laitimes

Independent IP authorization helps GPUs to replace domestically

author:Fat and thin 0072003

In recent years, domestic GPU companies have sprung up. Jingjiawei, Biren Technology, Moore Thread, Muxi Integrated Circuit and so on. New products have been released, so it is not busy.

On March 30, Moore Thread launched several new products in Beijing: the first unified system architecture MUSA, the first generation of full-featured GPU chip "Sudi", the first GPU desktop graphics card MTT S60, the first server computing card MTT S2000 for data centers, and the GPU physics engine AlphaCore.

Independent IP authorization helps GPUs to replace domestically

A key part of moore's thread chip design is licensed for IP from Apple's supplier, UK-based Imagination Technologies. Imagination Technologies is a well-known BRITISH chip designer whose Imagination PowerVR graphics technology has been licensed to many of the world's leading companies dedicated to providing key IP building blocks to support advanced and comprehensive solutions. Imagination has its own CPU IP in addition to its own GPU IP.

In 2017, American private equity firm Canyon Bridge filed an acquisition of Imagination. Canyon Bridge is backed by China Guoxin Holdings Co., Ltd., and the deal ended at £550 million.

Independent IP authorization helps GPUs to replace domestically

Nowadays, the main general computing + graphics GPU startups in China, such as Core Motion, Moore Thread, Biling, etc., all use Imagination IP or VeriSilicon authorized IP, which benefit from IMG's inexpensive IP library and licensing policy. The hardware specifications of each family do not differ much, and the final spelling is the adaptability of software and applications.

At present, the IP licensing of domestic graphics processors is either ARM Mali's Architectural, or VeriSilicon's Vivante, or Imagination lP.

Independent IP authorization helps GPUs to replace domestically

VeriSilicon's scalable Vivante graphics processing unit (GPU) IP family ranges from low-power, small IoT MCUs to powerful SoCs for automotive and computer applications. VeriSilicon's graphics processor technology supports the industry's mainstream embedded graphics acceleration standards Vulkan 1.0, OpenGL 3.2, OpenCL 1.2 EP/FP, and OpenVX 1.2, with autonomous and controllable instruction sets and dedicated compilers, supporting 250 billion floating-point operations per second and 128 parallel shader processing units.

The level of VeriSilicon Vivante IP is a bit worse than ARM, but it is expensive in autonomy. Its most important customer in China is Jingjiawei, Jingjiawei has completed the attack on the GPU market under the layers of barriers of foreign construction, and is also the only GPU manufacturer to achieve large-scale mass production in China. Recently, Jingjiawei decided to focus on the company's main business of GPU chips, terminate the general-purpose chip research and development and industrialization projects for the consumer electronics field, and concentrate on doing big things. This was followed by the recruitment of large software teams to develop graphics libraries and parallel computing.

Independent IP authorization helps GPUs to replace domestically

However, some insiders pointed out that there are not many GPU design companies or manufacturers specializing in the GPU field that we can see, the most prominent of which are Jingjiawei and Zhaoxin. The size of the mainland market and the number of manufacturers are not equal, and it is necessary to wait for industrial integration and concentrate resources in order to innovate.