What is a CPU? Explanation of its basic structure, task processing method, and applications, compared to a GPU
CPUs are semiconductor products that are installed in various electronic devices such as computers and
smartphones, and serve as the “brains” of these devices. GPUs are also semiconductor products that are installed
in computers and other devices, but are there any differences between them and CPUs?
This article explains the basic structure of CPUs and GPUs in a way that is easy to understand, how they process
tasks, their applications, and metrics for evaluating their performance.
1. Basic structure of the CPU and the GPU
Both the CPU (central processing unit) and the GPU (graphics processing unit) are called logic semiconductors, but they have different roles.
The basic structure of a CPU is a chip that integrates mainly several hundred million to several billion
transistors (semiconductor devices) on it. In other words, the CPU is an integrated circuit (IC), and it is
classified as a very large-scale integrated circuit (VLSI) or an ultra large-scale integrated circuit (ULSI)
within the IC category. In the 1970s, the number of transistors integrated into the world's first CPU was 2,300.
Today, CPUs are made up of approximately billions of transistors, which is about a million times more than back
then.
The size of the CPU is a few centimeters square, and the core is the chip part comprising countless transistors,
called a die. The core is connected to external components and devices via more than 1,000 connection terminals
(pins), and is placed in the center of the package to allow for heat dissipation and the like. In addition, most
cores are covered with heat-dissipating plates made of metal or other materials.
Like the CPU, the GPU also integrates a large number of transistors on a chip, but the main difference is that it has a very large number of cores. Although the number of operations that can be performed is limited compared to CPUs, which can handle a variety of complex processes, performance and efficiency can be increased by raising the number of cores and processing in parallel.
2. Differences in how the CPU and GPU process tasks
The CPU is the central processing unit and the GPU is the graphics processing unit, but what is the main difference between the two?
CPU and GPU task processing image
CPUs are good at processing different types of complex tasks in order and continuously. GPUs, in contrast, are small devices with many specialized cores that work together and are good at processing large amounts of data at once by dividing processing tasks across many cores and processing them in parallel.
Characteristics and applications of CPUs and GPUs
The CPU is responsible for complex operations and program control, serving as a sort of control tower. CPUs are particularly versatile when it comes to handling a wide range of processing and have high processing performance per core, but they are not good at handling multiple processes at the same time. Because of these characteristics, the CPU is responsible for a number of processes that support computer operations, such as sequential data processing and control, running the operating system (OS) and mouse, and executing applications and various commands.
On the other hand, a GPU is a semiconductor developed as an IC specialized in image processing, as indicated by
its name, “Graphics Processing Unit.” It specializes in high-speed graphics and data processing, and its strength
lies in parallel processing, which means handling multiple tasks simultaneously.
As a result, it can process large amounts of data simultaneously, making it ideal for high-speed rendering and
data processing in areas such as gaming, visual effects, and scientific computing. In recent years, the GPU has
also begun to be applied to machine learning and AI computing.
3. Applications of CPUs and GPUs according to their characteristics
CPUs and GPUs can be used for a variety of applications depending on their characteristics. In addition, by
combining the characteristics of the CPU and GPU in the right ways, the range of applications can be further
expanded.
This section explains the main applications of the CPU and GPU, as well as applications that arise from combining
the two.
Applications of CPUs
The main applications of CPUs, which are good at continuous processing, are to execute various programs and handle
processes related to starting up and running software.
For example, they are responsible for the smooth running of everyday applications such as document creation
software, web browsing, and basic video editing.
CPUs are also used for data management and computation on servers.
Applications of GPUs
GPUs are good at advanced video editing and graphic processing, and are the reason why we can now enjoy
high-quality 3D games on our home computers.
Processing VR applications is another area where GPUs excel.
Of particular interest is the use of GPUs in areas such as machine learning and deep learning.
Deep learning, which is essential for generative AI, is based on computations that mimic the processing of the
human brain, known as neural networks. A neural network consists of multiple layers that simultaneously perform
various pattern weightings and operations on the input data before passing it on to the next layer. GPU parallel
processing suits computations like this. The existence of GPUs is probably a major factor in the emergence of
generative AI, which is changing the world, as exemplified by ChatGPT.
GPUs are also used in supercomputers for research and testing in areas such as predicting climate change and finding new medical treatments.
Applications that arise from combining CPUs and GPUs
CPUs and GPUs each have their own advantages and disadvantages. The advantage of the CPU is that it has a greater
degree of versatility in the processing it can perform compared to the GPU, while the benefit of the GPU is that
it is better than the CPU at large-scale parallel processing.
For this reason, many computers have both a CPU and a GPU, and assigning each unit to a task that matches its
characteristics improves a computer's overall performance and accuracy.
For example, in video editing, the CPU is responsible for overall process control and basic editing tasks, while
the GPU is used for high-speed advanced graphics processing such as rendering and color adjustment, enabling
high-resolution video editing.
This method of utilizing CPU and GPU roles is the same for tasks such as building AI models.
4. Metrics for evaluating CPU and GPU performance
CPUs and GPUs share many common components that can also be used in relation to performance metrics. This section shows how to read the main metrics.
Number of cores
The core is the component that forms the heart of the CPU and GPU and is the unit that actually performs the arithmetic processing. In recent years, integrating multiple cores has become common, and such CPUs are called “multi-core CPUs.” Each core performs a single process, but if there are many cores, it is possible to have multiple processes running simultaneously.
Compared to CPUs, GPUs have far more cores. Specifically, whereas CPUs have tens of cores, GPUs have thousands.
Number of threads
The number of tasks that one core can process simultaneously is called a thread. Like the core, the more of these
the CPU or GPU has, the more processes it can run simultaneously in parallel.
In the past, each core having one thread was common, but now cores with two threads have also appeared.
The
number of threads that can be processed simultaneously is sometimes given along with the number of cores and is
used to describe CPU performance.
Clock frequency
The clock frequency is a factor that indicates the processing speed of the CPU and GPU. The higher the frequency,
the higher the performance.
Clock frequencies are expressed in units of “Hz (Hertz)” or “GHz (Gigahertz),” and most products today have high
values such as 2.60 GHz and 3.40 GHz.
Cache memory capacity
Cache memory is a storage device (memory) that temporarily stores data that is accessed more frequently, and it is
responsible for improving access speed. The capacity of the cache memory is expressed in units such as KB and MB,
and the higher the number, the faster the data can be accessed.
Cache memory used to be located outside the CPU, but thanks to technological improvements, it is now almost always
located on the CPU chip.
Benchmark score
A benchmark score is a list of evaluation results obtained by testing and quantifying the performance of each element of the CPU and GPU. Testing is done using benchmark software, and the higher the score, the higher the product's performance.
The following are some benchmark scores common to both CPUs and GPUs.
Floating-point operation performance (FLOPS)
This is a score that indicates how many floating-point operations the CPU or GPU can perform per unit of time. It is used to evaluate the performance of scientific computing and machine learning.
Tera operations per second (TOPS)
This score indicates the number of integer operations that can be processed per second, in units of one trillion. It is used to assess general computational ability.
Memory bandwidth
This is the data transfer rate between memory and the processor, and is expressed in units of bytes per second. High memory bandwidth is important for applications that process large amounts of data.
Energy efficiency (Performance per Watt)
This measures how much performance the CPU and GPU can deliver per watt.
5. Keep an eye on CPUs and GPUs as they continue to evolve
It is said that semiconductors will continue to evolve in the future in ways such as through increased
integration, and there are expectations for further improvements in the functions and performance of CPUs and
GPUs, which are types of semiconductors.
Regarding GPUs in particular, there is a lot of development underway on GPUs for deep learning. As their use in AI
is attracting increasing attention, greater effort will be put into their development.
As technological innovations in CPUs and GPUs emerge, be sure to keep a close eye on the new applications they make possible.