Development prospect of GPU industry in 2020

2020/11/17

Looking for the footprints of development from the world's giants

Function and classification of GPU

GPU (graphics processing unit, graphics processor) is also known as the display chip. It is mainly used in personal computers, workstations, game hosts and mobile devices (smart phones, tablet computers, VR devices) to run graphics operations.

Structure determines that GPU is more suitable for parallel computing. The main difference between GPU and CPU lies in the on-chip cache architecture and the structure of digital logic operation unit: the number of GPU cores (especially Alu computing units) is far more than that of CPU, but its structure is simpler than that of CPU, so it is called multi core structure. The multi-core structure is very suitable for sending the same instruction stream to the multi-core in parallel, using different input data to execute, so as to complete the massive and simple operations in graphics processing, such as the same coordinate transformation for each vertex, and calculating the color value of each vertex according to the same lighting model. GPU makes use of its advantages of processing massive data, and makes up for the shortcoming of long latency by improving the total data throughput.

Generally speaking, consumers will pay more attention to the performance of CPU (central processing unit) when purchasing consumer electronic products, such as mobile phones or laptops, such as the brand, series and number of cores of CPU, while GPU receives less attention. GPU (graphic processing unit), as well as graphics processor, is a kind of microprocessor which can do image and graphics related operations on personal computers, workstations, game machines and some mobile devices (such as tablet computers, smart phones, etc.). At the beginning of the birth of PC, there was the idea of GPU, and all graphics calculation was done by CPU. However, the speed of using CPU to do graphics calculation is slow, so a special graphics accelerator card is designed to help with graphics calculation. Later, NVIDIA proposed the concept of GPU, which promoted the GPU to the status of a separate computing unit.

CPU is generally composed of logic operation unit, control unit and storage unit. Although the CPU has multiple cores, the total number is not more than two digits, and each core has enough cache; the CPU has enough number and logical operation units, and has many hardware to accelerate branch judgment and even more complex logical judgment. Therefore, the CPU has super logical ability. The advantage of GPU lies in multi-core, the number of cores is far more than that of CPU, which can reach hundreds, each core has relatively small cache, and the number of digital logic operation units is small and simple. Therefore, GPU is more suitable for data parallel computing than CPU

There are two ways to classify GPU, one is based on the relationship between GPU and CPU, the other is based on the application class of GPU. According to the relationship with CPU, GPU can be divided into independent CPU and GPU. The independent GPU is usually welded on the circuit board of the graphics card, and is located under the fan of the graphics card. The independent GPU uses a dedicated display memory, and the video memory bandwidth determines the connection speed with the GPU. The integrated GPU is generally integrated with the CPU. The integrated GPU and CPU share a fan and cache. The integrated GPU has good compatibility because the design, manufacture and driver of the integrated GPU are completed by the CPU manufacturer. In addition, due to the integration of CPU and GPU, the space of integrated GPU is small; the performance of integrated GPU is relatively independent, and the power consumption and cost of integrated GPU are relatively independent due to the integration of CPU and CPU. Independent GPU has independent video memory, larger space and better heat dissipation, so the performance of independent graphics card is better; but it needs additional space to meet the complex and huge graphics processing needs, and provide efficient video coding applications. However, strong performance means higher energy consumption, independent GPUs require additional power supply, and the cost is higher.

According to the type of application terminal, it can be divided into pcgpu, server GPU and mobile GPU. Pcgpu is applied to PC. According to its product positioning, either integrated GPU or stand-alone GPU can be used. For example, if the PC is mainly light office and text editing, the general product will choose to carry integrated GPU; if the PC needs to produce high-definition pictures, edit videos, render games, etc., the selected product will carry an independent GPU. Server GPU is applied to servers, which can be used for professional visualization, computing acceleration, deep learning and other applications. According to the development of a series of technologies such as cloud computing and artificial intelligence, the server GPU will be mainly independent GPU. The mobile terminal is becoming thinner and thinner, and the internal net space of the terminal has declined rapidly due to the increase of multiple function modules. At the same time, as far as the video and image need to be processed by the mobile terminal, the integrated GPU has been able to meet the requirements. Therefore, mobile GPU generally adopts integrated GPU.