Intel launches its first discrete graphics card for server data centers

Intel officially released its first discrete graphics card for server data centers, code-named SG1. The official name is simply "Server GPU", based on the Xe LP low-power micro-architecture, designed for high-density, low-time Designed for extended Android cloud games and streaming services.

This is also Intel’s third Xe LP architecture discrete graphics product. Previously, Iris Xe MAX (code-named DG1) for thin and light notebooks and desktops was already available. Tiger Lake and Rocket Lake 11th-generation Core processors also integrate Xe LP architecture cores. Core graphics card.

Intel emphasized that as the world enters the era of billions of smart devices, the amount of data is increasing exponentially, and the focus must be shifted from a single CPU to a hybrid architecture that spans CPU, GPU, FPGA and other accelerators, that is, " XPU" vision.

The launch of server GPUs is the latest step in the transformation and expansion of the product portfolio in the Intel XPU era. In the future, we will continue to see discrete graphics products with Xe HPG, Xe HP, and Xe HPC architectures, covering desktops, data centers, high-performance computing, etc. field.

Among them, the Xe HPG graphics card has been successfully lit, the Xe HP graphics card is in trial production and will be launched next year, and the Xe HPC graphics card is under development, code-named Ponte Vecchio.

The server GPU released this time has a 128-bit width pipeline in terms of specifications, with 8GB LPDDR4 onboard as the video memory, and the number of core execution units is not disclosed, but it is obviously 96 at most.

H3C (H3C) has created an expansion card "XG310" based on Intel server GPU, 3/4 length, full height, single slot form, 8-pin auxiliary power supply, equipped with four server GPUs, and system bus PCIe 3.0 x16.

Each server can install up to four such expansion cards, which is a total of up to 16 GPUs.

With Intel Xeon Scalable processor, it allows service providers to independently expand the graphics card capacity without changing the number of servers, supporting more media streams and subscribers, and at lower cost.

Intel said that a typical dual-card system can support more than 100 concurrent users of Android cloud games, and it can be expanded to 160 concurrent users.

Currently, Intel is working with many software and service partners to bring new server GPUs to the market, including Tencent, Gamestream, and Ubitus.

Tencent revealed that using Intel server GPUs, more than 100 game instances can be generated on each dual-card server, and mainstream games have been tested, including "Glory of the King", "Peace Elite", "Legend Showdown", etc. , The effect is good, and the more the number of games, the lower the cost.

In terms of video streaming, Intel server GPUs support multiple video encoding formats, including AVC, HEVC (H.264), MEPG2, VP9, ​​AV1, and AV1 now supports video decoding.

And with Xeon extensible processor, media plug-in, it can also support FFMPEG.

In terms of software development, Intel simultaneously announced the latest oneAPI Gold toolkit. Developers can access Intel XPU through a common, open, industry-standard programming model, release the performance potential of the underlying hardware, reduce software development and maintenance costs, and are deploying The risk of accelerated computing is lower.

The Intel oneAPI project aims to achieve a unified and simplified cross-architecture programming model that can provide uncompromising performance, is not limited to a single vendor's dedicated code, and can achieve the integration of original code. Developers can also choose the best accelerator architecture for a specific problem, and there is no need to rewrite the software after a new architecture and platform.

The oneAPI Gold toolkit will be available for free locally and on Intel DevCloud in December, and a commercial version with global support from Intel technical consulting engineers will also be provided. Intel Parallel Studio XE and System Studio tool suite will also be migrated to oneAPI products.

The Intel DevCloud platform has also added new Iris Xe GPU hardware support. Developers can also openly access Iris Xe MAX discrete graphics cards, and Xe HP has been opened to specific developers.