Computer Hardware Components

A desktop PC brings several replaceable parts together inside one case. Start by asking what each part does. A large number on a specification sheet means little until you know whether it describes working memory, saved files or electrical power.

Follow the work through the parts

The CPU, or processor, carries out program instructions. RAM holds the programs and data currently in use. It is volatile: its contents disappear when power stops. Storage keeps the operating system, apps and saved files with the power off. An SSD and a hard drive both provide storage, although they store and retrieve data differently.

Think about editing a photograph. The saved file sits on the drive. Opening it puts data into RAM so the CPU can work on it quickly. Saving writes the result back to storage. Adding drive capacity gives more room for files; adding RAM gives open work more room. A PC can have plenty of free drive space and still run short of RAM.

The motherboard is the main circuit board. Its socket, slots, connectors and ports link the processor, memory, drives and external devices. The chipset provides many of those connections. The PSU converts wall-outlet AC into the low-voltage DC that the parts use, mainly 12 V, 5 V and 3.3 V. Heat sinks, fans and liquid coolers carry away heat produced during operation.

Know where the picture comes from

The GPU produces the display image. Integrated graphics are built into many CPUs and use some system RAM. A dedicated graphics card has a separate GPU and video memory, called VRAM. Dedicated cards provide much more graphics performance for demanding games, 3D work and video editing. The monitor, keyboard, mouse and speakers are peripherals outside the case.

Firmware stored on the motherboard starts the machine. BIOS or UEFI runs the power-on self-test (POST), then starts the operating system from a drive. A small coin-cell battery, commonly a 3 V CR2032, keeps the clock running while the machine is unplugged.

Read a complete specification

Consider: "8-core CPU, 3.8 GHz; 16 GB DDR5-5600 (2 × 8 GB); 1 TB NVMe SSD; 8 GB graphics card; 650 W PSU; microATX motherboard."

The CPU has eight cores and a 3.8 GHz base clock. The two memory modules provide 16 GB of RAM at 5,600 MT/s. The 1 TB figure describes persistent storage. The graphics card's 8 GB is VRAM, separate from system RAM. The PSU's 650 W is its output capacity, and microATX describes the board's physical format.

Find the installed parts

In Windows, Settings > System > About lists the CPU and RAM. Task Manager's Performance tab shows processor, memory, disk and GPU details. System Information (msinfo32) and Device Manager provide more detail. A Mac offers About This Mac and System Information; Linux tools include lscpu, free -h and lsblk. Firmware setup also lists major components.

Software cannot measure every physical clearance or identify every PSU cable. Read labels and manuals for those checks. Desktop towers usually allow individual replacements. Laptops and all-in-one systems use more compact designs, and soldered parts cannot be upgraded as plug-in modules can.

A motherboard must fit both the case and the components. Its size tells you where it mounts. Its socket, chipset and firmware tell you which processors it can run. Check these separately before choosing a board.

Match the board to the case

A form factor sets board dimensions, mounting-hole positions and the rear I/O position. Three common desktop formats are:

Mini-ITX means Information Technology eXtended. Larger E-ATX boards appear in workstations, but case support needs checking carefully. An ATX mid-tower commonly accepts smaller microATX and Mini-ITX boards through shared mounting positions. A Mini-ITX case cannot hold a larger microATX board. The case's supported-form-factor list settles the choice.

Match the processor in three checks

First check the exact socket. LGA designs put the pins in the socket and flat contact pads on the CPU. Examples include Intel LGA1700 and LGA1851, and AMD AM5, which is LGA1718. PGA designs put the pins on the CPU, as with AMD AM4. Similar-looking parts with different socket names are not substitutes.

Next check chipset support. The chipset provides many extra USB ports, SATA ports and PCIe lanes, and determines features such as CPU overclocking. Intel Z-series boards support overclocking with unlocked "K" processors; B- and H-series boards do not support that CPU-overclocking combination.

Finally check the board maker's CPU support list for the required firmware version. A processor can fit the socket yet fail to start because the installed firmware does not support it. The support list names the processor and the earliest suitable firmware.

Check memory and expansion together

The CPU's memory controller sets supported DDR generations, speeds and capacity limits. The board supplies slots for a particular generation and has its own capacity limit. A CPU that supports DDR4 and DDR5 still needs modules matching the chosen board: a DDR4 board cannot take DDR5 modules.

Count the expansion positions and check their spacing. A large graphics card can cover neighboring positions. Also count storage connectors, needed rear ports and memory slots. A higher-end chipset usually offers more connections, but those connections are useful only if the build needs them.

Server and some workstation boards have multiple CPU sockets. They need matching supported processors and provide more cores and memory channels. Ordinary desktop boards have one socket.

Choose for a compact office PC

Suppose an AM5 CPU includes graphics, and the build needs two memory modules and one NVMe SSD. An AM5 microATX or Mini-ITX board with DDR5 slots and a suitable M.2 slot can meet the requirement if the case supports it. A later requirement for a graphics card, capture card and network card rules out Mini-ITX's single expansion position. Check an ATX or microATX board's actual slot layout before ordering the larger system.

A motherboard has several kinds of connection because power, data and case controls do different jobs. Use the board diagram in the manual to identify each connector before attaching a cable or card.

Read PCIe slots correctly

PCI Express (PCIe) moves data along serial lanes. The common sizes are x1, x4, x8 and x16. A smaller card can fit a larger slot. The long x16 slot nearest the CPU is the usual graphics-card position. Physical length does not guarantee lane count: "x16 (x4 mode)" means a long slot wired for only four lanes.

PCIe 3.0 carries about 1 GB/s per lane in each direction; 4.0 about 2 GB/s; 5.0 about 4 GB/s. A card and slot negotiate their shared generation. A PCIe 4.0 x4 SSD in a PCIe 3.0 x4 slot works, but its link ceiling falls from about 7.9 GB/s to 3.9 GB/s. These are link limits, not guaranteed file-copy speeds. Older PCI uses a different parallel bus and slot; a PCI card cannot be fitted into PCIe.

Locate storage and power connections

SATA data ports have a small, L-shaped 7-pin connector. M.2 slots accept small card-edge devices held at the far end by a screw or latch. An older board may offer external SATA, or eSATA, on its rear panel. Some M.2, SATA and PCIe connections share lanes. Installing a drive in one slot can disable another port, so read the lane-sharing table before assigning drives.

The large 24-pin ATX connector powers the board. The 8-pin EPS12V connector near the CPU powers the processor; it often separates into 4+4 pins. Some boards have another CPU power socket. Fan headers use three pins for voltage control or four for pulse-width modulation (PWM). CPU_FAN serves the CPU cooler, and a PUMP or AIO header can serve a liquid-cooler pump.

Connect the case to its headers

A header is a group of pins for a cable from the case or an accessory. The front-panel header takes the power switch, reset switch and indicator LEDs. Other headers include USB 2.0 (9 pins), USB 3.x (19 pins), front USB-C, front audio and RGB lighting. Some boards have a TPM header. The nearby Clear CMOS jumper or button restores firmware settings to defaults.

The rear I/O panel exposes built-in USB, Ethernet and audio ports. It may also provide video outputs, Wi-Fi antenna connectors or a firmware-flashback button. Motherboard video outputs need supported integrated graphics enabled in the system.

Add the card that supplies the missing function

A graphics card adds a GPU; a sound card adds audio connections; a capture card records video from an HDMI source; a network interface card (NIC) adds Ethernet, Wi-Fi or fiber connectivity.

With the PC prepared for internal work, remove the correct case slot cover. Align the card with the slot, press it straight in, secure its bracket, and attach any required power. Start the PC and install its driver. Connect a monitor used by a dedicated GPU to that card's outputs. Plugging it into an inactive motherboard video output is a common cause of a blank screen.

A processor must run the required software, support the right hardware and suit the user's work. Core count and clock speed help describe it, but neither number alone tells you which CPU will perform best.

Check the architecture

The architecture is the instruction set the processor understands. Most desktop PCs use Intel or AMD x86-64 processors, also called x64. ARM processors power phones, tablets, Apple silicon Macs and many newer Windows laptops. Software must support the architecture. Emulation can run many x86-64 applications on ARM, sometimes with a performance cost, but it cannot replace a missing hardware driver. Check ARM driver support for devices and software that depends on drivers, such as VPN clients.

A 32-bit address space spans 4 GiB. Some x86 systems use Physical Address Extension (PAE) to reach more physical memory, depending on the operating system. A 64-bit processor with a 64-bit operating system supports much more and can still run many 32-bit applications. A 64-bit operating system cannot run on a 32-bit processor.

Separate cores, threads and clock speed

A core is a processing unit inside the CPU. Simultaneous multithreading, or SMT, lets a core handle two hardware threads; Intel calls its implementation Hyper-Threading. An eight-core CPU with two threads per core handles 16 threads. Hybrid CPUs combine performance cores and efficiency cores. Six P-cores plus eight E-cores make 14 cores, but do not assume every core supports SMT.

More cores help video encoding, compiling and virtual machines because these workloads can divide work into parallel parts. Many everyday apps and games rely heavily on a few fast cores. Clock speed, in GHz, counts billions of cycles per second. A 4.5 GHz CPU runs 4.5 billion cycles each second, but different designs complete different amounts of work per cycle. Use relevant benchmarks to compare different generations or architectures.

The base clock is a reference operating rate. Boost can stay above base as long as workload, temperature, power and current limits permit. Good cooling can allow sustained boost; it is not always a brief burst. Small, fast cache memories, called L1, L2 and L3, keep useful data near the cores. Extra L3 cache helps some workloads, especially many games.

Check graphics and power needs

Many CPUs include graphics; others do not. Intel desktop models with an "F" suffix, such as "12400F," need a graphics card for display output. A motherboard HDMI socket cannot supply graphics that the CPU lacks.

TDP is the maker's cooler-design rating, not maximum electrical draw. A 65 W AMD desktop CPU can have an 88 W package-power limit. Read Intel maximum turbo power or AMD package power tracking (PPT) when estimating PSU demand. Choose a cooler from its maker's compatibility list for the exact processor; cooler wattage labels are not standardized.

Choose for the task

An office PC without a graphics card can use a supported six-core AM5 CPU with integrated graphics. An eight-core AM5 alternative without graphics needs another card and may need stronger cooling. Its extra cores do not fix the missing display output. Server and workstation families such as Xeon, EPYC and Threadripper add features including ECC support and more memory channels; some also support multiple CPU sockets.

Cooling moves heat away from the CPU so it can keep working at its intended speed. A CPU that gets too hot lowers its clock through thermal throttling. If temperature keeps rising, the system shuts down to protect itself.

Choose a cooling path

An air cooler uses a metal heat sink, often with heat pipes, and a fan. A low-power system can use a passive heat sink. A liquid cooler moves heat from a cold plate to a radiator. An all-in-one (AIO) cooler is a sealed, ready-filled loop with a pump and radiator fans.

Thermal paste fills microscopic gaps between the processor's lid and cooler base. It improves heat transfer; it is not a replacement for a properly mounted cooler. Thermal pads perform a similar gap-filling job on parts such as M.2 drives and voltage regulators.

Seat the CPU and cooler

Prepare the PC for internal work before opening the socket. For an LGA CPU, open the load lever and plate. Align the CPU's triangle and notches with the socket marks. Lower it straight into position, then close the plate and lever. Keep the socket's protective cover. Never force the processor: LGA socket pins and PGA CPU pins bend easily.

Clean old paste from both surfaces with a lint-free cloth and isopropyl alcohol of at least 90%. Follow the cooler's paste instructions; a small dot in the center is a common method. If the base already has paste, add no extra. Remove any protective plastic film. Too much paste, missing paste and forgotten film are common fitting errors.

Install the correct bracket or backplate. Tighten screws gradually in a diagonal pattern to keep pressure even. Connect the cooler fan to CPU_FAN and an AIO pump to the specified pump or AIO header. A tall air cooler must clear both RAM and the case side panel. A radiator needs the correct mounting space, such as 240, 280 or 360 mm, plus room for its fans.

Give the heat somewhere to go

Case fans commonly draw cool air in through the front or bottom and exhaust warm air at the rear or top. Arrows on the fan frame show direction. Slightly more intake than exhaust creates positive pressure, encouraging air to leave through gaps while incoming air passes through filters. Keep those filters clean.

Position an AIO so its pump is not the loop's highest point. Air collects at the high point; air inside the pump causes noise and poor cooling. A top radiator, or a front radiator with tubes at the bottom and its top above the pump, avoids that arrangement.

Read temperature with clock speed

Firmware monitoring and hardware-monitoring apps show temperature and fan speed. Desktop CPUs often idle around 30–50 °C and reach 60–90 °C under heavy load. Maximum temperatures are model-specific, commonly around 95–100 °C. Some CPUs are designed to work near that limit under full load.

A video export that begins at 4.8 GHz but falls to 2.5 GHz as temperature reaches 100 °C shows throttling. Check fan operation, dust, mounting pressure, paste and film. Then check the cooler maker's suitability list for that exact CPU. A small cooler intended only for low-power processors may be unable to remove sustained heat even when it fits the socket.

Memory selection involves capacity, physical fit, generation, speed and channel layout. Getting one of these right does not make up for getting another wrong. A fast kit of the wrong DDR generation cannot be installed.

Match the module to the platform

RAM holds active programs and data. When it fills, the operating system uses a page file or swap space on the drive, which is much slower. About 16 GB suits ordinary office work; 32 GB or more helps heavy multitasking, virtual machines and video editing.

Desktops normally use DIMMs. Smaller PCs and laptops often use shorter SODIMMs. They do not fit each other's slots.

DDR generations have different notch positions and electrical designs. DDR4 and DDR5 DIMMs both having 288 pins does not make them interchangeable. DDR5 adds on-module power management and splits each module into two 32-bit subchannels.

Read the speed label

"DDR4-3200" describes 3,200 million transfers per second, or MT/s. It is often loosely labeled "3200 MHz." A 64-bit-wide module moves eight bytes per transfer. Its theoretical rate is:

That is 25,600 MB/s, giving the alternative label PC4-25600. DDR5-5600 similarly gives PC5-44800. CAS latency (CL) counts clock cycles before data arrives. At the same transfer rate, lower CL means less delay; do not compare CL numbers alone across different speeds.

Use the memory channels

A typical desktop CPU has two memory channels. Put a matched pair in the manual's recommended slots, usually A2 and B2 on a four-slot board. Two DDR4-3200 channels provide a peak 51.2 GB/s, double one channel's 25.6 GB/s. Two 8 GB modules and one 16 GB module have the same capacity, but the pair can use both channels. Two modules in A1 and A2 occupy only one channel.

Choose an identical kit and check the qualified vendor list (QVL) of tested modules. Mixed memory needs shared settings and may run more slowly or become unstable. Unequal channel capacities can use a dual-channel portion and a single-channel remainder on platforms that support that mode; matching kits simplify a predictable build.

ECC modules add check bits to correct single-bit errors. Both board and CPU must support ECC operation. DDR5's internal on-die ECC is different and does not provide the same system-level protection. Registered server memory, or RDIMMs, also needs a supporting platform; ordinary desktop UDIMM slots do not take RDIMMs.

Install and verify

Open the slot latches, align the notch and press the module straight down until it clicks into place. Check full capacity and speed in firmware or Task Manager. XMP or EXPO selects the kit's performance profile; without it, a DDR5-6000 kit may start at 4,800 MT/s. The CPU, board and kit must support the chosen settings. Give new memory time to train on first startup, especially with DDR5.

Storage choices balance response time, capacity, price and portability. A drive for active applications has a different job from one holding a large archive, so compare the workload before comparing advertised transfer speeds.

Compare moving disks and flash

An HDD stores data magnetically on spinning platters. A moving arm places its heads over the data. Typical spindle speeds include 5,400 rpm for quieter, cooler drives and 7,200 rpm for desktop performance. Enterprise drives have also used 10,000 and 15,000 rpm. Mechanical movement adds delay when the drive must find many small files. Clicking or grinding can signal failure.

An SSD stores data in NAND flash, with no moving parts. It is silent, more resistant to knocks and much quicker at random access. Flash wears with writing; endurance is often rated in terabytes written (TBW). Ordinary desktop workloads rarely reach the rating during normal service, but it matters for write-heavy work.

For starting apps, an SSD's low random-access delay matters more than the biggest sequential number. The 3.5-inch HDD format is about 102 mm wide. A 2.5-inch drive is about 70 mm wide; many SATA SSDs use it too. These industry names do not measure the outside case width.

Count capacity in the same units

Drive makers use decimal units: 1 TB is one trillion bytes. A GiB is 2³⁰ bytes. Windows commonly reports a binary quantity with a "GB" label, so a 1 TB drive appears near 931 GB before space used by formatting and other data. A 2 TB drive appears near 1.8 TB and an 8 TB drive near 7.3 TB. The unit conversion itself has not removed any bytes.

A video editor can use a 2 TB NVMe SSD for current projects and an 8 TB 7,200 rpm HDD for finished work. The SSD serves frequent work quickly while the HDD supplies affordable bulk capacity. Separate backups still protect the files.

Choose removable media the device supports

USB flash drives and SD or microSD cards are portable and easy to lose; encrypt sensitive files. Capacity families matter: SDHC reaches 32 GB, SDXC 2 TB and SDUC 128 TB. An SDHC-only reader cannot read an SDXC card simply because the card fits.

Optical media also differ. A CD holds about 700 MB; DVD holds 4.7 GB, or 8.5 GB dual-layer; Blu-ray holds 25 GB, or 50 GB dual-layer. ROM is read-only, R is write-once, and RW or RE is rewritable. A DVD drive cannot read Blu-ray. A Blu-ray drive normally reads DVDs and CDs. Where a case has no optical bay, an external USB drive supplies the function.

A storage label can describe shape, protocol or link speed. Keep those separate when matching a drive to a board. M.2 is a physical format; NVMe is a storage protocol; PCIe provides the lanes that NVMe commonly uses.

Connect a SATA drive

A SATA HDD or 2.5-inch SATA SSD needs two connections. A 7-pin L-shaped data cable runs to the motherboard. A wider 15-pin power cable runs from the PSU. SATA III signals at 6 Gb/s, with about 600 MB/s available after encoding; real SATA SSDs reach around 550 MB/s.

Secure the drive in a bay or tray before connecting it. A 2.5-inch drive may need a mounting bracket in a 3.5-inch bay. External SATA, or eSATA, carries data through a shielded connection but does not normally provide power. USB and Thunderbolt have largely replaced it.

Read M.2 size and protocol

"2280" means 22 mm wide and 80 mm long. Other lengths include 2230, 2242 and 22110. Move the mounting standoff to the correct length; a longer slot does not mean a shorter drive can be left unsupported.

The connector notch is its key. M-key drives commonly use PCIe x4 NVMe, although some M-key SATA drives exist. B+M cards have two notches and are commonly SATA or PCIe x2. E-key slots commonly take Wi-Fi cards. Keying is a useful clue, but the board manual's supported protocols decide compatibility. An NVMe drive in a SATA-only M.2 slot is not detected.

NVMe is designed for flash storage. It handles many requests in parallel with low delay, normally over four PCIe lanes. A newer PCIe drive can run in an older compatible PCIe slot at the shared speed. That backward compatibility does not turn a SATA-only slot into PCIe.

Install and account for shared lanes

With the system prepared for internal work, remove the slot heat sink if fitted. Insert the drive at roughly 30°, lower it to the standoff and secure the screw or latch. Remove protective film from the heat sink's thermal pad before refitting it.

Suppose the manual says:

• "M2_1: M key, 2242/2260/2280, PCIe 4.0 x4 only."
• "M2_2: M key, 2280, PCIe 3.0 x4 or SATA; SATA port 1 is disabled by an M.2 SATA drive."

A 2280 SATA card goes in M2_2, and a cabled drive must use a SATA port other than port 1. A PCIe 4.0 NVMe drive fits M2_1 for its higher speed; M2_2 limits it to PCIe 3.0.

Recognize other interfaces and finish setup

Older compact systems may have mSATA, a small SATA card format that is different from M.2. Servers use SAS, at 12 Gb/s for SAS-3 and 22.5 Gb/s for SAS-4. A SAS controller can support SATA drives; a SATA port cannot run a SAS drive.

A blank new drive can appear correctly in firmware yet be absent from File Explorer. In Windows, Disk Management initializes, partitions and formats it for use. Confirm which drive is new before making changes; a detected but unformatted drive is different from one missing at the hardware level.

RAID combines drives into a logical volume. It can improve speed, keep a system available after a drive failure, or do both. The RAID level determines how much capacity remains and which failures the array survives.

Compare the layouts

Striping spreads blocks across drives. Mirroring stores another copy. Parity stores calculated information that can reconstruct missing data. A controller, motherboard firmware or operating-system software can manage the array; examples include Windows Storage Spaces and Linux mdadm.

RAID 0 provides speed by striping, but one failed member breaks the array. RAID 1 writes the same data to both drives. Reads can be faster; mirroring does not double write speed. RAID 5 distributes parity and keeps more capacity, but parity work slows writes and a rebuild places heavy load on the surviving drives. RAID 6 uses double parity to tolerate a second failure, useful when large drives take a long time to rebuild. RAID 10 stripes across mirrored pairs for speed and protection.

Work out usable space

With four 4 TB drives, the raw total is 16 TB. RAID 0 uses all 16 TB. RAID 5 reserves the equivalent of one drive for distributed parity, leaving 12 TB. RAID 6 leaves 8 TB after two drives' worth of parity. RAID 10 leaves 8 TB because half the space holds mirrored copies. A two-drive RAID 1 made from 4 TB drives holds 4 TB.

For RAID 5 with equal drives, use:

For RAID 6:

Here, n is the number of drives and d is one drive's capacity. Different-sized members are normally treated as the smallest member's size for these layouts, wasting the extra space on larger members.

Follow the failures, not just their count

In RAID 10, pairs A/B and C/D can lose A and C and keep running. Losing A and B destroys the same mirror, so the array fails even though only two drives have failed. RAID 6 with four members survives any two failures; RAID 10 does not promise that.

A redundant array within its tolerance keeps operating in a degraded state. Replace the failed member promptly so it can rebuild. A hot spare is an installed spare that can start the rebuild automatically. RAID 0 has no degraded survival mode when a member fails.

Keep recovery separate from availability

RAID is not a backup. Deletion, corruption and ransomware can affect the logical volume and its redundant copies together. Keep separate backups. An office file store may use RAID 5 or RAID 6 to keep working after failure. A busy database may favor RAID 10. RAID 0 suits disposable scratch files only when their loss is acceptable or the originals exist elsewhere.

The PSU supplies every part with the voltage and current it needs. Its wall input, case size and output connectors must all match the build. A plug that looks close to the right shape is not enough.

Read the input and output labels

The PSU converts alternating current (AC) from the outlet into direct current (DC). The main outputs are +12 V for high-power parts such as the CPU, GPU, fans and drive motors; +5 V for USB and drive electronics; and +3.3 V for some board chips and slots. The +5 V standby line supports wake and charging features while the PC is plugged in.

Most modern units automatically accept 100–240 V AC at 50 or 60 Hz. This covers both the roughly 100–120 V and 220–240 V supplies used around the world. Confirm the printed range. Older units may have a red 115/230 selector: 115 selected on 230 V can destroy the PSU, while 230 selected on 115 V can prevent reliable startup. Set it correctly before connecting power.

The common inlet is IEC C14 and takes a C13 cord. With an auto-ranging supply, moving to another country normally requires a suitable local power cord, not a new PSU. Match the PSU's physical format too: ATX units are 150 mm wide and 86 mm tall, with varying depth; SFX units are smaller, at 125 × 63.5 mm across those dimensions.

Match each output connector

EPS12V and 8-pin PCIe use different wiring and keying. Read "CPU" or "PCIe" on the cable; never force either into the other's socket. SATA power carries 3.3 V, 5 V and 12 V; the older Molex connector provides 5 V and 12 V.

Understand the graphics-card power budget

A PCIe x16 slot supplies up to 75 W. A 6-pin cable adds up to 75 W; an 8-pin cable up to 150 W. Two 8-pin inputs plus the slot give:

That is a 375 W connection budget, not a statement that the card always draws 375 W. The newer 16-pin 12V-2x6, succeeding 12VHPWR, supports up to 600 W with the appropriate cable and equipment. Seat it fully until its latch clicks and avoid sharp bends at the connector; incomplete seating can overheat it.

Work safely around the supply

Switch off and unplug the PC before internal work; standby power remains while plugged in. Do not open a PSU: its capacitors can retain a dangerous charge after disconnection. Replace a faulty PSU as a complete unit. Handle its outside mounting and cables only, and follow the assembly precautions when handling boards.

A PSU's wattage is the maximum continuous DC power it can supply, not what it draws from the wall all the time. Choose capacity for the components, then evaluate efficiency, cabling and physical fit.

Add the load before adding headroom

Use the CPU's maximum turbo or package-power limit, not its TDP. Add the graphics card's board power, about 50 W for motherboard and memory, roughly 5–10 W per drive, a few watts per fan and a budget for USB devices. Read the graphics-card maker's recommended PSU rating as well.

For a build with a 125 W maximum CPU, 220 W graphics card, 50 W board and memory, 15 W of drives, 10 W of fans, and 30 W for USB and margin, the sum is 450 W. A 650 W PSU operates at about 69% of its rating at that load. The remaining capacity gives room for short spikes and future changes. An undersized supply can cause shutdowns or restarts under heavy load, even if the machine starts normally.

Separate efficiency from capacity

Efficiency is DC output divided by AC input. At 90% efficiency, a PSU delivering 450 W draws:

It takes 500 W from the wall and releases 50 W as heat. At 85%, the same output needs about 529 W. An efficiency badge does not reduce a 650 W unit's rated DC output to 585 W.

The 80 PLUS program measures efficiency at specified loads. These are the internal 115 V minimums:

Titanium also requires 90% at 10% load. The 230 V program uses different, generally higher thresholds. Cybenetics also assesses efficiency and noise. Higher efficiency reduces wasted energy and heat; an efficiency rating alone does not establish every aspect of build quality. A greatly oversized PSU can cost more and work at a less efficient very low load.

Check cables and resilience

Fully modular PSUs detach all cables; semi-modular designs keep main cables fixed; non-modular units keep them all attached. Modularity makes cable routing easier, but PSU-side pinouts are not standardized. Use the cables supplied for that exact unit. A cable from another model can physically fit and still send the wrong voltage to a component.

ATX 3.x supplies are designed to tolerate short power spikes from modern graphics cards. That feature complements adequate continuous capacity and the right GPU connectors.

Servers may use redundant PSUs. Remaining modules carry the required load if one fails, allowing supported hot-swap replacement. Separate circuits or UPS units reduce the chance that one power fault disables every module. Two modules on one failed circuit do not provide protection against that circuit's loss.

A port's shape tells you what fits. Its specification tells you what the connection can carry. This distinction matters most with USB-C, where identical-looking ports can offer very different data speeds, video support and charging power.

Identify the plug, then the speed

USB Type-A is the familiar rectangular computer plug. Type-B is square and common on printers and scanners. Mini-B appears on older cameras, Micro-B on older mobile devices, and wide USB 3 Micro-B on some external drives. USB-C is small, oval, reversible and has 24 pins. A blue Type-A insert often indicates USB 3, but color is not a reliable specification.

USB 5Gbps was previously called USB 3.0, USB 3.1 Gen 1 and USB 3.2 Gen 1. USB 10Gbps includes the Gen 2 names, while USB 3.2 Gen 2x2 supplies 20 Gb/s over USB-C. New speed-based labels make these comparisons easier. A USB-C port can still be USB 2.0, and a charging cable may carry slow data and no video. Check both port and cable specifications.

Check the power path

Standard USB 2.0 supplies 500 mA at 5 V, or 2.5 W; USB 3.x supplies 900 mA, or 4.5 W. USB-C can provide 3 A at 5 V, or 15 W. USB Power Delivery (PD) negotiates more power using higher voltages. At 20 V and 3 A the result is 60 W. At 20 V and 5 A it is 100 W, requiring a 5 A cable. Extended Power Range reaches 48 V and 5 A, or 240 W, with an EPR-rated cable. The source, device and cable must all support the required power.

Recognize Thunderbolt and older connections

Thunderbolt carries PCIe data, DisplayPort video, USB and power. Thunderbolt 1 and 2 used Mini DisplayPort-shaped connectors; USB devices need a suitable dock or adapter. Thunderbolt 3 and 4 use USB-C at 40 Gb/s and accept USB devices directly. Thunderbolt 5 offers 80 Gb/s in both directions and up to 120 Gb/s one way for demanding display traffic. Look for the lightning-bolt symbol and verify the version. Devices can be daisy-chained.

Apple Lightning is an older small reversible 8-pin connector used on iPhones, iPads and accessories. iPhone moved to USB-C with iPhone 15 in 2023. RS-232 serial connections still serve console and industrial equipment. Their 9-pin connector is commonly called DB9, more precisely DE-9; a USB-to-serial adapter needs its driver.

Find the slowest link

An SSD rated at 1,050 MB/s attached to USB 2.0 cannot approach that rate. Dividing 480 Mb/s by eight gives 60 MB/s before overhead. A suitable USB 10Gbps port and cable remove that particular limit. Replacing the drive while leaving the same slow port would not solve the link bottleneck.

A video connection needs a compatible signal, direction and supported display mode. Before buying a cable, identify the computer's output, the monitor's input, and the resolution and refresh rate you want to use.

Compare digital and analog connections

HDMI and DisplayPort carry digital video and audio. HDMI is common on TVs and monitors, with full-size Type A, Mini Type C and Micro Type D connectors. HDMI 1.4 supports 4K at up to 30 Hz, HDMI 2.0 offers 18 Gb/s and 4K at 60 Hz, and HDMI 2.1 reaches up to 48 Gb/s for modes such as 4K at 120 Hz. HDMI 2.2 raises maximum bandwidth to 96 Gb/s with Ultra96 cables. Verify the actual port features and mode support, not only the version label. An Ultra High Speed cable serves 48 Gb/s modes; Ultra96 serves the new maximum.

DisplayPort is common on graphics cards and PC monitors. Its plug often has a latch, and Mini DisplayPort is a smaller form. DP 1.4 has a 32.4 Gb/s raw link rate; DP 2.1 reaches up to 80 Gb/s. Multi-Stream Transport can daisy-chain supported monitors. DP can also travel through USB-C using DisplayPort Alt Mode.

Older DVI-D carries digital video, DVI-A analog video, and DVI-I both. Single-link DVI reaches 1920 × 1200 at 60 Hz; higher modes such as 2560 × 1600 need dual-link support and a dual-link cable. Standard DVI does not carry audio. VGA is analog, with a 15-pin, three-row DE-15 connector. It remains useful for old displays, but image quality softens at high resolution or over long cables.

Decide whether conversion is needed

A passive adapter changes the connector wiring. An active converter changes the signal. Many DP outputs support dual-mode DP++, allowing passive HDMI or DVI connections. HDMI output to a DisplayPort input needs an active converter. Digital output to VGA needs active digital-to-analog conversion. HDMI and DVI-D share compatible video signaling, so a passive cable can link them, usually without audio. Direction matters: do not assume an adapter works in reverse.

USB-C native video requires DisplayPort Alt Mode or Thunderbolt. A data-only USB-C port cannot drive a simple HDMI adapter. DisplayLink docks instead encode video as USB data and require a driver.

Choose the mode across the entire path

A monitor can support different refresh rates on different inputs. Many QHD monitors reach 165 Hz over DisplayPort 1.2 or later but a lower rate over their HDMI input. Color depth also affects bandwidth. Read the supported-mode table for each input.

For a card with HDMI 2.1 and DP 1.4, connect a DP monitor directly with DP. For an older Full HD screen with DVI-D and VGA, HDMI-to-DVI-D preserves digital video without an active VGA converter. If there is no image, check selected monitor input, full cable seating and whether the cable is attached to the active graphics card output.

Network cables differ in distance, speed, interference protection and connector type. Start with the required link and its route, then select the cable. A connector that fits does not establish the cable's supported rate.

Choose copper for the distance

Twisted-pair Ethernet has four wire pairs and RJ45, technically 8P8C, connectors. The category appears on its jacket.

UTP is unshielded twisted pair, common in offices. STP adds shielding to reduce interference near motors and power wiring; it needs suitable shielded, grounded connections. Plenum-rated cable produces less smoke and toxic gas in a fire and is required where applicable building rules call for it in air-handling spaces. Direct-burial cable has a waterproof jacket for underground routes without conduit. A speed category and an environmental jacket rating answer different questions.

Recognize the wire orders

T568A and T568B define the order of eight contacts. Listed from pin 1, T568B is white-orange, orange, white-green, blue, white-blue, green, white-brown, brown. T568A is white-green, green, white-orange, blue, white-blue, orange, white-brown, brown. The orange and green pairs trade positions; blue and brown do not.

The same order at both ends makes a straight-through cable. A at one end and B at the other makes a crossover. Modern Auto-MDI-X ports normally handle crossover automatically. A narrower RJ11 telephone or DSL plug, usually with two or four contacts in six positions, is not an Ethernet connection even if it slips into the larger jack.

A punchdown block terminates individual wires in metal contacts, pressed in with a punchdown tool. Wiring closets use these connections for wall-jack, patch-panel and telephone cabling.

Recognize coax and fiber

RG-6 coaxial cable has a center conductor, insulation and a shield. A screw-on F-type connector is common for the wall-to-cable-modem connection.

Fiber carries light through glass and is immune to electrical interference. Single-mode uses a core about 9 µm wide and supports links of many kilometers. Multimode uses a 50 or 62.5 µm core and commonly serves shorter high-speed links of a few hundred meters; actual reach depends on fiber grade and optics. Jackets are often yellow for single-mode and aqua or orange for multimode. Verify labels rather than using color alone. Never look into a fiber end or port; its laser light can injure eyes.

Fiber connectors include round twist-lock ST, square push-pull SC and small latched LC. LC pairs are common on switch SFP modules.

Match the route

An 80 m, 10 Gb/s office run needs Cat 6a, because Cat 6's 10 Gb/s reach is shorter. If that route is an air-return plenum, it also needs the required plenum jacket. A 2 km interbuilding link calls for suitable single-mode fiber and matching transceivers, not a longer copper cable.

A display specification describes its light source, pixel grid, motion and color. Choose these for the work, then configure the connection so the screen actually runs at the intended mode.

Compare panel technologies

An LCD controls light from a backlight. TN panels offer fast response and low cost but narrower viewing angles and weaker color. IPS panels offer wide angles and accurate color, with lower contrast and possible glow in dark scenes. VA panels have high contrast and deep blacks, with some dark-motion smearing and viewing angles between TN and IPS.

OLED pixels emit their own light. They can switch completely off for black, giving very high contrast and fast response. Repeated static images such as a taskbar can cause burn-in over time; full-screen brightness is often below an LCD's. Mini-LED is still LCD technology, using many small backlight LEDs in local-dimming zones. It improves contrast and HDR but can create blooming around bright objects. It is different from microLED, where individual pixels emit light.

Separate resolution from physical size

Resolution is the pixel grid: Full HD is 1920 × 1080, QHD 2560 × 1440 and 4K UHD 3840 × 2160. Full HD has 2,073,600 pixels; 4K has 8,294,400, four times as many. Native resolution gives the sharpest text. Scaling increases the size of interface elements while keeping that native pixel grid.

The aspect ratio describes width relative to height, commonly 16:9, with 16:10 and 21:9 alternatives. Pixel density combines resolution and physical size. Industry screen diagonals are stated in inches, with one inch equal to 2.54 cm. A 27-inch screen has a 68.6 cm diagonal. For a QHD screen at that size:

A 24-inch Full HD screen is about 92 PPI; a 27-inch 4K screen is about 163 PPI. The latter may need scaling, such as 150%, for comfortable text.

Configure motion and color

Refresh rate counts redraws each second in hertz. 60 Hz suits office work; 120, 144, 165 or 240 Hz can show smoother motion. The GPU, port and cable must support the mode, and the operating system must select it. A 144 Hz monitor left at 60 Hz does not run at its advertised rate. Variable refresh technologies such as FreeSync, G-SYNC and HDMI VRR match display timing to frame delivery to reduce tearing.

Color gamut describes the range a display can show: sRGB is common for web and office work, DCI-P3 for video/HDR, and Adobe RGB for print. A wide gamut does not guarantee accurate color; calibration still matters. Brightness uses nits, equivalent to cd/m².

Add touch input

A digitizer turns contact into coordinates. Capacitive touch detects fingers and supports multitouch; resistive touch responds to pressure, including gloves or a basic stylus. Active pens use a suitable pen digitizer. HDMI and DisplayPort video do not carry the USB touch-data path: an external touch monitor needs an extra USB cable. A suitable USB-C connection can carry video and touch together. Recalibrate when touches consistently land in the wrong place.

A peripheral can need a data connection, power, a driver and permission to use a feature. Check each need during setup. A device receiving power can still lack the data connection or software it needs to work.

Identify the device's role and connection

Keyboards, mice, scanners, barcode readers, webcams, microphones and drawing tablets provide input. Monitors, speakers and headphones provide output. Headsets, touch screens and external drives can do both.

USB is the most common wired connection. Bluetooth devices must be paired, typically within about 10 m. Other wireless devices use their own paired 2.4 GHz USB receiver. A receiver for one device is not necessarily a general Bluetooth adapter.

For a wireless device that stops responding, check charging or batteries, pairing and interference. USB 3 ports and cables can disturb a nearby 2.4 GHz receiver. A short USB extension or a USB 2.0 port can move the receiver away from the interference source.

Supply the driver and permissions

Plug and play lets the operating system recognize a device and load a suitable driver. Standard keyboards, mice, flash drives, headsets and most webcams commonly use built-in drivers. Drawing tablets, capture devices, some scanners and docks need the maker's software. In Windows, Device Manager uses a yellow warning symbol for missing or failed drivers.

A working microphone or webcam may still be blocked by privacy settings. Check the selected device in the application and the operating system's permission controls before replacing the hardware. A Bluetooth device may also be reconnecting to a nearby phone or another computer.

Connect audio correctly

Desktop 3.5 mm jacks commonly use green for speakers/headphones, pink for microphone input and light blue for line input. A headset with one four-pole TRRS plug combines headphones and microphone. A PC with separate input and output jacks needs a suitable splitter. USB headsets contain their own sound hardware.

Expand the available ports

A USB hub adds ports but does not create unlimited power or bandwidth. A bus-powered hub shares its upstream port's power among the devices. If several bus-powered drives need more than that, use a powered hub with its own adapter. Drives with their own adapters do not take their operating power from the hub. A monitor's built-in USB hub also needs its upstream data cable connected; an HDMI cable alone does not connect that hub.

A dock adds functions such as video outputs, USB, Ethernet and charging through one link. Native video requires DisplayPort Alt Mode or Thunderbolt. Charging requires enough Power Delivery wattage, and DisplayLink video needs its driver. Check the dock's delivered charging power, not only its adapter rating.

For a shared desk with two monitors, keyboard, mouse and Ethernet, a compatible Thunderbolt dock can consolidate connections. If the computer needs 65 W charging, the dock must deliver at least that amount to it. A KVM switch has a different purpose: one keyboard, video display and mouse can control two or more computers by switching between them.

A useful parts list starts with the user's work. Gaming puts a large share of the budget into graphics. Video editing benefits from CPU cores, RAM and fast storage. Office work often needs a quiet, reliable system with integrated graphics. Plan the whole system before ordering individual bargains.

Check the pairs that must match

For each part, record the model and the specification that affects its neighbors:

• CPU and board: exact socket, chipset support and required firmware.
• Memory and platform: DDR generation, DIMM or SODIMM, capacity limits and tested kits.
• Board and case: supported form factor and mounting points.
• GPU and case: card length, thickness and available expansion positions.
• GPU and PSU: recommended wattage and correct power connectors.
• Cooler and CPU: mounting kit and suitability for the exact CPU.
• Cooler and case: height, radiator positions and memory clearance.
• Storage and board: interface, M.2 key and length, and lane sharing.
• PSU and case: ATX or SFX format, depth and cable space.

Online compatibility tools are helpful, but the board manual, CPU support list, QVL and component makers' dimensions settle the final checks. A matching socket does not establish firmware support; an M.2 connector does not establish protocol support.

Allow for parts that occupy the same space

Compare dimensions in millimeters. If a case allows a 330 mm graphics card and the card is 320 mm, the nominal clearance is 10 mm. A front radiator and fans can consume that space. Check the maximum GPU length for the actual radiator layout, not the case with no radiator installed.

Cooler height works the same way. A 158 mm cooler fits a 160 mm limit on paper but leaves little margin. Check the cooler's memory clearance and required brackets as well. A card can fit in length but block an additional expansion card because of its thickness.

Audit a proposed build

A list contains a microATX AM5 board with DDR5 slots, a microATX case supporting an ATX PSU, an AM5 CPU, a compatible 158 mm cooler, a 320 mm GPU, a DDR4-3600 kit and a 550 W PSU. The case limits are 160 mm cooler height and 330 mm GPU length. The card requires a 650 W recommended PSU and one 12V-2x6 connection.

The cooler and card fit the stated clearances. The DDR4 kit cannot fit the DDR5 board, so replace it with suitable DDR5. The PSU is below the card maker's recommendation and lacks its connection; select at least a suitable 650 W unit with a compatible cable arrangement. Check remaining parts and firmware before treating the list as complete.

Include the complete working setup

Budget for an operating-system license, display, keyboard, mouse and any missing Wi-Fi capability. A no-card build needs a CPU with integrated graphics and usable board outputs. Check the intended operating system's hardware requirements too: Windows 11 needs TPM 2.0, Secure Boot-capable UEFI and a supported CPU. A physically compatible set of parts can still fail the user's software or peripheral requirements.

A planned assembly order reduces awkward handling and makes the first startup easier to check. Keep the board and case manuals open, lay out the screws, and identify connectors before parts make them hard to reach.

Prepare the work area

Use a Phillips #2 screwdriver, a flashlight and cable ties. Shut down, switch off the PSU and unplug the power cord. Use an antistatic wrist strap attached to bare case metal and an antistatic mat when available. Keep parts in antistatic packaging until needed, hold boards by their edges and avoid carpet. Keep drinks away and lift heavy cases carefully.

Electrostatic discharge (ESD) is a sudden transfer of static charge that can damage components. Good handling also prevents physical damage: do not touch socket pins, drag cards across the board or rest a board on a loose screw.

Build in a practical order

1. Rest the board on its box. Fit the CPU, RAM and M.2 drives. Fit the cooler now if it will not obstruct mounting.
2. Check case standoffs against the board's holes. Fit a separate I/O shield if required, then secure the board.
3. Fit the PSU with its fan toward a suitable filtered vent. Install the cooler if not already fitted and connect cooling headers.
4. Mount SATA drives, then seat and secure the graphics card in the main x16 slot.
5. Attach main ATX, CPU and GPU power, SATA cables and case headers. Route cables clear of fans and airflow.

Only place standoffs under mounting holes. An extra post touching the underside can short the board. Follow the front-panel pin diagram. LEDs need correct polarity; a reversed LED normally stays dark. Power and reset switches work either way around.

Complete POST before closing the case

Connect a monitor to the active GPU output, or the board output if using supported integrated graphics. Power on and watch the board's indicators. POST checks the processor, memory and video hardware before booting. Beep patterns and CPU/DRAM/VGA/BOOT debug LEDs vary by maker, so read the board's definitions.

New RAM, especially DDR5, can train for a minute or more with a black screen. Allow the expected training time before interrupting startup. Once setup opens, confirm the CPU model, full RAM capacity and every drive. Check fan operation and CPU temperature against that processor's normal behavior.

Enable a supported XMP or EXPO profile after the basic hardware check. Select the operating-system USB installer with the boot menu or boot order when ready to install. A BOOT indicator with no installed operating system can mean the machine needs a bootable device, not that its CPU or RAM is faulty.

If the button does nothing

Check the PSU switch and wall cord, then confirm the 24-pin and CPU power connections and the case power-switch header. Inspect and reseat internal connections only after switching off and unplugging again. A wrong front-panel pin pair can leave an otherwise sound build completely unresponsive.

Firmware setup controls how the computer starts and exposes hardware features to the operating system. Enter it with the maker's startup key, commonly Delete, F2 or F10, or through Windows advanced startup. A one-time boot menu often uses F11 or F12.

Choose boot and security settings

Older BIOS has largely been replaced by UEFI, which supports GPT boot disks, large boot drives and Secure Boot. Many UEFI setups are graphical, but a text screen can still be UEFI. People commonly call either setup "the BIOS."

Boot order sets which device is tried first. A one-time boot menu selects a device without permanently changing that order. Use it for an installer, then boot normally from the internal drive. Legacy compatibility mode, or CSM, must be off for Secure Boot and Windows 11. Fast boot can skip USB initialization and make startup keys unresponsive; use the operating system's firmware-entry option instead.

Secure Boot checks boot loaders against trusted signing keys. This helps stop untrusted code before the operating system starts. It does not encrypt files. A TPM stores or protects keys and records boot measurements. It can be a discrete chip or firmware feature, such as Intel PTT or AMD fTPM. Windows 11 requires TPM 2.0 and Secure Boot-capable firmware. BitLocker and Windows Hello use the TPM.

Save the BitLocker recovery key before clearing a TPM or changing the board, because a changed trust state can require recovery. A hardware security module (HSM) is a separate dedicated key-handling device used by organizations, such as a server card or network appliance for certificates or payments.

Distinguish access controls

A supervisor or administrator password protects setup settings. A user or power-on password controls startup. Clearing CMOS resets forgotten passwords on many desktop boards, but this is not a general solution for business laptops.

USB settings have separate effects. Disabling USB boot prevents starting an operating system from a flash drive. It does not stop Windows from using that flash drive afterward. To restrict removable copying, use the appropriate USB-storage or port controls.

Enable features and monitor cooling

Virtualization uses Intel VT-x or AMD-V, often called SVM Mode in AMD firmware. Enable it for supported hypervisors and features such as WSL 2. A virtualization error can come from this setting being disabled, not from too few CPU cores.

The hardware monitor shows temperatures, voltages and fan speeds. Fan curves choose speed according to temperature. A CPU_FAN error can occur if an AIO uses another header and nothing reports speed there. Correct the connections or supported monitoring setting rather than ignoring a stopped fan.

Update and restore deliberately

Firmware updates fix bugs, address security issues and support new CPUs. Use the exact board model's file and update tool, and maintain uninterrupted power. Some boards support CPU-less USB flashback. Update before replacing the old CPU when the new one needs newer firmware.

Updates may reset settings; Clear CMOS restores defaults. Recheck boot order, memory profile, fans and security settings afterward. A failing coin cell commonly causes a lost clock after unplugging. A desktop left connected can keep its clock from standby power, hiding the battery problem until disconnection.

An upgrade should remove a measured limit and fit the existing platform. Identify the current parts before shopping. A faster replacement that the board cannot use is not a useful upgrade.

Find the constraint during the slow task

Record board model, CPU, RAM capacity and slots, storage connectors, PSU label and case clearances. Task Manager and System Information reveal many installed parts; the manual and physical inspection supply limits and space.

Watch performance while the user does the work that feels slow. RAM constantly near full suggests more memory. A hard drive at 100% active time suggests a storage bottleneck, especially with paging or many small reads. A saturated CPU or GPU points to that part's workload. High utilization is a clue to investigate alongside the running processes, temperature and clock behavior, because malware, background work or throttling can also explain slowness.

Choose the upgrade that meets the need

Replacing an HDD with an SSD often makes an older PC much more responsive. Clone the existing system with migration software or install the OS fresh. A clone needs enough destination space for the data and a migration tool that supports the partition layout. An NVMe upgrade needs a compatible M.2 slot or a suitable PCIe adapter plus firmware boot support if it will be the boot drive.

For RAM, check generation, maximum capacity and channel layout. A machine with one 8 GB DDR4-2666 module and four slots can reach 32 GB with a supported 2 × 16 GB kit in the recommended slots. Keep the matched kit at its supported settings. Adding the old module changes the channel balance and can lower speed or stability; mixed capacity does not necessarily disable all dual-channel operation.

For a graphics card, check length, thickness, free slots, PSU recommendation and power connectors. Follow the driver maker's replacement procedure; remove an incompatible old driver and install the proper new driver. An otherwise suitable GPU can be blocked by a front radiator or an inadequate PSU.

Prepare a CPU change before removing the old CPU

Use the board's CPU support list. If the replacement needs newer firmware, update while the old supported CPU still runs. Then change the CPU and renew the thermal paste. When the desired CPU requires another socket or memory generation, budget for a platform replacement: board, processor and often RAM together.

Compare that total with replacing the PC. Check current OS support as well as speed; missing TPM 2.0 or an unsupported CPU can prevent a supported Windows 11 installation. Buying several parts for an aging platform may cost more than a suitable replacement system.

Protect data and handle replaced parts

Back up important data before migration or hardware changes. Before a drive leaves your control, sanitize it using a method suitable for that media and verify completion. Deleting files or quick-formatting is insufficient. SSDs can need their supported sanitize or cryptographic-erase method because ordinary overwriting may miss reserved flash areas. If sanitization cannot be completed and verified, arrange appropriate physical destruction. Take electronics and batteries to approved collection services under local rules.