What is CPU?

The Central Processing Unit (CPU), the main chip of your computer, phone, television, etc., is in charge of sending instructions to the motherboard's various electronic components. The CPU, also known as the "brain" of the computer, is the most flexible of the chips we're talking about. However, due to its versatility, it will consume more power and do some jobs more slowly than CPUs with greater specialization.

CPU Architecture

The CPU, which has been around for more than 50 years, is the one computing architecture that is most well-known and widely used. Because it is built to effectively handle serial instructions, the CPU architecture is occasionally referred to as a scalar architecture. In order to promote instruction-level parallelism (ILP) and speed up the execution of serial applications, CPUs have been optimized using a variety of ways.

Scalar pipelined CPU cores can process instructions split into many levels at a pace of up to one instruction per clock cycle (IPC) when there are no dependencies. Modern CPU cores are multithreaded superscalar processors that may execute many disordered instructions per clock cycle and have sophisticated techniques for locating instruction-level parallelism. This helps them perform better. They retrieve several instructions at once, determine the dependency graph between them, use advanced branch prediction techniques, and then execute those instructions in parallel at typically 10 times the IPC speed of scalar processors.

What is GPU

The General-Purpose Graphics Processing Unit (GPU) is a type of graphics processing unit that makes use of a graphics processing unit to do general-purpose computing tasks that are traditionally performed by the central processor unit. Like CPUs, GPUs are general-purpose processors. However, because GPUs have more cores, they can run more processes simultaneously and more efficiently handle floating-point calculations.

GPU Architecture

GPUs are processors made up of smaller, more specialized cores that operate in large parallel and are often more powerful than high-performance CPU cores. GPU architectures place less emphasis on the latency and performance of individual threads in favor of the combined throughput of all cores. Vector architectures—often referred to as GPU architectures—effectively handle vector data (arrays of integers). Less silicon is used for caching and control, and more is used for computing. As a result, in order to maximize performance and efficiency, GPU hardware investigates less parallelism at the instruction level. GPUs do not allow complicated branch prediction since they are ordered processors. As an alternative, they have deep pipelines and an excessive amount of arithmetic logic units (ALUs). Performance is achieved through multi-threaded execution of large independent data, which spreads the cost of simpler control and smaller caches.

What is FPGA?

The Field Programmable Gate Array (FPGA), albeit also a silicon-based semiconductor, is composed of a grid of programmable logic blocks (CLBs) that are connected via programmable interconnects. It is possible to design an FPGA to do certain tasks or applications. It supports programming in C/C++ and hardware languages and has a low latency and good energy efficiency ratio. But each time a function needs to be changed, it needs to be reprogrammed.

FPGA Architecture

Contrary to fixed-architecture CPUs and GPUs that can be software-programmed, FPGAs are reconfigurable, and the user can choose the computational engines they employ. When software is created for FPGAs, the compiled instructions turn into hardware elements that may all be performed concurrently and are spatially ordered on the FPGA architecture. Because of this, FPGA architectures are also known as spatial architectures.
FPGAs are collections of numerous small processing units that can contain millions of 1-bit adaptive logic units that can be programmed, tens of thousands of memory units that can be configured, and thousands of digital signal processing (DSP) units that can perform fixed-point and variable-precision floating-point operations. All of these resources are connected through a grid of programmable wires that can be activated as needed.

What is ASIC?

An "Application Specific Integrated Circuit" (ASIC) is a type of silicon chip designed for a specific logic operation. It is a special chip that carries out a specific hardware function. The ASIC chip's computational efficiency and power can be adjusted based on algorithm requirements, but they cannot be changed.
As a result, ASIC chips are small, energy-efficient, and have a high level of specialized computing capacity. Once changed, the application scope cannot be changed back, and doing so is expensive.

CPU vs. GPU vs. FPGA vs. ASIC

An "Application Specific Integrated Circuit" (ASIC) is a type of silicon chip designed for a specific logic operation. It is a special chip that carries out a specific hardware function. The ASIC chip's computational efficiency and power can be adjusted based on algorithm requirements, but they cannot be changed. As a result, ASIC chips are small, energy-efficient, and have a high level of specialized computing capacity. Once changed, the application scope cannot be changed back, and doing so is expensive.

Numerous logic circuits make up an FPGA, a general-purpose integrated circuit that may be "wired up" however you wish. Due to its lack of specialization, an FPGA set to be a CPU or GPU will always be slower than the real CPU or GPU. Contrarily, the FPGA can be programmed with logic to carry out computing tasks for which neither CPUs nor GPUs are particularly well-suited. Your FPGA can be programmed to do a particular set of complex computations on an input of data more quickly than a CPU. The FPGA still has the advantage of being able to be reprogrammed if you need to undertake another task suitable for it, even though an ASIC could be created to perform these same exact workloads faster than the FPGA.