Skip to content
Tutorials Center
My learning

Topics

Lesson 15 of 55 ยท Inside the Machine

Motherboard, Buses, GPU and Power

See how the motherboard connects the CPU, memory and devices, what buses and PCIe lanes are, why a GPU is good at parallel work, and how to read a spec sheet.

The processor, memory and storage you met in the last lessons cannot work alone. Something has to hold them, carry signals between them, feed them electricity and carry the heat away. That job belongs to the motherboard, the large circuit board that every other part plugs into, together with the power supply and the cooling system.

This lesson also introduces the GPU (graphics processing unit), a second kind of processor built to do huge amounts of simple work at the same time. By the end you will be able to read a product page and say what the numbers mean. It follows Storage: HDD, SSD and Beyond and ends the chapter before we climb up to logic and programs.

The motherboard: the city map of a computer

A motherboard is mostly copper wiring (called traces) printed in layers inside a board, with sockets and slots on top. The important parts are:

  • CPU socket. Holds the processor. Sockets are specific: a processor made for one socket shape will not fit another.
  • Memory slots. Hold RAM modules. The board decides which memory type (for example DDR4 or DDR5) it accepts.
  • Expansion slots. Mostly PCIe slots for graphics cards, network cards, capture cards and fast storage.
  • Storage connectors. M.2 slots for small SSD sticks and SATA ports for 2.5-inch drives and optical drives.
  • Chipset. A set of controller chips that manages traffic between the CPU and the many slower devices.
  • Firmware chip. Stores the UEFI/BIOS firmware that starts the machine (see Hardware and Software).
  • Back-panel ports. USB, video, audio and network connectors.
  • Power connectors. Where the power supply attaches.

Laptops and phones use the same ideas but squeeze everything onto smaller boards, often soldering the processor and memory in place. That is why you can upgrade a desktop piece by piece but usually cannot upgrade a thin laptop.

flowchart TB
  CPU["CPU"] ---|"memory channels"| RAM["RAM slots"]
  CPU ---|"PCIe lanes"| GPU["Graphics card"]
  CPU ---|"PCIe lanes"| NVME["M.2 SSD"]
  CPU ---|"link"| CH["Chipset"]
  CH --- USB["USB ports"]
  CH --- SATA["SATA drives"]
  CH --- NET["Network and audio"]

The fastest devices connect straight to the CPU; the chipset gathers the slower ones behind a single link.

The exact layout varies by processor generation. In some designs the chipset is a separate chip, in others its functions are built into the CPU package. What stays the same is the principle: fast, demanding devices get a short, direct path; slow, numerous devices share a hub.

Buses and lanes

A bus is a set of wires, plus rules for using them, that carries data between components. Older diagrams show one shared bus that everything listens to. Modern computers rely mostly on point-to-point links: each device has its own private connection, which avoids waiting for a shared wire.

Two numbers describe any link:

  • Width: how many bits move at the same time (or how many lanes there are).
  • Speed: how many times per second the signal can change.

Bandwidth is roughly width times speed, minus protocol overhead.

PCI Express (PCIe) is the standard connection for graphics cards, SSDs and other add-in devices. A PCIe connection is made of lanes. Each lane is two pairs of wires, one pair for each direction, so data can flow both ways at once. Slots come in sizes by lane count: x1, x4, x8 and x16. More lanes means more bandwidth. Each new PCIe generation roughly doubles the speed per lane.

Approximate usable bandwidth per lane, per direction:

PCIe generation Per lane
3.0 about 0.985 GB/s
4.0 about 1.97 GB/s
5.0 about 3.94 GB/s

You can compute any slot's bandwidth by multiplying.

JS PCIe bandwidth calculator

Press Run (or Ctrl + Enter) to see the console output here.

No output.

The numbers are theoretical maximums. A real SSD, cable or program rarely reaches them, but they explain why a drive in an x4 slot can be several times faster than a SATA drive, and why putting a fast device in a slot with fewer lanes than it wants wastes its speed.

USB (Universal Serial Bus) connects peripherals from keyboards to external drives. It is a family of speeds, which is why two identical-looking ports can behave very differently:

  • USB 2.0: 480 Mbit/s
  • USB 3.2 Gen 1: 5 Gbit/s
  • USB 3.2 Gen 2: 10 Gbit/s
  • USB4: up to 40 Gbit/s

Remember that network and USB speeds are quoted in bits per second while file sizes use bytes. Divide by 8 for a rough byte rate, so 480 Mbit/s is at most about 60 MB/s before overhead. The connector shape (USB-A, USB-C) says nothing about speed: USB-C is only the plug, and the cable and both ends decide the standard in use. Other common links are HDMI and DisplayPort for video, Thunderbolt for very fast external devices, and SATA for older-style drives.

The GPU: many small workers

A CPU has a few powerful cores that are excellent at long, branching, step-by-step work. A GPU has hundreds or thousands of much simpler execution units. It is built for jobs where the same small calculation is applied to a huge amount of data, such as colouring each of the roughly 2 million pixels on a 1920 by 1080 screen. Every pixel can be worked on independently, so thousands can be handled at once.

That same shape of problem appears outside games. Video encoding, 3D rendering, scientific simulation and training neural networks all do the same arithmetic on large arrays, so GPUs now power much of modern AI.

Two kinds of GPU exist:

  • Integrated GPU: built into the CPU package, shares system RAM, uses little power. Fine for everyday use, video and light games.
  • Dedicated (discrete) GPU: a separate card or chip with its own fast memory (VRAM) and its own power needs. Needed for demanding games and heavy compute.

A GPU is not simply a "faster CPU". It is faster only when the work can be split. A task with steps that depend on each other, where step 2 needs the result of step 1, gains almost nothing from a thousand workers. The following stand-in models both situations in counted steps (not wall-clock time, so the result is always the same).

JS Parallel versus serial work

Press Run (or Ctrl + Enter) to see the console output here.

No output.

The second half is Amdahl's law. If half your program cannot be split, the best possible speed-up is 2x, however many workers you add. The stand-in is a simplified model; real GPUs also pay to copy data to their own memory over PCIe, which is one more reason bus bandwidth matters.

Power and heat

Every watt that goes into a computer eventually becomes heat. The power supply unit (PSU) converts the wall's alternating current into the several low direct-current voltages the components need. Its rating is in watts, and it must cover the total of all parts at their peak, with some headroom. A quality unit that runs at moderate load is quieter, cooler and more efficient. Efficiency labels such as 80 Plus describe how much input power becomes useful output rather than waste heat.

Heat is the matching problem. Chips slow themselves down (thermal throttling) when they get too hot, and sustained high temperatures shorten life. Cooling uses:

  1. Heat sinks: metal fins that spread heat over a large area.
  2. Fans or liquid loops: move air or coolant across the fins.
  3. Thermal paste: fills microscopic gaps between a chip and its cooler.

Thin laptops and phones often have no fan, so they throttle quickly under long loads, even though the chip inside is the same family as a desktop part.

Never open a power supply or touch its internals. Capacitors can hold a dangerous charge even when unplugged. Also unplug a desktop before touching its parts.

Reading a spec sheet

A listing might say: 8 cores / 16 threads, 4.2 GHz boost, 32 GB DDR5, 1 TB NVMe PCIe 4.0 x4, 12 GB VRAM, USB4, 650 W PSU. Translate it with the questions below:

  • Cores and threads: how much work at once? (see The CPU)
  • RAM size and type: how many programs and large files fit? (see Memory)
  • Storage type, interface and size: SSD or hard drive, and which link?
  • GPU and VRAM: matters for games, video editing and AI work.
  • Ports and standards: not just "has USB-C", but which speed.
  • Power rating and cooling: will it stay fast under long loads?

A big number in one place does not rescue a weak link elsewhere. A fast GPU on an old CPU, or a fast SSD behind a slow port, is held back by the slowest part of the chain.

Check your understanding

Quick check ยท 5 questions

of 5 answered You scored / 5 ยท perfect!

  1. 1. What is the main job of the motherboard?

    The motherboard holds the sockets, slots and traces that link the CPU, memory, storage and other devices together.
  2. 2. A PCIe x4 slot compared with an x1 slot of the same generation has about...

    Each lane carries its own data, so four lanes give roughly four times the bandwidth of one lane.
  3. 3. Why is a GPU good at colouring millions of pixels?

    GPUs shine when the same small calculation is applied to lots of independent data.
  4. 4. Does a USB-C connector guarantee a fast transfer speed?

    The connector and the speed standard are separate things. A USB-C port can run USB 2.0 speeds.
  5. 5. A program is 90% parallel. With unlimited workers, the best speed-up is about...

    The 10% serial part still takes the same time, so the total can shrink to at most one tenth: 10x.

Summary

  • The motherboard links every component with traces, sockets and slots; the chipset gathers slower devices behind one link.
  • Buses and lanes carry data; bandwidth is roughly width times speed, and PCIe doubles per generation.
  • USB is a family of speeds; the plug shape does not tell you which one you have.
  • A GPU has many simple units and wins on work that can be split into independent pieces.
  • Amdahl's law: the serial part of a job limits the benefit of extra workers.
  • The power supply must cover peak demand, and heat decides how long a machine can stay fast.
  • When reading specs, look for the weakest link in the chain.

Next, we leave the hardware and look at how computers reason: Boolean Logic and Logic Gates.