Intel demonstrated numerous yet-to-be-released merchandise at its annual Intel Imaginative and prescient occasion. These merchandise are the upcoming Meteor Lake processors for shopper PCs due in 2023, Sapphire Rapids CPU for datacenters and supercomputers, and Ponte Vecchio compute GPU for synthetic intelligence (AI) and high-performance computing (HPC) functions. For the primary time Intel publicly demonstrated its Sapphire Rapids with HBM2E reminiscence.
Whereas Intel’s Meteor Lake, Sapphire Rapids, and Ponte Vecchio are processors designed for utterly completely different functions, there’s a factor that they’ve in widespread: all of them use multi-tile design. That is Intel’s newest chip constructing paradigm and was launched in mid-2020 and is ready to come back to fruition in 2022 ~ 2023.
Intel showcased its Meteor Lake take a look at chips final November, however again then the corporate demonstrated CPUs in miniature packaging geared toward low-power laptops in addition to 2-in-1 cellular techniques. This time round, the corporate displayed Meteor Lake processors in common packages for mainstream laptops, primarily based on pictures printed by PC Watch. In the meantime, it’s nonetheless unclear what sort of packaging desktop Meteor Lake merchandise are set to make use of.
Intel showcased its 4th Era Xeon Scalable ‘Sapphire Rapids’ processor with HBM2E in mid-October 2021. At the moment Intel demonstrated a CPU with every tile geared up with two HBM2E stacks. At its Imaginative and prescient 2022 occasion Intel uncovered a Sapphire Rapids with HBM2E that featured just one reminiscence stack per channel. Whereas we have no idea for certain, it seems like Intel plans to supply two forms of Sapphire Rapids with onboard HBM2E reminiscence: one that includes a single HBM2E reminiscence stack per CPU tile and one other with two HBM2E stacks per tile. Holding in thoughts that Sapphire Rapids can work with HBM2E reminiscence in several modes, it may be logical to supply two variations of the CPU for various functions.
Intel’s Ponte Vecchio is arguably the head of Intel’s multi-tile design method and guarantees to be one of many quickest compute GPU unveiled so far. The corporate has been displaying its Ponte Vecchio accelerator consisting of 47 tiles and ~100 billion transistors for a few 12 months now, so it isn’t significantly stunning that Intel displayed it at its Imaginative and prescient 2022 convention.
Intel’s Sapphire Rapids processor with HBM2E and Ponte Vecchio compute GPU are set to energy Argonne Nationwide Laboratory’s Aurora supercomputer rated for an as much as 2 FP64 ExaFLOPS throughput. Each merchandise are at present sampling with choose clients, so it’s pure to display them at an occasion. The one factor that’s maybe stunning is that Intel determined to not showcase them in motion.
“We are beginning to construct [Aurora],” mentioned Rick Stevens, affiliate laboratory director at Argonne Nationwide Laboratory at Intel Imaginative and prescient 2022. “Every Aurora blade could have two Sapphire Rapids CPUs linked to 6 Ponte Vecchio GPUs. It’s taking form in an enormous 10,000-square-foot space the place the crew is busy putting in racks, the storage system and cooling infrastructure.”