The first supercomputer from an exaFLOP: Arriving in 2021
The combination of Intel Xeon Scalable CPU, Optane DC Pwrsistent memory and Intel Xe GPU behind the first supercomputer capable of exaFLOP power.
The American Department of Energy (DOE) has announced that in the course of 2021, it will implement Aurora, a name that indicates the future supercomputer developed by Cray that will combine different types of Intel architectures to reach a level of processing power equal to an exaFLOP. Designed for HPC and artificial intelligence, this supercomputer will be used for different types of uses including the development of climate models and cosmological simulations.
To achieve this, Aurora will see the use of Intel Xeon processors, integrated into Cray’s proprietary supercomupter system named Shasta. In total, more than 200 cabinets will be installed, connected to each other through the Slingshot interconnection system always developed by Cray. Next to Intel Xeon Scalabole processors we also find Optane DC Persistent Memory and this is the real novelty, Intel GPU of the Xe family.
These are discrete Intel solutions not integrated within the processors, of which the company has anticipated the future existence on the occasion of its Architecture Day at the end of 2018 but without providing specific details. However, we know that these proposals are developed by Intel thinking of the need to process the datacenter: it is possible that Intel will also be able to derive a family of products for gamers from these solutions.
The debut of the first Intel Xe solutions is expected by 2020; it will be in 2021 that the Aurora system will be completed and installed at the Argonne National Laboratory. The total investment is indicated at over 500 million dollars: this is the first announcement related to Intel’s next-generation graphics solutions, and we imagine that it will not be the only one of this type.
