A supercomputer is a system that operates at the highest operating rate for computers. Usually, these have been used for engineering and scientific applications that need to manage huge databases or perform a large amount of calculations. The US Department of Energy "Argonne National Lab" named Nvidia's A100 graphics card paired with AMD's microprocessors for its latest Polaris processor, a move deemed critical due to Intel's latest Aurora-powered supercomputer. The system has encountered a problem due to manufacturing difficulties with Intel's Sapphire Rapids server chips.
Argonne National Laboratory
ANL (Argonne National Laboratory) is a national engineering and science research laboratory operated by Argonne for the United States Department of Energy. Located outside of Chicago, the facility is the largest national laboratory by scope and size in the Midwest. Argonne Laboratory paves the way for the first tested exascale supercomputer, Polaris. Here are some more details on Nvidia vs AMD 2021 future supercomputer.
Performance
ANL's latest entry will have two AMD processors per node and four Nvidia A100s for a total of 2,240 Nvidia graphics cards spread across 560 nodes. It will provide PetaFLOPS with up to 44 of FP64 performance, which means it will be a much smaller powerful device than the intentional Aurora supercomputer exascale that will provide up to a single ExaFLOP of continuous performance when it will reach by the end of 2022 or the beginning of 2023.
Unlike, Polaris will scale to a performance level of 1.4 "AI ExaFLOPS", which is not estimated with the normal FP64 load used to count supercomputer performance. This means that Polaris is not an exascale-class device. However, we have to see a deliberately Nvidia-powered computer that will hit that specific mark. Still, Polaris PetaFLOPs 44 performance is eligible for a spot in the top 10 of the Top 500 list of the fastest supercomputers in the world.
Features
The computer will spur transformative scientific research, such as research into particle collisions, the exploration of clean energy, and the advancement of cancer treatments to discover new methods of physics. In addition, it will bring the ALCF (Argonne Leadership Computing Facility) into the era of exascale artificial intelligence by allowing researchers to improve their scientific missions for Aurora, the future exascale system of Argonne.
Design specifications
As informed, we can say that Polaris should use around 2 megawatts of power at full power, which is down from Aurora's 60 megawatts. Speaking of Aurora supercomputer specs, Polaris will come with 560 knots compared to Aurora's deliberate 9,000 knots. The supercomputer will also use Cray's Sling networking hardware, which means the system is trained by HPE (Hewlett Packard Enterprise).
High Performance Computing Wire said the Epyc Rome 32-core 7532 computers will be used initially, then modified in March 2022 to the latest Epyc Milan 32-core 7543 chips. a 10 Slingshot to the 11 Slingshot Cloth to meet Aurora, verifying that Intel fails to interact with Aurora. It uses 40 Gen 10 Apollo server racks.
The AMD supercomputer delay is another setback for Intel's previous HPC attempts so far. In fact, Aurora was originally planned as a much lighter 180-PetaFLOP device debuting in 2018, but Intel Xeon The Phi processors were late and eventually canceled, leading to a complete restructuring of the Aurora system into an Exascale system powered by Intel Ponte Vecchio graphics cards and Sapphire Rapids graphics cards. In fact, Intel has just revealed that it has implemented a key feature of its Ponte Vecchio graphics cards, the Tile Xe, which boosts communication between collected computer parts, explicitly at the request of the DoE (Department of Energy) to Aurora. This determination required a whole year of progress work.
The Department of Energy has pledged that Aurora's Second Amendment will be delayed from its innovative 2021 launch, which is no surprise given that Intel's efforts with its upgraded Superfine node of 10 nanometers, renamed Intel 7, used for Rapids Sapphire processors. Now, as a result, Aurora is planning placement in the 2022-2023 timeframe, handing the tag of the world's first exascale computer to Frontier pushed by AMD. In addition, AMD's device will continue to advance and be the fastest.
With its diverse GPU-CPU architecture, Polaris is helping Argonne transition to the HPE Intel Aurora system, which falls on Intel's 2021-2022 roadmap delays (impacting Ponte Vecchio and Sapphire Rapids). It will be used by scientists from the Department of Energy's Exascale Computing Project as well as the ALCF Science Program to boot their codes for Aurora.
Advantages
Final Verdict
The new supercomputer has multiple graphics card nodes and supports programming devices. It has a similar Interconnect Slingshot that will be available on Aurora. Early Science Program works in normally traditional HPC simulation, but also in learning space and data. Thus, the company wanted optimized Python support, a number of optimized frameworks, and things accessible on Aurora for these applications to develop. This is available with HPE and Nvidia solutions. In the meantime, Polaris is now in the final stages of set-up and will be ready for initial research in 2022, with expanded availability to the study community in Q2 2022. We hope you enjoyed this review. We will come soon with another one.
