20 November 2019 – For many years the Numerical Algorithms Group (NAG) has been internationally recognised as the impartial source of expertise on the technical and business aspects of High Performance Computing (HPC). It is thus no surprise that AMD selected NAG to help evaluate the performance of the new 2nd Gen AMD EPYC™ processor as part of its recent highly successful launch.
In the build up to the launch, NAG benchmarked the 2nd Gen AMD EPYC processor (models 7742 and 7542) using a series of typical HPC applications codes including computational fluid dynamics (CFD), computational chemistry and materials science, reservoir modelling, and thermodynamics.
The benchmarking study shows that the new AMD CPU delivers excellent performance across a range of applications, especially where memory bandwidth was a key performance requirement. The processors shows a valuable performance increase from the previous generation CPU, along with good scaling properties. In many cases the performance was twice that of the competing Intel Cascade Lake processors. Coupled with the 2nd Gen AMD EPYC processor’s lower list prices, it is a leading proposition for HPC workloads.
The results of the benchmarking study are explored in further detail via a series of whitepapers available on the AMD website here
NAG’s Andrew Jones, Vice-President of HPC Services and Consulting, and trusted HPC community voice, said: “I’m delighted that the 2nd Gen EPYC processor has lived up to the expectations. The HPC market demands strong performance improvements at good price points and the new EPYC processor clearly delivers that, as shown by our benchmarking and by the series of AMD customer wins already announced in the HPC space. The EPYC processor has also performed well in many of the confidential benchmarking services we provide for HPC buyers in industry and the public sector for procurement evaluations and technology planning.”
Kevin Mayo, Director, HPC Software and Solutions Engineering, AMD, said: “NAG’s impartiality and expertise were crucial to help ensure we had a solid basis for our market leadership positioning during the product planning and very successful launch of the 2nd Gen AMD EPYC processor. The NAG team delivered expert benchmarking analysis that helped us to fine tune the EPYC product and its software ecosystem, and provided excellent data on our performance.”
NAG HPC team helped to varying degrees on most of the following work, either as a consultant/tuning guidance/reviewing results/etc., or in actually running the benchmarking tests
Case studies including competitive comparisons vs. Intel Cascade Lake
- ANSYS® Fluent® SOLVING REAL WORLD PROBLEMS
- MULTI-DOMAIN SIMULATION WITH ESI VIRTUAL PERFORMANCE SOLUTION
- STRUCTURAL ANALYSIS WITH ALTAIR RADIOSS™
AMD Prior Generation Comparisons
- NAMD MOLECULAR DYNAMICS SIMULATION
- GROMACS MOLECULAR DYNAMICS SIMULATION
- REAL WORLD SIMULATION WITH LSTC LS-DYNA®
- WEATHER MODELING WITH WRF
Figure 1 shows outstanding performance on the CONUS12km model. The 32-core EPYC 7542 shows a ~19% generational improvement, while the 64-core EPYC 7742 shows an impressive ~65% improvement.
NAG is acknowledged around the world as a leader in the provision of numerical libraries for technical computing, and as the trusted source of impartial expertise in the business and technical aspects of HPC. In May 2020, NAG will celebrate 50 years since its formation as an inter-University collaborative venture in the UK. Over the following decades NAG has grown into an international not-for-profit business, working collaboratively with partners in industry and academia to deliver robust algorithms via the flagship NAG Library, and other numerical computing software products. NAG also helps customers around the world in industry, government and academia to deliver the best impact at the best cost from HPC and technical computing, through expert consulting and services, including code performance and scalability, technology evaluation, benchmarking, training, cost models, strategic advice, and more.
AMD, the AMD logo, EPYC, and combinations thereof are trademarks of Advanced Micro Devices, Inc.