At the ISC High Performance 2021 event, Intel held its "Accelerating the Possibilities with HPC" keynote where it talked about its next-gen server and HPC platform. The company first started talking about its Sapphire Rapids server CPUs and Ponte Vecchio graphics back at SC19, however, this time Intel has provided a lot more detail.
First up is the Sapphire Rapids. As mentioned earlier, it's Intel's next-gen server processor family which will succeed its Ice Lake Xeon scalable processors. Like Alder Lake will do in the mainstream consumer market, Sapphire Rapids will also bring several new features in the server space which include:
- DDR5 memory
- PCIe 5.0
- Compute Express Link (CXL) 1.1
Sapphire Rapids will also introduce for the first time the integration of HBM for CPU use, something that has generally been reserved for graphics and acceleration.
In terms of features, the next-gen Xeon will carry Advanced Matrix Extensions (AMX) DL Boost that will help in deep learning (DL) inference and training. It will also pack Intel Data Streaming Accelerator (DSA), the company's next-gen I/O acceleration technology.
Intel hopes to start production of Sapphire Rapids in Q1 2022 and plans to ramp up production in the next quarter.
The company also detailed its Xe HPC plans a bit more. According to the firm, Ponte Vecchio, which will be the GPU powered by the Xe HPC architecture, is currently running validation testing.
Ponte Vecchio is being designed to work in the OCP Accelerator Module (OAM) form factor which will make scaling the accelerators easy work for enterprise customers of Intel. For example, above (image on the right) you can see an x4 subsystem of Ponte Vecchio GPUs.
Ponte Vecchio will be Intel's first Exascale GPU and its power, paired with Sapphire Rapids, will be leveraged inside upcoming supercomputers like Aurora.