DISPLAY / Input / output
Display panel
A world of pixels
The display creates images from millions of pixels. Panel type, brightness and resolution determine what you see.
A 120 Hz display looks smooth; it does not guarantee accurate colours.
COMPUTER ENGINEER
A / HARDWARE LABORATORY
Unpack three architectures layer by layer. Locate components, connect their roles, then explore how they work from beginner explanations to engineering concepts.
These are representative architectural cutaways, not service CAD for a specific product. Soldered chips separate for explanation, not as repair instructions. Layout and installed capabilities vary by product.
01 / COMPONENTS
The guide follows the chosen device. It also explains functions inside model groups and optional hardware. Changing level updates both component and technology explanations.
32/32 components · Desktop
DISPLAY / Input / output
A world of pixels
The display creates images from millions of pixels. Panel type, brightness and resolution determine what you see.
A 120 Hz display looks smooth; it does not guarantee accurate colours.
KEYBOARD / Input / output
The input matrix
The keyboard sends text and commands to the computer when you press keys.
Keyboard shortcuts rely on simultaneous key detection.
MOTHERBOARD / Platform and trust
Where every component meets
The motherboard is the main circuit board connecting computer components. The processor, memory, storage and graphics processor communicate through it. It is the starting point for checking compatibility when choosing new hardware.
Example: A seemingly compatible processor may still require a BIOS update.
CPU / Compute and architecture
The command center
The processor executes your computer's instructions. It is involved when you open an application, perform a calculation or move a character in a game. It does not do everything alone; it constantly exchanges data with other components.
Example: Compiling code with dozens of browser tabs open shares processor resources between tasks.
GPU / Compute and architecture
Graphics and parallel computing
The graphics processor creates the image you see on screen. It renders game scenes and is also used in video processing and some AI applications. Some computers have integrated graphics; others use a separate graphics processor.
Example: 4K gaming, image processing and model training stress different GPU resources.
RAM / Memory
The workbench
RAM temporarily holds information currently used by open applications. Think of it as a desk: enough working space makes it easier to handle several tasks. Information in RAM is lost when the computer powers off.
Example: A browser with many tabs, an IDE and a virtual machine increase RAM usage together.
SSD / Storage
Persistent memory
An SSD stores your files, games and operating system even when the computer is off. Unlike RAM, it is persistent. A fast SSD reduces startup and file-loading times.
Example: An SSD speeds up game loading; it does not increase frame rates in every game by itself.
VRM / POWER / Power
The right voltage
Motherboard power circuits adapt energy from the battery and adapter for components such as the processor. A laptop does not contain a separate large PSU like a desktop computer.
A low-powered adapter can limit system performance.
COOLING / Cooling
Balance that sustains performance
Working components generate heat. Heatsinks and fans remove it to keep the system stable. Poor airflow can slow even a powerful computer during sustained workloads.
Example: Rising temperatures during a long render may cause a performance drop at the thermal limit.
FANS / Cooling
Moving warm air out
Fans move warm air out of the laptop. Their speed and noise may increase during demanding tasks.
A soft bed cover can block bottom air intakes.
WIRELESS / Networking
Wireless connectivity
The wireless module connects to the internet over Wi-Fi and to headphones and other devices over Bluetooth.
A faster internet plan cannot fix a weak Wi-Fi signal by itself.
SPEAKERS / Input / output
From signal to sound
Speakers turn the audio signal produced from digital content into sound waves you can hear.
Higher wattage alone does not mean better sound.
USB / HDMI / Input / output
Connect to the outside world
USB, HDMI and audio ports connect the computer to external displays, drives, mice and headphones.
Not every USB-C port can output video to a monitor.
FIRMWARE / Platform and trust
Starting the boot process
When the computer starts, fundamental software prepares the hardware before the operating system. Modern devices generally use UEFI.
If a new drive will not boot, check the boot order.
RTC / Platform and trust
Keeping time
The clock circuit helps keep time when the computer is off. Some models use a separate small battery; others do not.
Not every laptop requires a traditional CMOS battery.
FLEX / ANTENNA / Platform and trust
Internal connections
Thin cables connect the display, camera, keyboard, touchpad and antennas to the motherboard.
An image that cuts out when the display moves can indicate a cable problem.
CHASSIS / Mechanical structure
The supporting frame
The chassis holds components together, protects them from impacts and leaves room for airflow.
The same processor can sustain different speeds in different chassis.
PSU / Power
From the wall to system rails
A desktop PSU converts AC electricity into DC rails used by components. CPU and GPU loads need suitable cables and connectors; wattage alone does not identify a good supply.
An incompatible modular cable can damage drives and boards; leave PSU servicing to qualified personnel.
HDD / Storage
Rotating platters, persistent data
An HDD stores data on magnetic platters. A head moves to the right location, making small random operations slower than SSDs. It can offer capacity and cost advantages for large archives.
Even with RAID, an accidental deletion can propagate across the array.
PCH / I/O / Platform and trust
A connection hub for peripherals
Some USB, SATA and other devices communicate through a chipset rather than directly with the CPU. Modern systems integrate some functions into the CPU or SoC; not every board uses the same separate chipset.
A SATA port becoming unavailable after adding an M.2 drive can be lane sharing rather than failure.
TPM / Platform and trust
Trust for keys and measurements
A TPM supports security operations and key protection. It can be discrete or firmware-based. It participates in disk encryption and device trust, but does not make a whole computer secure by itself.
Secure Boot, Measured Boot and disk encryption do not replace one another.
NIC / Networking
Moves packets onto the physical network
An Ethernet adapter connects the machine to a wired network. Link speed is not the same as internet throughput; switches, cables, peers and software affect the result.
A 10 Gb/s link cannot produce file data faster than the source disk supplies it.
VRAM / Memory
The GPU’s local working space
Discrete GPUs use graphics memory for images and compute data. Integrated GPUs commonly share system memory. More VRAM does not automatically mean a faster GPU in every task.
Fitting a model in memory and meeting latency requirements are separate conditions.
NPU / Compute and architecture
Hardware specialised for selected AI workloads
An NPU can accelerate selected AI operations using hardware different from a CPU or GPU. Not every machine has one, and not all software can use it.
Unsupported operators can cause CPU fallback when moving a model onto an NPU.
SoC / Compute and architecture
Multiple functions in one integrated system
A SoC integrates processing and several controllers into one system approach. It may include GPU, NPU and memory interfaces, with capabilities varying by product.
Exploded packages in the animation are conceptual, not repair or upgrade instructions.
TIM / Cooling
The contact path for heat
Paste or thermal pads fill small gaps and support heat transfer. They do not replace a cooler, and not every gap needs the same material or thickness.
If the CPU is hot but its cooler is relatively cool, investigate contact and mounting.
RAID / HBA / Storage
Drive access and array management
An HBA connects drives to the host; a RAID controller can present arrays. RAID is not backup, and not every system needs a dedicated hardware card.
Parity and snapshots do not remove the need for separately protected backups.
RISER / Platform and trust
Places expansion cards to fit the enclosure
A riser changes expansion-card orientation and placement, often horizontally in a rack enclosure. It does not create new bandwidth.
A physically x16 slot does not always carry sixteen electrical lanes.
ECC / RAS / Memory
Detecting and correcting memory faults
ECC helps detect and correct supported fault classes. It does not prevent every bit error or all data corruption. Support is required across the system.
Do not automatically classify every DDR5 machine as an ECC workstation.
NUMA / Compute and architecture
Memory locality matters
In some multi-processor systems, each CPU accesses its local memory more directly. Remote memory can have different latency. More cores do not speed every application proportionally.
Installing two CPUs does not guarantee balanced memory or PCIe use.
ADC / DAC / Input / output
Between analogue and digital
Audio circuits convert microphone input to digital data and playback to analogue output. Speaker, amplifier and codec are separate functions.
Investigate software buffers and the analogue chain together when diagnosing recording latency.
THERMAL / POWER / Platform and trust
The system monitoring itself
Temperature, fan, voltage and power sensors help monitor operation. Not every reported number measures the same physical point.
Do not compare machines without matching sensors, workloads and conditions.
02 / TECHNOLOGIES AND RELATIONSHIPS
From instruction sets to CXL, data integrity and mainframe architecture. These chapters compare device families at your chosen explanation level.
An instruction set defines operations software can describe to a processor. Two compatible processors need not have equal speed or power use.
Technical referenceA core is a hardware execution resource. Software threads differ; creating hundreds of threads does not create hundreds of cores.
Technical referenceCPU caches keep useful data close to execution. They do not replace RAM or storage. A miss can require accessing a more distant level.
Technical referencePrograms use address spaces mapped onto physical memory. Some data can be paged to storage; that does not make an SSD equivalent to RAM.
Technical referenceCapacity, transfer rate and latency differ. DIMMs serve desktops and servers, SO-DIMMs many laptops; some memory is soldered.
Technical referencePCIe connects expansion devices. x1/x4/x8/x16 describe lane counts; a large physical slot does not guarantee the same electrical width.
Technical referenceM.2 is a physical format/interface arrangement; NVMe is a storage protocol family. Not every M.2 drive is NVMe. SATA, SAS and NVMe differ.
Technical referenceSSDs use flash cells. A controller maps blocks to physical cells; the visible address is not a permanently fixed cell.
Technical referenceRAID can tolerate some drive failures, while backups serve another purpose. It alone does not address deletion, ransomware or common failures.
Technical referenceUSB-C describes a connector. Shape alone does not establish speed, video or power capability; both endpoints and the cable must support it.
Technical referenceA GPU draws images and supports suitable parallel compute. Graphics and inference use different software paths; a strong GPU does not accelerate every program.
Technical referenceOLED pixels emit light; LCD uses backlighting. IPS is an LCD approach, while mini-LED commonly describes backlighting. These are not equivalent feature categories.
Technical referenceRadio communication depends on antennas, environment and peers. A fast internet plan does not repair weak Wi-Fi through walls.
Technical referenceA NIC carries packets; copper or fibre provides the physical path. Network capacity differs from application data-production rate.
Technical referenceFirmware initialises hardware and starts OS boot. UEFI is a firmware interface, distinct from the OS and its drivers.
Technical referenceComponents do not always run at full power. Clock/voltage scaling, idle states and sleep change consumption.
Technical referenceCooling transfers heat to the environment. Laptops, towers and racks have different space and airflow. Fan noise alone does not prove performance quality.
Technical referenceA hypervisor shares hardware among VMs. Containers and VMs have different isolation models. Assigned vCPUs are not guaranteed dedicated physical cores.
Technical referenceCXL can enable memory and coherent device scenarios on supported platforms. A PCIe slot or SSD does not automatically support it.
Technical referenceCPUs, GPUs and NPUs suit different AI work. An AI label does not establish support for every model or tool.
Technical referenceServers often prioritise continuity and serviceability. Two PSUs or ECC alone do not guarantee uninterrupted applications; software and operations also matter.
Technical referenceServer describes a service role or system class. Mainframes are a different platform family designed for large-scale transactions, I/O and continuity. Not every rack server is one.
Technical referenceThe limiting path constrains a system. Replacing a CPU does not fix every slowdown. Identify the actual workload and stage first.
Technical reference03 / READ THE ARCHITECTURE
| Decision | Desktop | Laptop | Rack server |
|---|---|---|---|
| Power | ATX PSU; component budget | Adapter + battery + charging | Redundant PSUs where supported |
| Memory | DIMMs; platform-dependent ECC | SO-DIMM or soldered LPDDR | Validated ECC DIMMs and population |
| Cooling | Case airflow, tower/AIO | Heat pipes/chambers; constrained space | Fan wall, ducts, inlet policy |
| Service | Modular cards/cables; compatibility | Model-dependent upgradeability | Supported hot swap, BMC and RAS |