Intel demonstrated a number of yet-to-be-released products at its annual Intel Vision event. These products are the upcoming Meteor Lake processors for client PCs due in 2023, Sapphire Rapids CPU for datacenters and supercomputers, and Ponte Vecchio compute GPU for artificial intelligence (AI) and high-performance computing (HPC) applications. For the first time Intel publicly demonstrated its Sapphire Rapids with HBM2E memory.
While Intel's Meteor Lake, Sapphire Rapids, and Ponte Vecchio are processors designed for completely different applications, there is a thing that they have in common: they all use multi-tile design. This is Intel’s latest chip building paradigm and was introduced in mid-2020 and is set to come to fruition in 2022 ~ 2023.
Intel showcased its Meteor Lake test chips last November, but back then the company demonstrated CPUs in miniature packaging aimed at low-power laptops as well as 2-in-1 mobile systems. This time around, the company displayed Meteor Lake processors in regular packages for mainstream laptops, based on images published by PC Watch. Meanwhile, it is still unclear what kind of packaging desktop Meteor Lake products are set to use.
Intel showcased its 4th Generation Xeon Scalable 'Sapphire Rapids' processor with HBM2E in mid-October 2021. At that time Intel demonstrated a CPU with each tile equipped with two HBM2E stacks. At its Vision 2022 event Intel exposed a Sapphire Rapids with HBM2E that featured only one memory stack per channel. While we do not know for sure, it looks like Intel plans to offer two types of Sapphire Rapids with onboard HBM2E memory: one featuring a single HBM2E memory stack per CPU tile and another with two HBM2E stacks per tile. Keeping in mind that Sapphire Rapids can work with HBM2E memory in different modes, it might be logical to offer two versions of the CPU for different applications.
Intel’s Ponte Vecchio is arguably the pinnacle of Intel's multi-tile design approach and promises to be one of the fastest compute GPU unveiled to date. The company has been showing its Ponte Vecchio accelerator consisting of 47 tiles and ~100 billion transistors for about a year now, so it is not particularly surprising that Intel displayed it at its Vision 2022 conference.
Intel's Sapphire Rapids processor with HBM2E and Ponte Vecchio compute GPU are set to power Argonne National Laboratory's Aurora supercomputer rated for an up to 2 FP64 ExaFLOPS throughput. Both products are currently sampling with select customers, so it is natural to demonstrate them at an event. The only thing that is perhaps surprising is that Intel decided not to showcase them in action.
"We are starting to build [Aurora]," said Rick Stevens, associate laboratory director at Argonne National Laboratory at Intel Vision 2022. "Each Aurora blade will have two Sapphire Rapids CPUs connected to six Ponte Vecchio GPUs. It’s taking shape in a massive 10,000-square-foot area where the team is busy installing racks, the storage system and cooling infrastructure."