He will be ready next year.
Despite the fact that Intel has not yet launched a single product with even entry-level Xe GPUs, a year ago the company introduced the Aurora supercomputer with such GPUs at its core. At the time of the formal announcement, Aurora was the first Exascale class supercomputer introduced, adjusted for the fact that it should be built only in 2021.
And now we have a lot of technical details of this installation. To begin with, the timing is still the same – in 2021. It is next year that the customer represented by the Argonne National Laboratory should receive a new installation. Whether Aurora will be the first supercomputer of the Exascale class at the time of launch is still unknown.
But we know that Aurora will consist of an unknown number of computing nodes, each of which will contain two Xeon processors of the Sapphire Rapids line and six GPUs of the Ponte Vecchio line.
At the same time, these six GPUs will be combined into a single graphics accelerator, that is, you can probably talk about one multi-chip graphics processor. More than 10 PB of shared system memory and the Cray Slingshot Fabric interconnect is also reported.
From the rest, it can be noted that the Xeon processors of the Sapphire Rapids line will be based on the updated Willow Cove architecture – the basis of the Tiger Lake processors. Such CPUs will be produced using 10-nanometer technology, while Ponte Vecchio GPUs will be produced using seven-nanometer technology.