Fundamentals of CPU, GPU, and Storage in the Xinchuang Industry

Xinchuang industry (Xinchuang Industry Chain Full Overview 2022), i.e., the information technology application innovation industry, aims to achieve autonomy and controllability in the IT field and ensure national information security.

From the perspective of the industry chain, Xinchuang industry mainly consists of four parts: basic hardware, basic software, application software, and information security. Among them, chips, complete machines, operating systems, databases, and middleware are the most important industry‑chain links.

The Central Processing Unit (CPU) is the core of computation and control in a computer, comprising an arithmetic logic unit, a control unit, and registers. The ALU performs calculations and tests, the control unit fetches, decodes, and directs data flow, and registers temporarily store data during instruction processing.

CPU operation can be divided into fetch, decode, and execute: the CPU fetches instructions from memory, decodes them into control signals, and finally the ALU performs the computation.

An instruction set is the collection of all commands that a CPU can execute. CPUs are classified into CISC (Complex Instruction Set Computing) and RISC (Reduced Instruction Set Computing). CISC CPUs are mainly x86 architecture, while RISC CPUs include ARM, RISC‑V, MIPS, POWER, and Alpha.

In terms of ecosystem, CPUs based on x86 and ARM have good compatibility with downstream software and hardware, offering abundant products and user friendliness. CPUs based on MIPS and Alpha perform well in high‑performance computing and embedded industrial control but have less market penetration.

Graphics Processing Unit (GPU), also called visual processor or display chip, is a microprocessor specialized for image and graphics related computations in PCs, workstations, game consoles, and some mobile devices.

GPU generates 3D graphics, maps them to pixels, computes final colors, and outputs the result. The pipeline typically includes vertex processing, rasterization, texture mapping, pixel processing, and output. GPUs use a stream‑parallel model that allows independent parallel computation of data rows.

CPU is designed for low‑latency tasks with a few ALUs and control logic, excelling at serial processing and logical control. GPU is designed for high throughput with many ALUs, excelling at parallel processing of dense data, and is also used for AI training.

Independent GPUs are packaged on separate graphics cards with dedicated video memory; performance depends on GPU and memory bandwidth, but they consume more power and generate more heat. Integrated GPUs share a die with the CPU and use system memory, offering lower power consumption and heat but generally lower performance.

Computer storage includes various memory types and enterprise storage. Based on medium, storage is divided into optical, semiconductor, and magnetic storage. Semiconductor storage includes RAM and ROM. RAM is volatile main memory directly exchanged with the CPU, while ROM retains data without power.

DRAM and Flash (including NOR and NAND) dominate the current storage market; in 2020, DRAM accounted for over 53% of the global storage market, making it the largest segment.

The basic hardware layer includes complete machines, as illustrated by the following diagrams.

Xinchuang industry aims to achieve autonomy and controllability in the information technology field and safeguard national information security.

Statement: Thanks to the original author for their hard work. All reproduced articles are credited; contact us for copyright issues.

Recommended reading: More architecture‑related technical summaries are available in the “Architect’s Full Store of Technical Materials Pack” e‑book (37 volumes).

Order the full architecture e‑book pack for 198 CNY (original price 350 CNY) to receive free updates.

Warm reminder: Scan the QR code to follow the public account and click “Read Original” to get the “Architect’s Full Store of Technical Materials Pack” details.