In the newly dropped list of TOP500 supercomputers, AMD and NVIDIA have taken the lead and have their game on a high when it comes to high-performance computing technology. Eight of the 10 fastest supercomputers are being powered by NVIDIA’s GPU accelerated systems, and 4 of the 50 highest-performance supercomputers across the world are being powered by AMD’s EPYC processors, as claimed by AMD.
NVIDIA’s 8th generation GPU design which is also their first based on architecture design “A100” has been a real game changer for them in the supercomputing technology. This A100 is considered as one of the biggest generational leaps by NVIDIA, by any of its GPUs ever. This multi-instance GPU has more than 54 billion transistors, and is designed for data centers, and is also optimized for interference and HPC. It also delivers speed improvements as much as twenty times over Volta and also has 3rd generation Tensor cores.
The new InfiniBand interconnect technology from Mellanox, which was bought by NVIDIA in 2019 for $7 billion is also included in the DGX A100, which is their latest DGX system. For this purpose, NVIDIA integrated the A100 with InfiniBand switches of Mellanox to create Selene, that has made it to No. 7 spot in the TOP500 list in its debut. NVIDIA claims that more than 1 exaflop of AI processing is provided by Selene, which was designed in a time less than a month. They also claim it to be the fastest industrial supercomputer in the US, and that it is also very energy-efficient.
We’ve entered a new era, said NVIDIA product marketing director Paresh Kharya in a briefing. He said it by referring to the point that their scientific programming has changed and that is why NVIDIA is so successful in the TOP500 list. He claimed that this new era includes expansion beyond traditional modelling and simulation workloads to include AI, data analytics, edge screening, and big data visualization. He said that by combining with Mellanox, NVIDIA’s position is stronger than ever.
Meanwhile, in the high performance computing market, AMD is touting momentum continuously as is shown in their recent claim of their EPYC processors powering 4 out of 50 highest super-performance supercomputing systems across the globe. The new HPC installations at Purdue University, CERN and Indiana University, all have the new processors of AMD. Not only this, the HPC cloud instances from Google, Oracle cloud and Amazon Web services all have AMD’s processors.
The four EPYC systems that have been named in the TOP500 list are Belenos, Mahti, Joliot-Curie and Uninett sigma2. All four of these come from French firm Atos.
It is also known now that the upcoming fastest exascale supercomputers, El Capitan and Frontier, will both have EPYC processors by AMD. El Capitan is specifically significant as it is being developed to protect US nuclear stockpile, and it is expected to come online in 2023. Frontier will be used at Lawrence Livermore National Laboratory and is expected to arrive there in 2021.
AMD says that they are helping the industry advance towards the exascale supercomputing technology through their open software programming environment, along with their EPYC CPUs and Radeon Instinct accelerators. Forrest Norrod, GM and SVP of AMD’s data center and embedded systems group, said that through their cloud computing environments and their support of the top supercomputing clusters, AMD are proud to strengthen the global HPC ecosystem.