Intel unveils Xe graphics cards for servers and cloud gaming

Intel has previously shown product plans for the various architectures included in the collective name Xe. These range from the simple and energy-efficient circuits in the Xe-LP, which includes the Iris Xe Max, up to game-focused Xe HPG and at the server and supercomputer level Xe HP / HPC. However, the company’s goal is not only to have different capable circuits for different purposes, but to let developers scale up code according to the Xe capacity available.

The hub of the vision is the One API, a model and collection of interfaces that allow developers to write code that can scale up and down depending on the capacity available, applicable graphics cards, processors, programmable circuits (FPGAs) and other available computing devices. The company calls the model “XPU”, ​​as the name implies, One API is an interface that makes all types of computing devices available to developers.

Intel is demonstrating the concept by also unveiling the graphics card H3C XG310 GPU, where four graphics circuits of the energy-efficient type Xe-LP take place. Xe-LP is also used in laptops in the form of Intel Iris Xe and Iris Xe Max. Each individual circuit is equipped with 8 GB of LPDDR4 graphics memory. Since up to four H3C XG310 GPUs can run in parallel, it allows servers to run 16 graphics circuits and 128 GB of graphics memory in parallel.

The graphics cards are connected through PCI Express 3.0 x16 slots. A graphics card supports 60 contemporary players for heavier Android games streamed via cloud services, and 90 players for simpler titles. Four parallel graphics cards thus increase this to 240 simultaneous players in a Server GPU configuration. According to the company, the largest development interfaces such as DirectX and Vulkan in One API are already supported today, and starting next year, they will also be supported as part of the streaming video functions in One API Video Processing.

In addition to scalability for server-side graphics cards, Intel also highlights some other innovations One API introduces. One such is low-level compilation with the Intel Implicit SPMD Program Compiler (ISPC), which runs on top of a layer called One API Level Zero. As the name suggests, it is an abstraction layer that provides direct access to the hardware, and thus provides reduced latency compared to traditional high-level interfaces. ISPC is a variant of the programming language C and one of the functions it is used for is calculations of ray tracing with Intel OSPRay on the company’s server graphics card.

The next One API feature with a focus on games is what is called Project Flipfast, whose goal is to improve the gaming experience for Linux games running at the server level. With Project Flipfast, a Linux system running in a virtual machine can still have full access to the full performance of the graphics card. It also enables the host system and the virtual guest systems to share data without performance penalties. The purpose of Project Flipfast is simply to give game developers access to more of the hardware performance of streaming Linux games.

Intel announces that gaming companies such as Tencent Games and Ubitus are early partners for the H3C XG310 GPU and One API. The graphics cards are being delivered to partner companies now and the sharp version of One API will be available to developers in December 2020.

Read more about Intel Xe: