In the annual Supercomputing event, Intel has announced that its next-gen Xeon processors, codenamed Sapphire Rapids, will feature 64GB of HBM2e memory. In addition to that, the upcoming GPUs, codenamed Ponte Vecchio, will feature 408MB of L2 cache. The company also revealed that it is working with a broad range of partners to bring OEM systems to market with both Sapphire Rapids and Ponte Vecchio.
Intel also revealed that its flagship supercomputer, Aurora, will come with 54,000 Ponte Vecchio GPUs that will be paired with 18,000 Sapphire Rapids processors. Aurora is a project that Intel has been working on for quite some time now. After the Xeon Phi was scrapped, Intel finally decided to use the Sapphire Rapids processors for Aurora.
Each unit of Aurora will feature two of the newest Sapphire Rapids CPUs that will include four compute tiles, DDR5, with PCIe 5.0 connectivity. Aurora will also be using SPR with built-in High Bandwidth Memory (SPR+HBM) with 64GB of HBM2e using 8-Hi stacks.
Sapphire Rapids is based on the Golden Cove architecture that is etched onto the Intel 7 process. The processors support AMX, Advanced Matrix eXtensions, that are targeted at improving performance in training and inference workloads. Intel will be delivering the first batch of the hardware to the Argonne National Laboratory by the end of the year. Intel is expected to deliver the rest of the machines through 2022. The company is also trying to ramp up the production for mainstream use through Q1 2022 for a wider spread launch.