriate for these types of applications. Platforms that are tailored for these workloads should be available soon, allowing for OpenShift simulations and sophisticated analytics.

Integration of Edge Computing and IoT

Reduced power consumption

EPYC processors of the future might concentrate on power efficiency, which would make them perfect for edge computing situations where power limitations are an issue. By doing this, OpenShift deployments can be made closer to data sources, which will lower latency and boost responsiveness.

IoT device management

EPYC-based servers have the potential to function as central hubs for the management and processing of data from Internet of Things devices. On these servers, OpenShift can offer a stable foundation for creating and implementing IoT applications.

Environments with Hybrid and Multiple Clouds

Uniform performance across clouds

major cloud providers will probably offer EPYC-based servers, which will guarantee uniform performance for hybrid and multi-cloud OpenShift setups.

Cloud-native apps that are optimised

EPYC-based platforms are designed to run cloud-native applications effectively by utilising microservices and containerisation.

VMmark User Guide

After their introduction in 2017, AMD EPYC processors have become the preferred option for both public and private cloud deployments. High core counts, the fastest x86 memory bandwidth in the market, and cutting-edge security features make 4th generation AMD EPYC processors ideal for demanding private cloud workloads. AMD EPYC processors offer a strong platform for building private cloud infrastructures, as do VMware by Broadcom products including VMware Cloud Foundation (VCF) and VMware vSphere Foundation (VVF), which comprise programmes like vSphere, vCenter, vSAN, and NSX. The strategic partnership between AMD and VMware By Broadcom helps organisations install and manage virtualized environments that meet their performance and scalability needs while optimising energy efficiency and TCO.

Two typical deployment scenarios for VCF/VVF exist:

Conventional VVF infrastructure typically comprises discrete networking, storage, and computation gear. Running this model requires managing each part independently, which frequently calls for specialised knowledge and equipment, but it produces the best results. The computational component is virtualized using software like VMware vSphere, which is controlled by a virtual infrastructure administrator. This configuration’s storage component depends on specialised Storage Area Network (SAN) or Network Attached Storage (NAS) devices under the control of a Storage Administrator. A contemporary method of managing data centres is called hyper-converged infrastructure (HCI), which integrates networking, storage, and processing into a single, cohesive system. Typically, HCI systems pool storage resources throughout the entire infrastructure for simple scaling and management by a Virtual Infrastructure Administrator. This is made possible by Software Defined Storage (SDS), which is made possible by products like vSAN. Compared to conventional VCF/VVF installations, this simplified method can maximise savings while enhancing agility and scalability. VMmark Benchmark

By offering thorough performance measurements that mimic real-world workloads to evaluate CPU, memory, storage, and networking performance, the VMmark benchmark helps enterprises to evaluate the effectiveness and scalability of their virtualized systems. Businesses may decide on hardware configurations, resource allocation, and workload management methods with knowledge thanks to this comprehensive picture of system capabilities under various workload scenarios.

By applying the smallest unit, or “tile,” of load, VMmark3 employs a weighted scoring system to assess a server’s performance. Nineteen virtual machines are included of each tile, which represents a typical enterprise virtualization scenario by executing a wide collection of tasks both concurrently and collaboratively. Each tile’s static nature allows for the size and scalability of the amount of work for every distinct VMmark3 cluster.

With a growth in hosts, sockets, and cores, there are usually more tiles. Each tile also serves as a representation of the maximum theoretical score and a Quality of Service (QoS) metric, which are used to appropriately scale each publication: an excessive number of tiles exceeds the benchmark latency requirements, while an insufficient number of tiles limits the maximum score.

While application performance accounts for a piece of the total score, the quantity of tiles is essential to comprehending the potential of a VCF deployment. Many of the capabilities that are exclusive to VCF, like XvMotion, Storage vMotion, vMotion, and the Distributed Resource Scheduler (DRS), are also utilised by standard virtualized deployments. By utilising these capabilities, VMmark workloads offer a more comprehensive view of what can be accomplished with the entire system as opposed to just one server. Depending on the underlying technology, these infrastructure processes can vary in scope and duration, accounting for 20% of the total score.

Benefits of Generational Performance AMD is committed to innovation and keeps delivering notable gains in efficiency and performance for important applications. As can be seen in Figure 1, AMD EPYC processors have significantly improved in performance with each iteration when used in VCF scenarios. These assessments include single- and dual-processor installations in both normal settings and vSAN configurations. The consistent dedication of AMD towards meeting the dynamic and ever-increasing demands of contemporary virtualized infrastructures is demonstrated by the notable performance improvements attained with every new generation of AMD EPYC CPUs.

Performance of Throughput

Thanks to continued AMD processor technology advances and collaborative VMware By Broadcom optimisations, the preceding section of this blog revealed constant performance increases across successive AMD EPYC processor generations. The performance domination of AMD EPYC processors over several processing generations is demonstrated in Figure 2, which highlights their unwavering commitment. In VMmark3 performance tests conducted on clusters comprising two dual-socket servers for a total of four processor sockets, the 4th generation AMD EPYC 9654 CPU outperforms even the most recent 5th generation Intel Xeon Platinum 8592+ processor.

Per-Core Efficiency

When it comes to power, space, and server count optimisation, per-server throughput is a crucial measure. Additionally, because per-core performance is frequently a major issue in software licencing models based on core count, it is especially important for workloads that are SLA-critical.

The performance improvements of 32-core AMD EPYC 9374F processors over 32-core Intel Xeon Gold 6548Y+ processors and 64-core AMD EPYC 9554 processors over 64-core 5th Gen Intel Xeon 8592+ processors are shown in Figure 3.

Power Effectiveness

In order to minimise cooling requirements, maximise utility costs, and support sustainability goals, modern data centres must have efficient power consumption. In comparison to both 4th and 5th Gen Intel Xeon processors, Figure 4 illustrates how 4th Gen AMD EPYC 9654 CPUs provide higher power efficiency.

Space for a Data Centre VMmark Scores Data centres have tremendous incentives to optimise space, electricity, and cooling for sustainability and TCO. Less servers that may offer more virtual machine consolidation without sacrificing the speed required for corporate operations can be used to achieve these goals. Suppose AMD workloads need to achieve business objectives with resources and quality of service equal to a total VMmark3 score of 500. This aggregate score can be attained with just 26 servers running 4th Gen AMD EPYC 9654 processors, as shown in Figure 5.

AMD will require 38 servers if they choose to employ servers with 5th generation Intel Xeon 8592+ processors. It may take up to 112 servers for customers with 2nd Gen Intel Xeon 8280 processor-based servers to get the same aggregate score! That’s more than 4 times the amount of AMD EPYC 9654 processor-based servers required to meet the same workload in business! This comparison emphasises the significance of not only power efficiency but also making the best use of server rack space in datacenters.

VMmark Benchmark Results This article demonstrated how VMmark victories and the next generation of AMD EPYC processors can drive consistent performance leadership, giving organisations optimal hardware consolidation and unparalleled x86 workload performance that may help reduce total cost of ownership. Businesses can safely depend on AMD EPYC processors to satisfy their changing virtualized infrastructure demands while maximising available power and capacity. Together, AMD and VMware by Broadcom are still dedicated to promoting innovation that continuously yields record-breaking virtualized environment performance and efficiency. Businesses can achieve new levels of agility, scalability, security, and reliability in their IT infrastructures by utilising Broadcom’s combination of experience in AMD and VMware.