Intel is increasing the performance of its Aurora supercomputer to 2 ExaFLOPS, looking for ZettaFLOPS

This is not the first time that the performance of the Aurora supercomputer has been increased, after all, this system was originally prepared for the American Argonne National Laboratory as early as 2018, when it was to offer the performance of 180 PetaFLOPS. However, it was later announced that the system would not be ready until 2021, but its performance would reach 1 ExaFLOPS, thus beginning the so-called exascale era.

However, AMD has joined in and announced this year’s 1.5 ExaFLOPS Superier supercomputer, which will probably be completed on time and the El Capitan system with 2 ExaFLOPS is set up, so it is clear that Aurora will not be the first or most powerful in the new era. . Respectively, it was obvious, but we are learning that its performance is changing again and it is about 2 ExaFLOPS.

In addition, Intel plans to make Aurora the first system with such performance, or as such, it should arrive before El Capitan in 2023. We are talking here, of course, the performance in Double Precision, or FP64 (Rpeak), in which supercomputer performance is usually reported.

And how does Intel want to ensure that at all? It is said that this will also be possible thanks to the higher performance of the Ponte Vecchio accelerators than expected, which was confirmed by Pat Gelsinger. But will it be such that it ensures twice the performance of the whole system? We learn that thanks to the TSMC N5 process used, the powerful GPU component will have higher clocks as well as a higher number of active EU units per tile, as well as a higher number of accelerators themselves, which will be used throughout parts of the system.

In addition to the Ponte Vecchio, Aurora should also rely on Sapphire Rapids’ generation Xeon server processors, but GPU accelerators in particular provide the performance needed. And the Ponte Vecchio itself will first be deployed in Aurora, but will later be available for various HPC systems.

Specifically, they will also be the first European supercomputers based on SoC Rhea from SiPearl, which are being developed within the EIP – European Processor Initiative. There are basically 72 ARM Neoverse Zeus cores, and it was SiPearl that decided to also use the Ponte Vecchio accelerators and with them the necessary oneAPI.

And what about the zettascale era? Of course, it is still many years in the future, but Intel believes that it can bring it closer long before it takes the transition from petascale to exascale, or 12 years. He wants to have hardware for computers with performance in ZettaFLOPS already in 2027, but we will not know how he intends to increase the performance of the HPC system by a thousandfold in about five or six years.

Prices of related products:



Source: Svět hardware by www.svethardware.cz.

*The article has been translated based on the content of Svět hardware by www.svethardware.cz. If there is any problem regarding the content, copyright, please leave a report below the article. We will try to process as quickly as possible to protect the rights of the author. Thank you very much!

*We just want readers to access information more quickly and easily with other multilingual content, instead of information only available in a certain language.

*We always respect the copyright of the content of the author and always include the original link of the source article.If the author disagrees, just leave the report below the article, the article will be edited or deleted at the request of the author. Thanks very much! Best regards!