At AMD, we believe our EPYC processors are the world's best data center CPUs. Our customers agree. From powering the world's most powerful supercomputer -- the Frontier system at Oak Ridge National Laboratory -- to having more than 800 public cloud instances with every major cloud service provider, EPYC processors have quickly gained wide adoption. Enterprises in every industry vertical are increasingly turning to EPYC CPUs to power their on-premise environments.
In fact, AMD EPYC now boasts more than 300 world records for performance and efficiency on premise and in the cloud. Did you know, however, that more than 50 of those records are in data management and analytics? This means that when it comes to running the world's most mission-critical, foundational business functions, there's no better choice than AMD EPYC. Oracle Database and Microsoft SQL Server are the two most commonly used databases by large enterprises. Whether you're running the Oracle Exadata Database Platform or Microsoft SQL Server, AMD’s 4th Gen EPYC processors deliver compelling performance. Let’s look at the Oracle Exadata platform first to see how AMD EPYC has helped bring superior computing to databases.
The Oracle Exadata Database Platform
Running Oracle Exadata Database workloads is serious business. Exadata delivers high-speed transaction processing, analytics, and real-time vector search capabilities. The Exadata system is built for mission-critical workloads. These are the workloads that make online financial transactions happen at the speed of light -- while identifying fraudulent transactions in real time. These database workloads keep the transactions humming on the internet, especially during the deluge of Cyber Monday or Single’s Day shopping periods; they allow telcos to manage their networks and keep up with growing demands of the global 5G network.
To support these workloads -- with the assurance of performance, scale, availability, and security -- Oracle built the Oracle Exadata Database Platform. It's an easy-to-deploy platform comprising Oracle database, hardware and software.
After years of relying on Intel CPUs to run the Oracle Exadata Database Platform, Oracle in 2023 made a major change. Looking for superior performance and energy efficiency, Oracle opted to use 4th Gen AMD EPYC processor (formerly code-named “Genoa”) on the latest generation X10M Exadata system.
Consider the difference EPYC CPU’s makes: With AMD EPYC CPUs, the Exadata X10M Database Platform has increased its database server cores per socket by 3x, from 32 to 96 cores when compared with the previous Exadata X9M-2 system. With Oracle’s optimized software, more cores with EPYC CPUs simply equates to more database transactions and faster analytics.
Meanwhile, with this impressive core count, the dual-socket X10M Database server can now handle the same workloads that typically required expensive and power hungry 8-socket servers in the past. With fewer, smaller servers needed to get the job done, an enterprise can reclaim valuable real estate and realize power savings within the data center.
AMD and Oracle collaborated to optimize the performance of the new Exadata X10M server, ensuring that database performance scales linearly with a core count of up to 192 per Database Server (2 X 96 core CPUs). Meanwhile, with the 4th Gen AMD EPYC processor, the Exadata system software can encrypt and decrypt data faster than other components deliver that data to the chip. On top of everything, the Exadata system software was optimized for AMD EPYC’s reliability, availability, and serviceability (RAS) capabilities -- think of Platform First Error Handling, which enhances the uptime of critical database workloads.
Microsoft SQL Server
Now, let’s look at Microsoft SQL Server and how AMD EPYC shines with industry leading performance for database transaction processing. To judge performance of the processor as well as complete server systems, enterprises widely use the TPC (Transaction Processing Performance Council) benchmarks. TPC.org maintains a repository of benchmarks that are audited to ensure authenticity of the benchmark test results. Let’s look at two popular TPC benchmarks for evaluating database performance: TPC-H and TPC-E.
The TPC-H benchmark measures decision support systems that examine large volumes of data, execute complex queries, and answer critical business questions. This benchmark shows that a non-clustered system running Microsoft SQL Server 2022 with a 64-core 4th Gen AMD EPYC processor delivers the best results -- the highest number of queries per hour. The AMD-based system delivers 14% higher performance than a system running the latest 5th Gen Intel Xeon 8592+. In other words, the number of queries per hour is 14% higher allowing the enterprise to analyze data much quicker to deliver business results.
The above results are for a 10TB size database. In addition, AMD EPYC powered systems also deliver the best non clustered performance for 3 TB (EPYCWR-869) and 1 TB databases (EPYCWR-865), according to the TPC-H results.
Meanwhile, the TPC-E benchmark measures on-line transaction processing (OLTP). It involves a mix of 12 concurrent transactions of different types and complexity. The transactions are either executed online or triggered by price or time criteria.
Once again, for the TPC-E benchmark, the AMD-based system delivers the best performance for a non-clustered server when running Microsoft SQL Server 2022. The system based on the 4th Gen AMD EPYC processor delivers 7% higher performance when compared to 5th Gen Intel Xeon system.
5th Gen Intel Xeon 8592+
Conclusion
AMD has demonstrated that with both Oracle Exadata and Microsoft SQL Server databases, its EPYC CPUs offer the best the market has to offer. When managing and analyzing data that's critical for keeping your business up and running, it's not worth cutting corners.
With 5th Gen EPYC processors on the horizon, AMD continues to raise the bar in the server market. In the meantime, AMD's unmatched 4th Gen EPYC CPUs are the clear choice for the best database performance for the most demanding database workloads.
