Intel has completed its one API cross-architecture tool suite. It will be made available to developers free of charge next month. The founder also announced the launch of its first dedicated GPU for the data center. Tested with Tencent, it is dedicated to cloud gaming and video streaming.
Intel officially launches its first dedicated GPU for servers this November 11th. Named “Intel Server GPU,” and based on the Xe LP (low power) architecture, it is intended for cloud gaming on Android and video streaming. It incorporated 8GB of DDR4 memory and was obviously designed to work in tandem with Xeon Scalable processors.
Intel states that eight of these GPUs (at a rate of four per PCIe x16 card from H3C) can handle more than 100 Android cloud gaming sessions simultaneously, with a maximum of 160 sessions depending on the games and server
configuration. Allen Fang, deputy general manager of Tencent’s XianYou cloud gaming platform, said he could surpass 100 gaming sessions on two-card servers for their most popular games: King of Glory and Arena of Valor.
This Intel Server GPU is now available for
sale. Next year, it will be followed by processors based on the “high power” variant of the Xe architecture, which will aim to compete with Nvidia’s A100 and “could exceed 40 TFLOPS,” according to Intel. A version dedicated to high-performance computing, called Xe HPC, is also planned.
A unified and rational software platform is essential to successfully establish itself on product lines ranging from the laptop System-on-Chip to the HPC
chip. Intel has been working for some time on cross-architecture tools (“XPU”) that simplify development on CPU, GPU, FPGA, and others while making the most of a system’s performance. The idea is to take advantage of GPU acceleration’s possibilities while running a un optimized x86 code. Called OneAPI, these tools have now gone “gold” and will be made available free of charge to developers in December.