Guided pathThis is part of Understand Transformers & BERTBack to the path

What is a Computer?

P3.computing-fundamentals.01 · Audience: guest, it-ml, language-pro

Every AI model runs on ordinary hardware — a processor, some working memory, and a disk. Before we can explain why large language models need special machines, we need a plain-language picture of the three parts of a computer and what each one does. No electronics required: a professional kitchen is all the intuition you need.

Step 1 / 4Three parts of a computer

A computer is like a professional kitchen. The CPU is the chef — the one who actually does the work, one task at a time, very fast. RAM is the chef's countertop — where the active ingredients sit, close to hand, but with limited space. Storage (an SSD or hard drive) is the pantry — enormous capacity, but you must walk there to fetch things. Everything a computer does is moving information between these three places.

ComponentWhat it doesKitchen analogy
CPU (processor)Executes instructions — adds numbers, compares values, decides what to do next (3–5 billion operations per second)The chef: skilled and fast, but only two hands
RAM (working memory)Holds the data and programs in use right now; the CPU reads and writes it constantly (fast, but limited)The countertop: instantly reachable, only a few square metres
Storage (SSD / hard drive)Keeps files permanently — programs, documents, model weights — even when the power is off (huge, but slower)The pantry: enormous capacity, a few steps away
🗣️ From a linguist's perspective: CPU, RAM, and storage as three time-scales of memory
Cognitive science distinguishes three memory systems: working memory (what you actively hold in mind right now — very fast, very limited), short-term memory (what you have recently encountered), and long-term memory (everything you know — vast, but slower to retrieve). A computer's three parts map onto this exactly: RAM is working memory, the CPU's small internal cache is a notepad between RAM and the chef, and storage is long-term memory. Every AI computation orchestrates all three — a language-model file sits in storage, is loaded into RAM when the program starts, and is processed number by number by the CPU (or GPU) at run time.
Quick recap — the three parts
  • A computer has three tiers: the CPU (executes instructions), RAM (working memory — fast but limited), and storage (permanent — large but slower).
  • The CPU handles one or a few operations at a time; it excels at complex logic but is not built for the massive parallelism AI requires.
  • RAM must hold the entire model while it runs — large models (GPT-3 ≈ 700 GB) exceed the RAM of any consumer machine.
  • Storage is permanent but slow; loading a model means copying it from storage into RAM every time the program starts.
📚 Go Further

Plain-language reads on how computers work.

TypeResource
VideoCrash Course Computer Science — how CPU, RAM, and storage work, from first principles (YouTube, free)
VideoComputerphile — How Computer Memory Works (YouTube)
ReferenceJulia Evans — How does a computer know what 2 + 2 is? (jvns.ca)
Ask the mentor about this module

Ask a question about this content. The mentor explains and grounds its answer in what you are studying; asking is recorded as a learning signal, not a grade.

Ctrl/Cmd + Enter to send