Micron and AMD have recently conducted benchmark tests on their new DDR5 memory and fourth-generation AMD EPYC processors for data centers. These tests were designed to measure the performance of high-performance computing (HPC) workloads, which are essential for solving complex problems in fields such as weather prediction, earthquake modeling, and chemical analysis.
In the past, HPC workloads were typically hosted on supercomputers, but with advances in computer architecture, they are now often run on large clusters of high-performance servers. These servers need to have the right combination of computing power, architecture, memory, and storage infrastructure to meet the scalability, low latency, and high-performance requirements of critical workloads. However, as the performance and throughput of server CPUs continue to increase, DDR4 memory is unable to provide sufficient memory bandwidth to support the growing number of high-performance cores.
That's where Micron's DDR5 memory comes in. The official benchmark tests showed that the DDR5 memory doubled the memory bandwidth compared to DDR4, providing much-needed support for the high-performance cores of the AMD EPYC processors. This is a significant improvement and will help data centers to run HPC workloads more efficiently and effectively.
According to the official benchmark tests, the combination of Micron's DDR5 memory and the fourth-generation AMD EPYC processor using the Zen 4 server architecture was able to double the memory bandwidth compared to DDR4. This was demonstrated through the use of the STREAM 1 benchmark tool, which measures the peak memory bandwidth of high-performance computing systems.
The increase in memory bandwidth was a significant improvement and will help data centers to run data-intensive workloads more efficiently and effectively. The combination of Micron's DDR5 memory and AMD's Zen 4 server architecture is a strong choice for those looking to run high-performance computing workloads at scale.
The official benchmark tests conducted on Micron's DDR5 memory and the fourth-generation AMD EPYC processor used the Alma 9 Linux kernel 5.14 and the STREAM.f version released on November 29, 2021 as the software stack for the workload. This software stack was chosen for its ability to accurately measure the performance of the memory and processor combination and provide reliable results.
The use of the Alma 9 Linux kernel and STREAM.f allowed for the testing of the DDR5 memory's ability to support the high-performance cores of the AMD EPYC processor, resulting in a two-fold increase in performance compared to DDR4. This improvement will be beneficial for data centers running data-intensive workloads, such as those in the fields of weather prediction, earthquake modeling, and chemical analysis.
The partnership between Micron and AMD has been successful in developing a strong combination of memory and processor for data centers looking to run high-performance computing workloads. The joint server laboratory in Austin has allowed for efficient memory verification and joint workload testing, ensuring the quality and reliability of the product.
The test setup included a DDR4 system equipped with a third-generation 64-core 3.7 GHz AMD EPYC processor and 64GB of RDIMM memory at 3200 MHz. The DDR5 system was equipped with a fourth-generation 96-core 3.7 GHz AMD EPYC processor and 64GB of RDIMM memory at 4800 MHz.
The results of the tests showed that the DDR5 system doubled the memory bandwidth per socket to 378 GB/s. This increase in memory bandwidth is significant and will allow customers to run larger artificial intelligence/machine learning (AI/ML) projects or take advantage of DDR5's increased memory bandwidth for more high-performance computing.
The combination of Micron's DDR5 memory and the fourth-generation AMD EPYC processor is a strong choice for data centers looking to run high-performance computing workloads at scale. The joint server laboratory in Austin, established by Micron and AMD, has allowed for efficient memory verification and joint workload testing, ensuring the quality and reliability of the product.
The HPC workload code used targeted weather and climate prediction. The WRF model, which performs well on architectures supporting high-performance floating-point processing, high memory bandwidth, and low-latency networks, was used to test the continental United States (CONUS) at a horizontal resolution of 2.5 kilometers.
The software stack for this workload included the Alma 9 Linux kernel 5.14, WRF 2.3.5 & 4.3.3, and Open MPI v4.1.1. The test setup included a DDR4 system equipped with a third-generation 64-core 3.7 GHz AMD EPYC processor and 64GB of RDIMM memory at 3200 MHz, and a DDR5 system equipped with a fourth-generation 96-core 3.7 GHz AMD EPYC processor and 64GB of RDIMM memory at 4800 MHz.
The results of the tests showed that the Micron DDR5 with the fourth-generation AMD EPYC processor was able to achieve 1.3567 time steps per second, compared to 2.8533 time steps per second for the DDR4 system. This faster performance means that data centers can use larger databases or run more models for weather predictions, improving the accuracy of the forecasts.
OpenFOAM, an open-source high-performance computing workload for computational fluid dynamics (CFD), was used to simulate the steady airflow around a motorcycle and rider. OpenFOAM is widely used in multiple industries, including consumer product design and aerospace design, to help reduce development time and cost.
OpenFOAM is able to perform load balancing calculations based on the number of processes specified by the user, allowing it to decompose the grid into multiple parts and assign them to different processes for a solution. After the solution is complete, the mesh and solution are recombined into a single domain.
The combination of Micron's DDR5 memory and the fourth-generation AMD EPYC processor is a strong choice for data centers looking to run high-performance computing workloads, such as those related to CFD. The joint server laboratory in Austin established by Micron and AMD has allowed for efficient memory verification and joint workload testing, ensuring the quality and reliability of the product.
the software stack used for the computational fluid dynamics (CFD) workload included the OpenFOAM CFD software (version 8) with a motorcycle mesh size of 600 x 240 x 240, the Alma 9 Linux kernel 5.14, and Open MPI v4.1.1.
The test setup included a DDR4 system equipped with a third-generation 64-core 3.7 GHz AMD EPYC processor and 64GB of RDIMM memory at 3200 MHz and a DDR5 system equipped with a fourth-generation 96-core 3.7 GHz AMD EPYC processor and 64GB of RDIMM memory at 4800 MHz.
The combination of Micron's DDR5 memory and the fourth-generation AMD EPYC processor is a strong choice for data centers looking to run high-performance computing workloads, such as those related to CFD. The joint server laboratory in Austin established by Micron and AMD has allowed for efficient memory verification and joint workload testing, ensuring the quality and reliability of the product.
The test results showed that Micron's DDR5 portfolio improved the performance of OpenFOAM by a factor of 2.4. OpenFOAM is a widely used high-performance computing software platform, particularly in universities and R&D centers, that can utilize high-bandwidth memory and high-performance CPUs with dense cores to achieve highly parallel operations.
Another software platform tested in the benchmark tests was CP2K, an open-source quantum chemistry tool that can be used for a variety of applications, including the simulation of solid-state biological systems. CP2K can provide a common framework for different modeling methods and was tested using the density functional theory (DFT) of water (H2O) with a simulation box containing a total of 6,144 atoms (2,048 water molecules).
The software stack used for the molecular dynamics workload included H2O-DFT-LS.NREP4 and H2O-DFT-LS, as well as the Alma 9 Linux kernel 5.14. The test setup included a DDR4 system equipped with a third-generation 64-core 3.7 GHz AMD EPYC processor and 64GB of RDIMM memory at 3200 MHz and a DDR5 system equipped with a fourth-generation 96-core 3.7 GHz AMD EPYC processor and 64GB of RDIMM memory at 4800 MHz.
The test results showed that Micron's DDR5 portfolio improved the performance of molecular dynamics by a factor of 2.1. The performance of such workloads is known to significantly increase as the number of cores and memory bandwidth increase.
Micron's DDR5 memory, combined with the fourth-generation AMD EPYC processor, has been officially tested for various high-performance computing workloads. The test results show that the DDR5 portfolio improves memory bandwidth by a factor of 2, improves OpenFOAM performance by a factor of 2.4, and improves molecular dynamics performance by a factor of 2.1. These improvements are significant and will allow data centers to run larger artificial intelligence/machine learning (AI/ML) projects, improve weather and climate prediction accuracy, and run more complex computational fluid dynamics simulations.
Only a small number of high-performance computing workloads have been tested so far, and the results are preliminary. However, the combination of high-performance, high-bandwidth memory with the latest server processors such as the fourth-generation AMD EPYC processors opens up new possibilities for HPC customers. The joint server laboratory in Austin established by Micron and AMD has allowed for efficient memory verification and joint workload testing, ensuring the quality and reliability of the product.
0 Comments