Intel shuts down Xe-based graphics chips for servers

When Intel unveiled the company’s strategy for the graphics circuits in the Xe series in the summer of 2020, these were divided into four different variants with the names LP, HP, HPC and HPG. The latter is the graphics circuit that takes place in the imminent game-oriented graphics cards Arc, while HP and HPC instead target data centers and servers.

Specifically, the HP series, also called “Arctic Sound”, is a scalable platform that is produced in what is called tiles, where each tile contains 2,048 cores. Tests with preliminary copies have been demonstrated to be capable of a computing power of about 10 TFLOPS, where the worsting variant with four tiles can handle upwards of 40 TFLOPS or higher. Now Intel’s graphics manager Raja Koduri announces that Xe HP will not be a commercial product for either companies or consumers.

Unlike the rest of the Xe family, which are wholly or partly manufactured by TSMC on their nodes, the HP series used Intel’s 10 nm Enhanced Superfin process. The fact that Intel chooses not to further develop that segment also means that they have no commercially available Xe product left that is completely self-produced.

Xe HP survives in, among other things, Intel’s cloud platform Devcloud, but it nevertheless leaves a certain void that it is currently unclear how Intel intends to fill. The HPG series can do that in theory, but without it tileconfiguration that the HP series has, the corresponding performance is unlikely to be achieved. The HPC series “Ponte Vecchio” on the other hand has the opposite problem, where the large circuits are expensive to manufacture and intended for even heavier calculations.

Source: Anandtech

Read more about Intel’s graphics investment: