What is GPU?Explain the difference and structure from the CPU!-FabCross for engineer

Here, we will briefly explain the GPU.GPU is an abbreviation of Graphics Processing Unit, and is a processor (semiconductor chip) that performs the necessary operations when performing image processing of 3D (3D) graphics.The following describes the differences and structures from CPU (Central Processing Unit), a processor that controls and calculates computers used in personal computers, and also describes the field of use of GPUs.

1.What is GPU?

The GPU is an arithmetic device or processor specializing in real -time image processing represented by computer games.For example, in 3D games, precise images are projected on the monitor, but to realize this image, a huge amount of operations is required.The GPU is in charge of this.In other words, it is a processor suitable for performing a fixed and vast amount of operational treatment.In order to perform these calculation processing, thousands of core, which is the core part of the processor, is mounted on one processor, and it is like a factory that can perform large amounts of processing through human sea tactics.。

The current high -performance GPU is connected to a high -speed video memory (VRAM), and coordinates such as peak processing (processing to move the peaks of polygons (polygon) or queue) and pixels (pixels) processing.It is equipped with a large number of programmable arithmetic (programmable shader unit) specialized in graphics shading (shadow calculation).

In addition, in the field of high performance computing (HPC), GPGPU (General-Purpose Computing on Graphics Processing Units), which uses GPU hardware, which has a better parallel operation performance than the CPU, for more common operations, is performed.It has become such a dedicated product that does not have a video output terminal for such fields, and a high -end product equipped with a operator specialized for deep learning -based AI (artificial intelligence).

Recently, many GPGPUs have also appeared, using a structure that can repeat the relatively simple numerical calculation of GPUs in parallel in parallel to many data.By using GPGPU, a server that has the same performance as a supercomputer can be built at a lower price.

The differences between GPU and CPU are compared to the role of 1, 2, 2, type of computational processing, 3, core number, 4, calculation speed, 5, applicable method, and applicable fields.

GPU is a processor that performs the calculation process required for imaging such as 3D graphics, while the CPU is a processor that corresponds to the brain of a personal computer or server, and the entire computer calculation processing is used.It is done.

The feature of the GPU is that the role as a processor is in charge of the calculation processing necessary for image descriptions such as 3D graphics, and the type of computational treatment that I am good at is parallel computational processing, 3.The number of core is thousands.4. Regarding the calculation speed, the GPU can achieve the calculation speed of several to 100 times or more of the CPU.5. For the application method or the applicable field, it is a method that is mounted on a graphic board that is useful for displaying the image processing of the personal computer more precisely or displaying 3D images smoothly.

In recent years, GPUs have been used instead of conventional CPUs for deep learning needed for AI, which is rapidly increasing in demand.Furthermore, the field of image recognition is developing rapidly due to the improvement of deep learning technology in AI, and a graphic board equipped with a GPU is also used in this field.Similarly, from around 2010, when deep learning was widely applied to pattern recognition issues, the recognition accuracy was increased by the use of deep learning methods to build acoustic models in the field of voice recognition.Again, a graphic board equipped with a GPU is used.

On the other hand, the characteristics of the CPU are 1, for the role of a processor, in charge of the entire computer, the command, control and operational processing of the processor, specifically, HDD (hard disk drive), semiconductor memory, OS (OS (OS). Operating System), programs, keyboards, mice, etc. It is a command tower that collects information sent from the entire computer, 2, and the type of computational treatment that you are good at is continuous operational treatment, HDD. CPUs are good at reading program data, loading and execution to semiconductor memory. 3. The number of cores is several. In addition, 4. The calculation speed is about 1/100 of the GPU of the GPU. 5. The application method or the applicable field is mounted on the motherboard, and the products that are installed are desktop personal computers, laptons, smartphones, tablets, feature phones (galaxy), large computers, and supercomputers. Other CPUs exist in all products in the world, including cars, medical equipment, washing machines, TVs, refrigerators, industrial robots, Blu -ray recorders, car navigation systems, digital cameras, game consoles, and microwave ovens. We perform those orders, control and calculation.

2.GPU structure (installation method)

Explain the structure of the board equipped with a GPU.The GPU has a single -type GPU and an integrated GPU, and the single -type GPU is equipped with a GPU on the graphic board.The graphic board is a board dedicated to the image, and the image is described on the monitor by receiving an order from the CPU.The integrated GPU includes a CPU or other memory integrated on a motherboard, and a CPU built -in GPU with a GPU built into the CPU.When a graphic board is installed, the function of the integrated GPU is disabled.

Explain the graphic board equipped with GPU.The graphic board is a board dedicated to image description equipped with a GPU on a personal computer, and the image is displayed on the monitor by receiving an order from the motherboard equipped with the CPU.

Although the graphic board is not essential to a personal computer, it is possible to display the image more detailed and display 3D images smoothly by installing it.Some graphic boards are equipped with a dedicated circuit that describes 3D images faster.

The disadvantage of installing a graphic board is that increased power consumption increases operational costs for personal computers.Because the GPU has a high fever, a cooling device such as a fan is attached.

GPUs may also be installed on the motherboard.Some of the motherboard chipsets have a built -in GPU and are called "built -in GPUs (integrated GPUs)".

The advantage of using the built -in GPU is power -saving / space -saving, but in general, the image processing capacity is much higher in the graphic board, so when using 3D graphics, the graphic board is installed.Recommended.

3.Used in deep learning

　GPUs, designed for 3D graphics applications that require vast operation, have attracted attention and have begun to be used for applications other than image processing. One of them is deep learning needed for AI, which has rapidly increased in demand in recent years. Deep learning means deep learning, a deep learning, which is deep, that is, a neural network, which has a large number of layers. In a broader sense, a type of expression learning that enables learning feature expressions (also called internal expressions and potential expressions) from local characteristics to high -abstract characteristics from data. (Source: Japan Encyclopedia). Deep learning, which performs a huge amount of operation based on such a large amount of data, has begun to use GPU servers equipped with GPU instead of conventional CPUs.

Image recognition is one of the pattern recognition technology that grasps the characteristics from the image and identifies the object. Humans can guess from their own experience, from their own experience, looking at images such as photos. However, computers have no accumulation or experience of human -like memory, so they cannot judge what is reflected. Therefore, in the image recognition, the computer is given a large amount of images from the database, and the computer is automatically "learning" the characteristics of the object. Then, the computer can "understand" the characteristics of the image from the image data, and if an image with the same characteristics is given, it can be guessed that it is a "one". In recent years, the field of image recognition has developed rapidly due to improvement of deep learning technology in AI. In the field of image recognition, a graphic board equipped with a GPU is also used.

Voice recognition is a technology that outputs the content of the utterance by analyzing the audio signal obtained by the sound of the microphone.Acoustic model that describes the sound characteristics of the phonemical signal, the pronunciation dictionary that describes the words contained in the language trying to recognize, and how the word is connected.The language model that describes whether it becomes a sentence is comprehensive, and the most reliable word column is estimated and output to the voice signal given as input.From around 2010, when deep learning was widely applied to pattern recognition issues, the recognition accuracy was increased by the use of deep learning methods to build acoustic models.Again, a graphic board equipped with a GPU is used.

4.Compare the main GPU!What do you recommend?

There are two main types of GPUs that perform image processing.They are single -type GPUs and integrated GPUs.The single -type GPU has a drawback such as "high heat generated" and "high power consumption" compared to CPU integrated GPUs.

The integrated GPU includes a "type built into the central chip motherboard" and a type that is built into the CPU (built -in GPU).The mainstream in the Intel specification is that it is built into the CPU.The characteristics of the integrated GPU are that they are low in performance than single ones, but low -priced, good cost performance, low heat consumption, and low power consumption.

We look at what kind of field and business the integrated GPU and the integrated GPU are suitable for each.First of all, the single -type GPU can perform high images and video processing, such as image processing such as 3D graphics, video processing such as videos, and 3D graphics, etc., and creative business such as design.It is also demonstrated.Integrated GPUs are suitable for clerical work such as document creation, viewing of the Internet, email, and programming.In other words, the integrated GPU is suitable for office work that does not require high performance for screen treatment.In addition, because the power consumption is low and the battery is improved, it can be said that it is suitable when using mobile on the go.

summary

As described above, the difference from the CPU was revealed about the GPU.The main role of the GPU is the calculation process necessary for image descriptions such as 3D graphics, and the type of computational processing that is good at 2 is parallelarian processing, 3, the number of cores is the number.For thousands, 4, computational speed, GPU can achieve an calculation speed of several to 100 times or more of the CPU, and for the application method or applicable field, the image processing of the PC.It is mounted on a graphic board that helps to display more precisely and display 3D images, and is now rapidly adopted for deep learning, image recognition, and voice recognition required for AI.For this reason, the role of the GPU will be more and more important, and the scope of its application will increase.