Intel Xe (architecture of GPU)

2019: Announcement

On November 17, 2019 within the Supercomputing 2019 conference in Denever (USA) the Intel company provided new architecture for server graphic processors. She received the name Xe, and the first solution on its basis — Ponte Vecchio.

It is the discrete graphic processor ground, as well as all architecture of Xe, for high-performance computing (HPC) and machine learning.

Intel provided new architecture for server graphic processors

GPU will be made on 7-nanometer technology process which was mastered by AMD earlier. However, if AMD she intends to release 7-nanometer chips in 2020, then Intel — in the 2021st, notes the PCWorld edition.

Ponte Vecchio uses space configuration of Foveros — the same that in processors for devices with two Lakefield displays.

Also in Intel pay attention to use of semiconductor bridges of EMIB (Embedded Multi-Die Interconnect Bridge; allows to integrate crystals which perform different functions: computing crystals, crystals of memory and input-output), high-speed memory of HBM (the version is not specified) and fast interconnect of CXL (Compute Express Link). On the basis of the last the Xe Link interface for communication between GPU will be created.

The technology CXL minimizes sharing of resources for performance improvement, reduces complexity of a program stack and reduces the total cost of a system.

Intel intends to separate graphic accelerators on architecture of Xe into two families: in one of them solutions for data centers, in another — for consumer video cards and the built-in graphics will be presented.

In new processor architecture the unified oneAPI API interface which is earlier not used by Intel which, as envisioned by developers, is designed to simplify programming for CPU, FPGA, GPU and other accelerators by means of uniform tool kit is applied.^[1]

Notes