Benchmarks for comparison in multiple HPC applications

At the annual Supercomputing Conference (SC), there has always been an unofficial theme. In recent years, the focus has centered on Machine Learning and Deep Learning, following a period where large-scale data-intensive computing and the potential of cloud computing to reshape supercomputing were key topics. These themes have one thing in common: they are not about CPUs themselves, but rather about performance improvements and ecosystem development under the X86 architecture. Last year, hardware became a central topic at the conference, especially with the rise of GPU-based supercomputers entering the Top 500 list. However, these were still accelerators, not the main processors. This year at SC17, the core processor returned as an unofficial highlight. The ARM-based hardware and software ecosystem was widely showcased, with Cray demonstrating fully integrated systems capable of competing with Intel’s latest offerings. The ARM-based “Isambard” supercomputer, set to launch next year at the University of Bristol, will feature 10,000 cores using Cavium’s ThunderX2 ARM processors. The team behind Isambard has long been working on ARM-based HPC systems, conducting extensive research and development. One of the most well-known ARM-based HPC projects is the Mont Blanc system at the Barcelona Supercomputing Center, which initially used dual Cortex-A15 ARM chips and now runs on ThunderX2. McIntosh-Smith and his team released benchmark results comparing a Cray 8-node cluster with a 32-core ThunderX2 ARM processor against Intel Skylake and Broadwell solutions. The results showed that memory-intensive applications ran significantly better on ThunderX2 than on Skylake. For floating-point intensive tasks, Skylake had an edge due to its wider vector units, but ThunderX2 performed comparably to Broadwell. If high-bandwidth memory continues to improve, the results could become even more impressive. The benchmarks for OpenFOAM, an open-source CFD application commonly used in HPC, showed that ThunderX2 outperformed both Skylake and Broadwell. Similarly, weather and climate simulations, such as those using Nemo, demonstrated improved performance on memory bandwidth-heavy applications. McIntosh-Smith noted that while compute-intensive applications like GROMACS, CP2K, and VASP show smaller performance gaps between processors, memory-bandwidth-heavy applications clearly reveal differences. This is because, although X86 processors benefit from wider vector units, ThunderX2 compensates with more cores and higher clock speeds. With the introduction of the high-end Cray XC50 system, we can expect more real-world performance data from ARM-based supercomputers at future SC conferences. The Isambard project architecture, featuring the ThunderX2 processor integrated with the Aries interconnect chip, demonstrates Cray’s strong commitment to ARM in HPC. McIntosh-Smith believes that future ARM options may adopt similar deployment strategies. He predicts that ARM-based HPC will eventually match the vector capabilities of other CPU manufacturers. The next generation of ARM processors is expected to offer vector widths comparable to those of other vendors. What’s particularly interesting is that these results were achieved with minimal optimization—just a few hours of fine-tuning. McIntosh-Smith emphasized that this rapid progress is the result of years of dedicated work on ARM-based HPC systems. As hardware matures, the software required for HPC on ARM will advance quickly. With the official release of ARM processors for supercomputers, the industry is entering a new era of architectural transformation. The ARM architecture is set to open a new chapter in HPC, offering a wider range of processor choices.

High Frequency Transformer

High Frequency Transformer,High Voltage High Frequency Transformer,High Requency Transformer Used For Machine,High Power High Frequency Transformer

Guang Er Zhong(Zhaoqing)Electronics Co., Ltd , https://www.geztransformer.com