Intel Sapphire Rapids Xeon with up to 64GB of HBM2e memory
Today at Supercomputing 2021 Intel made a new disclosure on 3rd Gen Xeon Scalable CPUs known as Sapphire Rapids.
Intel already revealed it will launch two types of Xeon’s based on Sapphire Rapids architecture, either with or without HBM2e memory. Today at SC2021 Intel confirmed that the latter will feature four stacks of HBM2e memory, each with 16GB of capacity. This means that in total, Xeon Scalable will feature up to 64GB of memory.
This disclosure is not exactly news to VideoCardz readers. We have already revealed this fact many months ago. The Sapphire Rapids CPUs based on the Eagle Stream platform is to feature up to 56 cores and TDPs up to 350W. Next-Gen Xeon platform is also to expand Intel support for PCIe Gen5 and DDR5 technology.
Furthermore, Intel confirmed L1 and L2 cache configurations of Ponte Vecchio HPC GPU. This is a data center accelerator that will be used alongside Xeon Sapphire Rapids in supercomputers such as Aurora. The Ponte Vecchio GPU is to offer 4MB of L1 Cache and 144 MB of L2 cache per tile. In total it will offer 408 MB of L2 cache. This GPU is to feature multiple fabrication nodes, such as Intel 7 for the base tile, TSMC N7 for Xe-Link, and TSMC N5 for computing tile.
Aurora Supercomputer, sponsored by the United States Department of Energy (DOE) and designed by Intel and Cray for the Argonne National Laboratory will feature more than 54,000 Ponte Vecchio GPUs and 18,000 Sapphire Rapids processors.
Intel also revealed a roadmap for its data center product. It confirms that the company is already planning ahead and will launch a successor to each of the 2022 products in the following year. It is, however, very short on details to say the least. But in Intel’s defense, neither did AMD confirm its plans for post-2022 data center products (except for EPYC Bergamo).
Source: HardwareLuxx, AnandTech