Processors: What are CPU, GPU, TPU, DPU, QPU, NPU, ASIC and FPGA?

Digital devices such as computers, smartphones, tablets, televisions, cameras, video game consoles, toys and the like contain a wide variety of components. Under the big umbrella term “chip” you can open up other different subcategories. One of them is the processors. A processor is there to make calculations or process data in a certain way and output the results. Processors in various forms are used as integrated circuits, individually or as part of a single-chip system (SoC). Below you will find out what special tasks the CPU, GPU, TPU, DPU, QPU, NPU, ASIC and FPGA have.

Main processor: what is a CPU?

The CPU (Ccentral Pprocessing Unit) is the central processor or central processing unit of a computer system. It is the main component in many computers, servers and other everyday devices that is responsible for executing commands and processing data. The CPU interprets data and executes commands given by software applications and the operating system.

The most important functions of a CPU include:

• Command execution: The CPU executes instructions written in machine language, with each instruction representing a specific action.
• Data processing: The CPU can perform arithmetic and logical operations on values ​​and data. This includes addition, subtraction, multiplication, division and comparison operations.
• steering: The CPU controls the flow of data within the computer system. It sets the pace for the execution of commands and synchronizes the sequence of operations.

Modern CPUs consist of multiple cores, which allows them to process multiple tasks in parallel (multicore architecture). The performance of a computer CPU was once often measured in megahertz (MHz) and is now more commonly measured in gigahertz (GHz), which represents the number of cycles per second. Other performance factors to consider when choosing a CPU include architecture (x86, RISC-V or ARM), cache size and number of threads.

Special Processor: What is an ASIC?

ASIC stands for “ApplicationSspecific Iintegrated Circuit” (application-specific integrated circuit). This processor is specifically designed and used for a specific application or task. Unlike general processors such as CPUs (Central Processing Units) or GPUs (Graphics Processing Units), which are suitable for a wide range of applications, ASICs are designed to perform a specific function extremely efficiently. They are only made for their calculations or algorithms. ASICs are installed in USB chargers and network switches, for example. 

Some features of ASICs are:

• Specificity: ASICs are tailored for a specific task or application. This allows for high efficiency and performance related to this specific function, but makes it less flexible compared to general processors.
• Efficiency: Because ASICs are tailored to a specific task, they can often perform that task much more efficiently and quickly than general purpose processors. This is particularly important in areas where speed and energy efficiency are critical.
• Applications: ASICs are used in various areas including telecommunications, network equipment, cryptocurrency mining (e.g. ASIC miners for Bitcoin), embedded systems, and many other specialized applications.
• Development costs: Developing ASICs can be expensive and time-consuming because they need to be customized for a particular application. This makes ASICs particularly useful for applications in which high volumes are produced and the costs can be amortized through mass sales.

ASICs are not as flexible as CPUs or GPUs, but when it comes to specialized and compute-intensive tasks, they can offer superior performance. If you want a special processor structure but want to use it individually for different applications, then you should choose an FPGA.

Special but flexible: What is an FPGA?

FPGA stands for “FfieldProgrammable Guntil Array” (field programmable gate arrangement). This processor, unlike ASICs, is programmable. This means that an FPGA can be configured by the user after production. For example, these types of processors are found in modern retro gaming solutions that are intended to imitate various consoles. The FPGA can transport the structure of old hardware, but is not limited to simulating a specific device, but can, for example, serve as an SNES, Sega Genesis, Sega Saturn or similar, depending on the game used. Roughly speaking, FPGAs are between CPUs and ASICs in form and function. 

Some features of FPGAs are:

• programmability: FPGAs are flexible and can be programmed as needed. This makes it possible to customize the functionality of the circuit after manufacturing. This is particularly useful when requirements change or when prototypes are developed.
• Gates and connections: FPGAs consist of a matrix of programmable logic blocks (gates) and routing resources. These blocks can be connected together in different ways to create complex digital circuits.
• Applications: FPGAs are used in a variety of applications including signal processing, wireless communications, image processing, automotive, networking and many others. They are particularly useful in areas where customizability, performance, and parallel processing are important.
• development time: Compared to ASICs, FPGAs have shorter development times because they are programmable after manufacturing and do not require special customization for a specific application.

Although FPGAs may not be as efficient as ASICs for a specific application, they offer the advantage of programmability and flexibility, making them a suitable choice for various development and prototyping scenarios. Developers can create and customize their own digital logic to implement specific functions without having to redesign physical circuits.

The graphics unit: What is a GPU?

A GPU (Grapics Pprocessing Unit) is a specialized type of processor primarily responsible for processing graphics and image data in a computer. Unlike the CPU, which serves general purposes, the GPU is designed to process large amounts of data in parallel. Originally, GPUs in the form of "graphics cards" were designed primarily for displaying 3D graphics in video games and computer graphics applications, but over time they have evolved into powerful accelerators for a wide range of computationally intensive tasks.

The key features and functions of a GPU include:

• Parallel processing: Unlike the sequential processing of data by a CPU, a GPU can perform many operations at the same time (in parallel). This is particularly advantageous for tasks that require large amounts of data to be processed simultaneously, such as image and video editing or scientific calculations.
• shader: GPUs use special programs called shaders to create complex graphical effects. These shaders can be customized for various purposes, including simulating light and shadows, rendering textures, and other visual effects and animations.
• CUDA and OpenCL: Many modern GPUs support frameworks such as CUDA (Compute Unified Device Architecture) from NVIDIA or OpenCL (Open Computing Language), which allow developers to use the GPU's computing power for general purposes. This enables acceleration of applications beyond pure graphics processing, such as scientific simulations, machine learning and cryptocurrency mining.

Overall, GPUs offer a powerful solution for compute-intensive applications that can benefit from processing large amounts of data in parallel. In the following video you can see a simplified model representation of image processing by a CPU (slow and in individual steps) and a GPU (fast and with overall image creation). The simplified representation should make the comparison understandable:

Application of AI: What is an NPU?

The abbreviation NPU stands for “Neural Pprocessing Unit” (Neural Processing Unit). An NPU is a specialized hardware component designed to accelerate tasks related to artificial intelligence (AI) and machine learning, particularly those involving neural networks.

Here are some important features of NPUs:

• Neural Networks: NPUs are designed to meet the specialized needs of neural networks used in many AI applications such as image recognition, speech recognition, machine translation and other complex pattern recognition tasks.
• Parallel processing: Neural networks require intensive calculations, especially matrix operations and linear algebra. NPUs are designed to efficiently process these calculations in parallel, resulting in accelerated execution of AI models.
• Acceleration of inferences: NPUs are often specialized to speed up inference operations (applying a trained model to new data). An example of this would be the “Neural Engine” in Apple SoC models, such as the A chip in the iPhone or the M chip in the Mac.
• Energy Efficiency: As artificial intelligence is used in various applications such as smartphones, IoT devices and other embedded systems, the energy efficiency of NPUs is often an important feature to minimize power consumption. But they can also be used efficiently in desktop computers with a mains connection to reduce the power consumption of AI applications.

NPUs are an example of specialized hardware designed to meet the growing need for computing power for AI applications. They can be used in various devices, from cloud servers to edge devices to consumer products, to optimize the execution of AI algorithms. One example is Apple’s “Neural Engine”.

Machine Learning: What is a TPU?

TPU stands for “Tsensor Pprocessing Unit.” It is a specially developed hardware from Google for machine learning and tensor-based calculations. TPU is designed to perform TensorFlow tasks, which is a widely used open source machine learning framework, particularly efficiently.

Some features of TPUs are:

• Tensor based: TPUs are specifically optimized for processing tensors. Tensors are multidimensional data structures used in many machine learning applications, particularly neural networks.
• Efficiency: TPUs are designed to perform machine learning calculations more efficiently than traditional CPUs or GPUs. This is because they can perform certain operations that are common in machine learning algorithms particularly quickly and efficiently.
• Google Cloud TPUs: Google offers Cloud TPUs that enable developers to leverage the computing power of TPUs via the Google Cloud Platform. This makes it possible to run sophisticated machine learning models in the cloud, saving costs for locally applied specialized hardware.
• Applications: TPUs are used in various machine learning applications, including image recognition, language processing, translation, and other tasks that require intensive neural network computations.

It is important to note that TPUs are specifically optimized for tensor-based calculations and are particularly effective in machine learning applications. Therefore, compared to general CPUs and GPUs, TPUs can provide higher performance and power efficiency in these specific use cases.

Data Processing: What is a DPU?

As DPU (Data Pprocessing Unit) generally refers to a component or unit in a computer system that is responsible for processing (certain) data. These can be different types of processors or specialized hardware designed to process data in a particular way. Here are some possible interpretations:

• General data processing unit: The term DPU can simply be a general name for a computing unit that is responsible for processing data in a computer or other electronic system. This could include a CPU (central processing unit) or other type of processor that executes instructions and processes data according to instructions from software applications.
• Specialized data processing unit: In some cases, DPU may indicate a specialized hardware unit optimized for specific types of data processing tasks. For example, this could be a unit designed for graphics data processing, signal processing or machine learning. DPU can therefore be used as an umbrella term for GPUs, NPUs, TPUs and the like.
• Network data processing unit: In networks, DPU can indicate a unit that is specifically optimized for processing data in the context of network communications. This could include functions such as routing data packets, performing security checks, or processing network analysis. In networks, a “Deep Packet Inspection Processing Unit” can also be used for IT security, which inspects data packets and is also abbreviated as “DPU”.

The exact context in which the term is used determines what type of data processing and what type of processing unit is meant. However, typically DPU refers to a hardware component that specializes in processing data in a fixed form.

Quantum Computer: What is a QPU?

The abbreviation QPU stands for “Quantum Pprocessing Unit” (quantum processing unit). Unlike classical computers, which are based on bits that can assume the state 0 or 1, quantum computers use qubits that can be in a superposition of states. This allows quantum computers to solve specific problems, such as certain cryptographic applications or complex optimization problems, potentially faster than traditional computers.

A few important points about QPU:

• Quantum superposition: Qubits can represent several states at the same time due to the state of quantum superposition. This opens up the possibility of carrying out certain calculations in parallel.
• Entanglement: Qubits can be entangled with each other so that the state of one qubit is linked to the state of another qubit. This allows for strong correlation between the qubits, which can be useful in certain applications.
• Quantum encryption: Quantum computers might be able to break quantum encryption systems, but they could also be used to enable secure quantum communications.
• Expansion instead of replacement for existing hardware: Quantum computers are intended to open up new areas of computing. In addition to encrypting and decrypting data, complex simulations are also possible uses. However, quantum computers are not intended to replace current computer technology as a “super version” of it. So anyone hoping for a literal “quantum leap” for their gaming setup will unfortunately be disappointed.

It is important to note that quantum computers are still in the development phase and there are many technical challenges that need to be overcome before they can be used for general applications. The term QPU refers to the special processing unit in a quantum computer that is responsible for executing quantum mechanical calculations. Companies and research institutions around the world are working intensively on the development of quantum computers and QPUs to improve the performance and reliability of this technology.

Multiple units in one: what is an SoC?

An SoC stands for “Ssystemon/a-Chip” (system on a chip). It is an integrated circuit that combines several components and functions of a complete computer system on a single chip - i.e. several integrated circuits on a single silicon plate. These components can include processors, memory, interfaces, communication modules, and others that are typically found individually plugged onto a motherboard (motherboard or logic board) of a traditional computer.

Some important features of an SoC are:

• Integration: SoC integrates various components on a single chip, which saves space, accelerates data exchange and can improve energy efficiency.
• Energy Efficiency: By integrating components on one chip, energy losses due to communication between different chips can be minimized.
• Compactness: SoCs are often available in compact and small form factors, making them ideal for use in mobile devices, Internet of Things (IoT) devices, laptops, mini-computers, and other space-constrained environments.
• Applications: SoCs are used in a variety of devices, including smartphones, tablets, embedded systems, IoT devices, smart TVs, etc. Due to their energy efficiency (and Apple's efforts since 2020), they are also increasingly being used in computers.
• versatility: Because SoCs can integrate various functions, they are extremely versatile and can be customized for a wide range of applications. You can ensure the implementation of modern operating systems and apps developed for them.

Examples of SoCs can be found at Apple, from Mac, iPad and iPhone to Apple Watch and Apple TV to the Studio Display and the Vision Pro. In addition to the graphics unit (CPU), an M chip in the Apple Mac includes, among other things: You can also find a graphics unit (GPU), the neural engine (NPU), the media engine (unit specializing in video encoding and decoding) and the like. A wide variety of processors are used here in order to be able to complete a wide variety of tasks as quickly and efficiently as possible.

Johannes Domke

After graduating from high school, Johannes completed an apprenticeship as a business assistant specializing in foreign languages. But then he decided to research and write, which resulted in his independence. For several years he has been working for Sir Apfelot, among others. His articles include product introductions, news, manuals, video games, consoles, and more. He follows Apple keynotes live via stream.