High performance computing (HPC) has become something essential in many sectors. and scientific research is, without a doubt, one of the areas that has benefited the most from the advances that have been made in this regard in recent years. Part of these advances have been possible thanks to AMD EPYC processors, which offer a high level of performance and high efficiency, and AMD Instinct graphics accelerators have also played an important role.
To improve their research capabilities, universities and scientific entities around the world have to bet on high-performance computing. This is essential to continue advancing and to be able to face new challenges, but it also entails a major investment which, in the end, can end up placing said entities and educational centers in a difficult situation.
The Royal Stockholm Institute of Technology had to face that decision recently, and in the end he decided to bet on an HPE Cray EX cluster, which is configured with second generation AMD EPYC processors with 64 cores and 128 threads each, and with AMD Instinct graphics accelerators. This move has allowed them to access a total power of 13.5 PFLOPs, an impressive figure when compared to the 2.28 PFLOPs that their previous system was capable of offering.
If we analyze that power a little more in depth, and contextualize it with the consumption of the system, we find very interesting data. This cluster offers a performance at the CPU level of 6 GFLOPs per watt consumedwhile the graphics accelerator block is capable of reaching a performance of 60 GFLOPs per watt consumed. This translates into high efficiency, and places this system as one of the most efficient in the world.
At the hardware level we know that this cluster, an HPE Cray EX system called Dardel, has a total of 1,270 nodes in the part dedicated to AMD EPYC CPUs and 56 nodes in the section dedicated to graphics accelerators. Each CPU node has two AMD EPYC 7742 processors with 64 cores and 128 threads at 2.25 GHz, and such nodes are mostly accompanied by 256 GB of RAM, although some have 512 GB, 1 TB and even 2 TB of RAM. Each GPU node has four AMD Instinct MI250X accelerators, they use an AMD EPYC processor in single socket configuration and each has 512 GB of RAM.
This new supercomputer will allow the Royal Stockholm Institute of Technology to continue advancing in the investigation of very important things, ranging from a cure for Alzheimer’s to the development of a sustainable shipping system.
