PC Components lessons

Meet the parts

Up close, the CPU is a silicon chip, the die, under a metal lid. The lid spreads the die's heat into the cooler.

A RAM stick carries a row of memory chips under its heat spreaders. A notch in its edge stops it going in the wrong way.

The GPU chip sits on the graphics card, ringed by its own fast memory, VRAM. A big cooler with fans covers them both.

3D example: a CPU, a DDR5 stick and a graphics card, each seen whole, then taken apart: the lid lifted off the dies, the spreaders off the chips, the cooler off the GPU and its VRAM.

Practice: Match each part to its job

Match each part to its job (MatchPairs), four random pairs a round.

Cores, threads and clock speed

A core is one complete processor inside the CPU. Desktop CPUs in 2026 have 6 to 24 cores, so they can work on that many jobs at once.

A thread is one stream of instructions. Many CPUs run two threads on each core, so one can use the core while the other waits.

The clock speed says how many cycles each core runs per second. At 5 GHz a core ticks 5 billion times a second, so every step of its work happens sooner.

Example: a video export split into 96 chunks shared across the cores, and a game's one-thread logic, timed from the core count and clock speed (sliders).

Practice: Pick the CPU

Pick the better of two CPUs for a job that splits up or one that runs on one thread (ChoiceQuiz), random specs.

Fetch, decode, execute

A program is a list of simple instructions stored in memory. The CPU runs them one after another in a loop called the fetch, decode, execute cycle.

It fetches the instruction at the address in its program counter. It decodes it to work out what to do, then executes it, with the ALU doing any maths.

Results go into registers, tiny stores inside the core. The program counter moves on to the next address, and the loop starts again.

Example: a four-instruction program runs step by step: the program counter, the fetched instruction, what it decodes to, and the registers and memory it changes.

Practice: Run the program

Trace a random short program and type the value left in register A (MathAnswerInput).

Cache: memory on the chip

One read from RAM takes about 90 nanoseconds, which is 450 ticks of a 5 GHz clock. So the CPU keeps copies of recent data in its cache, fast memory on the chip itself.

Cache comes in levels. L1 is tiny and answers in about 1 nanosecond, L2 is bigger, and L3 holds tens of megabytes shared between the cores.

Finding the data in cache is a hit, and going out to RAM is a miss. Programs reuse the same data a lot, so most reads are hits.

Example: twenty reads at a chosen hit rate; hits take 1 ns and misses 90 ns, and the average read time is computed (slider).

Practice: Fastest to slowest

Put the kinds of memory in order, first by speed and then by size.

Put the memory levels in order, fastest first, then biggest first (OrderSteps).

RAM: fast but forgetful

RAM holds the programs and data in use right now. It is volatile: everything in it is lost when the power goes off.

Desktops in 2026 use DDR5 sticks, and 32 GB is a common amount for gaming. A speed such as DDR5-6000 means 6,000 million transfers a second, each moving 8 bytes.

Two sticks in the right slots run in dual channel, so the CPU reads both at once. That doubles the bandwidth, the amount of data moved each second.

Memory bandwidth: \text{bandwidth} = \text{MT/s} \times 8\ \text{bytes} \times \text{channels}

Example: one stick or two, at a chosen DDR5 speed; transfers flow down each channel and the bandwidth is computed (slider and toggle).

Practice: Work out the bandwidth

Type the bandwidth in GB/s for a random DDR5 speed in single or dual channel (MathAnswerInput).

Storage: hard drives and SSDs

A hard drive (HDD) stores data on spinning magnetic disks, read by an arm that swings across them. Each read waits for the arm to move and the disk to turn, about 10 milliseconds.

An SSD stores data in flash memory chips with no moving parts, so any read takes about 0.1 milliseconds. A SATA SSD uses the old drive cable and tops out near 550 MB/s.

An NVMe SSD plugs into an M.2 slot and talks over PCIe lanes. A PCIe 4.0 drive reads about 7,000 MB/s, and the newest PCIe 5.0 drives reach about 14,000 MB/s.

Example: a hard drive's arm seeks to scattered blocks while its disk turns, against an SSD reading any block straight away; reads per second computed from the timings.

From drive to screen

Launching a game reads its files from the drive into RAM. The CPU sets up the level, then the GPU draws its frames.

The drive's read speed sets the load time. The same 8 GB takes 40 seconds from a hard drive but just over a second from an NVMe SSD.

Each step away from the CPU is slower. If a cache read took one second, a hard drive read would take almost four months.

Load time: \text{load time} = \frac{\text{size}}{\text{read speed}}

Example: a game loads from the drive into RAM, the CPU sets it up and the GPU draws it, on a hard drive, a SATA SSD and an NVMe SSD; load times and the memory hierarchy computed from data, then the learner swaps the drive and the size.

Practice: How long to load?

Work out how long a drive takes to load a game's data.

Type the load time in seconds for a random amount of data from a random drive (MathAnswerInput).

The GPU: thousands of small cores

The GPU is built to do the same small job many times at once. A mid-range card in 2026 has several thousand simple cores, and a top card has over 20,000.

A 1080p frame has about 2 million pixels, and each one needs its colour worked out. A CPU shares them between a few fast cores, while a GPU spreads them across thousands.

The GPU has its own memory, VRAM, for textures and finished frames. Mid-range cards carry 12 to 16 GB.

Example: the same frame shaded by a 16-core CPU and a 6,144-core GPU; frame times computed from cores, clock and work per pixel.

Practice: CPU or GPU?

Sort tasks into CPU work and GPU work (SortGroups), two rounds.

The motherboard

The motherboard joins every part with layers of copper wiring. The CPU sits in its socket, with four RAM slots beside it.

The long PCIe x16 slot holds the graphics card, and the M.2 slot holds an NVMe SSD flat on the board. The chipset adds extra USB ports, SATA ports and PCIe lanes.

Power arrives through the 24-pin main connector and an 8-pin CPU connector. The rear I/O ports show through the back of the case.

3D tour of an ATX motherboard: socket, RAM slots, PCIe x16, M.2 with its SSD, chipset, power connectors and rear I/O.

Practice: Find the slot

Tap the place on a top-down motherboard where each part or cable goes.

Sizes and compatibility

Motherboards come in standard sizes called form factors. ATX is 305 by 244 mm, micro-ATX is 244 mm square and mini-ITX is 170 mm square.

Smaller boards have fewer slots but fit smaller cases. A case lists the form factors it takes, and the longest graphics card that fits.

Parts must also match each other. The CPU needs a board with the same socket, the board takes one RAM type, and the power supply needs the right plugs.

Example: ATX, micro-ATX and mini-ITX drawn to scale, lined up on their rear I/O corner, with their slots counted.

Practice: Spot the clash

Tap the one part in a random parts list that does not fit the rest: socket, RAM type, board size, card length or power plug.

The power supply

The power supply (PSU) turns mains electricity into the steady DC the parts need. Most of it goes out at 12 volts, with 5 V and 3.3 V for smaller parts.

Its rating, such as 750 W, is the most it can deliver. To pick one, add up what each part draws at full load, then add about 30% headroom for spikes and upgrades.

The graphics card is usually the biggest load. A mid-range card draws 200 to 300 W, and a top card can pull 575 W.

Example: each part's draw stacks up, 30% headroom is added and the smallest standard PSU that covers it is picked; chips swap the CPU and graphics card.

Practice: Size the power supply

Pick the smallest standard PSU for a random build with 30% headroom (ChoiceQuiz tiles).

Cooling the CPU

Nearly every watt a part draws ends up as heat. A CPU slows itself down, called throttling, when it nears about 95 to 100 °C.

A cooler presses a metal base onto the CPU, with thermal paste filling the tiny gaps. An air cooler carries heat up pipes into fins.

An AIO liquid cooler pumps liquid from a block on the CPU to a radiator. Fans blow air through the radiator and out of the case.

3D tour of a built gaming PC: the AIO pump on the CPU, its tubes, the radiator and its fans, the case fans and the power supply under its shroud.

Airflow through the case

Case fans move the warm air out. The usual layout pulls cool air in at the front and pushes it out at the back and top.

The more air flows through, the less it warms up on the way. Hot air also rises, so the top is a good place to let it out.

Example: air flows through a case side view, warming as it passes the parts; the exhaust temperature is computed from the heat and the airflow (chips and slider).

Practice: Set the fans

Set each fan so that air flows one way through the case.

Tap each case fan to set it to blow in or out, for a front-to-back airflow.

Overclocking

Overclocking runs a part faster than its rated clock speed, usually with a little more voltage. You gain speed, but the part draws more power and makes more heat.

Power climbs faster than speed, because the voltage counts twice. A 10% faster clock that needs 10% more voltage uses about 33% more power.

Switching power: P \propto V^2 \times f

Undervolting is the safer cousin: the same speed at less voltage, so less heat. Any overclock needs its extra heat carried away, by a bigger cooler, good airflow and the GPU's own fans.

Example: clock and voltage sliders drive the power through P ∝ V²f; the chosen cooler sets the temperature it settles at, and the chip's bin decides whether it is stable, throttles or crashes.

Practice: Work out the power

Use the power rule to find what an overclock or an undervolt costs.

Type the new power draw after a random clock and voltage change (MathAnswerInput).

What and where to overclock

Only some parts allow it. A CPU must be an unlocked model on a board whose chipset allows overclocking, and a GPU lets you raise its core clock, memory clock and power limit.

RAM is the easy start: switch on its rated profile, called XMP or EXPO, and it runs at the speed on the box. SSDs, power supplies and most other parts can't be overclocked.

Some settings live in the BIOS, the UEFI setup screen, some in Windows software and some in both. GPU settings are software only, while memory profiles and timings are BIOS only.

Example: a monitor shows the UEFI setup screen, where a pointer turns on the EXPO profile and sets the CPU ratio and voltage, then a GPU tuning app, where it raises the clocks, power limit and fan curve.

Practice: BIOS or software?

Sort each overclocking setting by where you can change it.

Sort overclocking settings into BIOS only, software only, or both (SortGroups), two rounds.

Binning and the silicon lottery

No two chips come out quite the same. Tiny differences across the silicon wafer let some chips run faster, or at lower voltage, than others.

Makers test every chip and sort them into bins. The best become the faster, pricier models, and a chip with a faulty core can be sold with that core switched off.

Even two chips of the same model differ. How far yours overclocks is called the silicon lottery.

Example: a simulated wafer of chips is tested one by one, each scored from a seeded spread that weakens towards the edge, then sorted into bins; the counts show why the top bin is rare.

Overclocking safely

Push too far and the PC crashes or freezes, losing unsaved work. An unstable chip can also make silent errors: wrong results with no warning.

Faster clocks bring more heat and fan noise. Too much voltage also wears a chip out sooner, called degradation, and overclocking can void the warranty.

Raise the clock in small steps, and test each one with a stress test and some gaming while you watch the temperatures. If a chip gets too hot anyway, it throttles or shuts down to protect itself.

Quick check on overclocking risks and safe steps (ChoiceQuiz), random questions.

Booting up

Pressing the power button wakes the PSU, which powers the board. The CPU then runs the firmware, called UEFI, from a chip on the motherboard.

UEFI runs the POST, a power-on self-test that checks the CPU, RAM and graphics. If something is wrong, it shows an error code or beeps instead.

Next it finds the boot drive and runs its boot loader, which loads the operating system into RAM. The OS starts its drivers and shows the sign-in screen.

Example: a monitor shows each stage of a boot: power on, POST checks, boot loader, operating system loading and sign-in.

Practice: Order the boot

Put the steps of a boot in the order they happen.

Put the boot steps in order (OrderSteps).

Buses, lanes and bottlenecks

Parts send data over buses, sets of wires carrying bits side by side. Graphics cards and SSDs use PCIe, built from lanes that each carry data both ways at once.

Each PCIe generation doubles a lane's speed: about 1 GB/s for PCIe 3.0, 2 GB/s for 4.0 and 4 GB/s for 5.0. A slot's width, such as x4 or x16, is its number of lanes.

A link runs at the older generation and the fewer lanes of its two ends. Whenever one step in a chain is slower than the rest, it holds back everything else: it is the bottleneck.

PCIe link speed: \text{bandwidth} = \text{lanes} \times \text{speed per lane}

Example: a card or SSD in different slots; the link takes the lower generation and the fewer lanes, and the bandwidth is computed against what the device could do.

Practice: Find the bottleneck

Tap the slowest step in a random chain of parts and links, with speeds in MB/s and GB/s.

Review

The drive stores data, RAM holds what is running, the CPU runs instructions and the GPU draws frames. Each step away from the CPU is bigger but slower.

The motherboard joins the parts, which must share a socket, a RAM type and a size. The power supply needs headroom, and fans carry the heat out.

Mixed questions from across the course (ChoiceQuiz).