Quick Answer

A GPU, or Graphics Processing Unit, is a specialized processor designed to perform thousands of calculations simultaneously. It was originally developed for drawing computer graphics, but modern GPUs also accelerate artificial intelligence, video processing, engineering simulations, scientific computing and cryptocurrency-related workloads.

A GPU product may pass through several different companies:

A GPU company such as NVIDIA, AMD or Intel designs the processor.
A semiconductor foundry such as TSMC, Samsung Foundry or Intel Foundry manufactures the silicon chip.
Memory manufacturers such as Samsung, SK hynix and Micron manufacture GDDR or HBM memory.
Packaging companies connect the GPU die, memory and substrate.
Graphics-card manufacturers such as ASUS, MSI, Gigabyte, Sapphire and Zotac assemble complete add-in cards.
The finished chip and card undergo electrical, functional, thermal, performance and reliability testing.

Therefore, there is no single accurate number of “GPU manufacturers.” The answer depends on whether we count GPU designers, chip foundries, integrated GPU designers, graphics-card assemblers or companies making specialist AI accelerators.

What Is a GPU? Complete Guide to GPU Technology, Manufacturing, Components and Testing

What Does GPU Mean?

GPU stands for Graphics Processing Unit.

It is an electronic processor optimized for processing many similar operations in parallel. A conventional Central Processing Unit, or CPU, is designed to handle a smaller number of complex, sequential tasks efficiently. A GPU contains a much larger collection of smaller computing resources that can work on many data elements simultaneously.

For example, displaying a three-dimensional game scene requires calculations for millions of pixels, vertices, textures, shadows and lighting effects. Because many of these calculations are similar, a GPU can divide them among numerous processing units and execute them in parallel.

Modern GPUs are used for much more than gaming:

Windows and desktop display rendering
Gaming and 3D graphics
Video editing and encoding
CAD, CAM and architectural applications
Animation and visual effects
Artificial intelligence training and inference
Scientific simulations
Medical imaging
Engineering calculations
Data analytics
Cloud computing
Virtual desktops
High-performance computing

GPU, Graphics Card and Video Card: Are They the Same?

These terms are related, but they do not mean exactly the same thing.

Term	Meaning
GPU	The processing chip that performs graphics or parallel calculations
Graphics card	A complete circuit board containing a GPU, memory, power circuitry, cooling and display connectors
Video card	Another commonly used name for a graphics card
VRAM	High-speed memory used by the GPU
Integrated GPU	Graphics processor built into or packaged with a CPU or system-on-chip
Discrete GPU	A separate graphics processor with dedicated resources, usually including its own memory
External GPU	A discrete graphics card placed in an external enclosure and connected to a computer

Calling an entire graphics card a “GPU” is common in everyday conversation, but technically the GPU is only its central processing device.

CPU vs GPU

A CPU and GPU are both programmable processors, but their architectures and intended workloads differ.

Feature	CPU	GPU
Primary strength	Complex sequential processing	Highly parallel processing
Core design	Fewer powerful cores	Many smaller execution units
Latency	Optimized for fast response to individual tasks	Optimized for high total throughput
Cache	Large and sophisticated per-core cache hierarchy	Designed to feed many parallel operations
Common work	Operating systems, applications, databases	Graphics, AI, video and simulations
Memory	Uses system RAM	Usually uses dedicated VRAM or shared unified memory
Branching	Handles irregular logic effectively	Performs best with similar parallel operations

A GPU is not universally faster than a CPU. It is faster only when the workload can be divided into enough parallel operations and suitable software is available.

Why Was the GPU Invented?

Early computers used the CPU to perform most display calculations. As graphical user interfaces, high-resolution displays and 3D games became more advanced, graphics workloads became too demanding for general-purpose processors.

Dedicated graphics hardware evolved through several stages:

Basic display controllers generated text and simple images.
Two-dimensional accelerators handled lines, shapes and window operations.
Fixed-function 3D accelerators processed triangles, textures and lighting.
Programmable shaders allowed developers to control graphics calculations.
Unified shader architectures used flexible processors for multiple graphics stages.
General-purpose GPU computing enabled scientific and business applications.
Modern GPUs added hardware for ray tracing, artificial intelligence and media processing.

How Does a GPU Work?

A GPU receives instructions and data from the CPU or another host processor. It divides the work into many smaller operations and distributes them across groups of parallel execution units.

A simplified graphics pipeline works as follows:

The application and CPU prepare the scene.
Geometry data describing objects is sent to the GPU.
Vertex processing calculates object positions.
Primitive assembly creates triangles or other geometric shapes.
Rasterization determines which pixels are covered by those shapes.
Pixel or fragment processing calculates colour, lighting and textures.
Depth, blending and other output operations combine the results.
The completed frame is written to a frame buffer.
The display engine sends the image to the monitor.

Modern GPUs may alter this pipeline using mesh shaders, compute shaders, ray tracing and AI-assisted rendering.

Parallel Processing Inside a GPU

A modern GPU does not simply contain thousands of completely independent CPU-style cores. Marketing terms such as CUDA cores, stream processors, execution units and shader processors describe arithmetic resources organized into larger processing blocks.

Depending on the manufacturer, these groups may be called:

Streaming Multiprocessors by NVIDIA
Compute Units or Work Group Processors by AMD
Xe-cores or subslices by Intel

A workload is divided into many threads. The GPU schedules groups of related threads together so that they can perform similar instructions on different data.

This model is highly efficient for operations such as:

Applying the same colour calculation to many pixels
Multiplying large matrices
Processing thousands of image elements
Simulating particles
Evaluating AI neural-network layers

Performance can fall when nearby threads require substantially different instructions, because the hardware may have to execute different paths separately.

Important Technologies Inside a Modern GPU

Shader or General-Purpose Processing Units

These programmable arithmetic units execute graphics shaders and compute instructions. They handle floating-point, integer and logical operations.

Different companies use different names, so core counts should not be compared directly across unrelated architectures.

Texture Mapping Units

Texture Mapping Units, or TMUs, retrieve, filter and map texture data onto 3D surfaces. They help calculate how images and material details appear on rendered objects.

Raster Operations Units

Raster Operations Units, frequently called ROPs, help perform final pixel operations such as blending, colour output, anti-aliasing and writing data to the frame buffer.

Ray-Tracing Hardware

Ray tracing simulates the movement and interaction of light rays. Dedicated ray-tracing hardware accelerates calculations involving:

Ray and triangle intersections
Reflections
Shadows
Global illumination
Transparency
Bounding-volume hierarchy traversal

Real-time games frequently combine traditional rasterization with selected ray-traced effects.

AI or Matrix Accelerators

Modern GPUs may contain specialized matrix-processing engines. NVIDIA calls some of these Tensor Cores, while AMD and Intel use their own architectural terms.

They accelerate operations used in:

AI model training
AI inference
Image reconstruction
Upscaling
Frame generation
Denoising
Matrix multiplication

Cache Memory

GPUs use several cache levels to reduce repeated access to slower external memory. The exact arrangement depends on the architecture, but may include:

Instruction cache
Data cache
Texture cache
Per-processing-block cache
Shared last-level cache

Cache capacity alone does not determine performance. Cache design, hit rate, workload and memory bandwidth are also important.

Memory Controllers

Memory controllers transfer data between the GPU and its video memory. Their design affects memory bandwidth, power consumption and supported memory type.

Display Engine

The display engine generates output for interfaces such as:

HDMI
DisplayPort
Embedded DisplayPort
USB-C DisplayPort Alternate Mode

Supported resolutions, refresh rates and features depend on the specific GPU, graphics card, connectors, firmware, drivers and monitor.

Video Encode and Decode Engines

Dedicated media engines accelerate formats such as H.264, H.265/HEVC and AV1. Exact codec and quality support varies by GPU generation.

These engines reduce the need to use general shader resources for video playback, streaming and transcoding.

PCI Express Interface

A discrete graphics card normally communicates with the computer through PCI Express. Workstation and data-centre accelerators may also use high-speed proprietary or industry-standard interconnects between accelerators.

Power-Management Circuitry

The GPU chip contains sensors and control logic for:

Voltage
Clock frequency
Temperature
Power consumption
Idle states
Workload-based boosting
Thermal protection

Dynamic voltage and frequency scaling adjusts performance and power use according to workload and operating conditions.

What Components Are Used in a GPU Chip?

The silicon GPU die contains enormous numbers of microscopic electronic structures.

Important elements include:

Transistors
Logic gates
Arithmetic and logic units
Register files
Schedulers
Load/store units
Cache memory
Memory controllers
Clock-distribution circuits
Power-management logic
Media engines
Display controllers
Input/output interfaces
On-chip interconnects
Security and virtualization logic
Test and diagnostic circuitry

Individual transistors are connected through multiple microscopic metal-interconnect layers. These layers form the data paths and circuits needed to operate the processor.

What Components Are Used in a Graphics Card?

A complete discrete graphics card contains much more than the GPU chip.

Component	Purpose
GPU package	Performs graphics and compute processing
VRAM chips	Store textures, frame buffers, models and working data
PCB	Electrically connects all card components
VRM	Converts input power into stable lower voltages
MOSFETs or power stages	Switch and regulate power
Inductors or chokes	Smooth the electrical current
Capacitors	Filter and stabilize voltage
VBIOS or firmware chip	Stores card initialization and operating information
PCIe connector	Connects the card to the motherboard
Auxiliary power connectors	Supply additional power from the PSU
Display connectors	Provide output to monitors
Heatsink	Transfers heat away from components
Heat pipes or vapour chamber	Spread heat across the cooling assembly
Fans	Move air through the heatsink
Thermal paste	Improves heat transfer between the GPU and cooler
Thermal pads	Transfer heat from memory and power components
Backplate	Provides protection, rigidity and sometimes heat spreading
Sensors	Monitor temperature, current, voltage and fan operation

Not every graphics card contains every component. Laptop GPUs, workstation modules and data-centre accelerators use different packaging and cooling arrangements.

What Is VRAM?

VRAM is the memory available to the GPU. A discrete graphics card commonly uses GDDR memory, while high-end AI and high-performance computing accelerators may use HBM.

GDDR Memory

Graphics Double Data Rate memory is mounted around the GPU package on the graphics card’s PCB.

Common generations include:

GDDR5
GDDR5X
GDDR6
GDDR6X
GDDR7

Micron describes GDDR7 as graphics memory intended for bandwidth-intensive workloads including graphics, gaming and AI. Actual speed and capacity depend on the selected memory devices and GPU memory controller. Micron GDDR7 product information

HBM

High Bandwidth Memory uses vertically stacked memory dies placed very close to the processor through advanced packaging. It offers exceptionally high bandwidth and is widely used in premium AI and data-centre accelerators.

TSMC’s CoWoS platform, for example, can integrate large processors with multiple HBM stacks through an interposer. TSMC CoWoS technology

Does More VRAM Always Mean a Faster GPU?

No. VRAM capacity is only one specification.

Performance also depends on:

GPU architecture
Number and capability of processing units
Memory type
Memory speed
Memory-bus width
Cache system
Clock frequency
Power limit
Cooling
Drivers
Application optimization

Insufficient VRAM can cause stuttering, reduced texture quality or out-of-memory errors. However, adding capacity does not make a weak processor automatically faster.

How Is a GPU Designed?

GPU design begins years before mass production.

1. Product Requirements

Engineers define targets for:

Performance
Power consumption
Die size
Memory capacity and bandwidth
Display features
Video engines
AI capabilities
Manufacturing cost
Product markets
Reliability

2. Architecture Design

Architects determine how processing units, caches, interconnects, memory controllers and specialized engines will work together.

3. Logic and Circuit Design

Hardware-description languages are used to define digital logic. Analog circuits, high-speed interfaces and power systems require specialized engineering.

4. Simulation and Verification

Before manufacturing, teams simulate and formally verify the design. Verification attempts to identify:

Incorrect calculations
Timing problems
Protocol errors
Memory-coherency issues
Deadlocks
Security weaknesses
Power problems
Rare instruction combinations

Verification is one of the largest and most expensive parts of advanced chip development.

5. Physical Design

Electronic Design Automation software converts logical circuits into a physical arrangement of:

Standard cells
Memories
Processing blocks
Interconnects
Clock trees
Power-delivery networks
Input/output structures

Engineers check signal timing, voltage drop, electrical interference, heat density and manufacturability.

6. Tape-Out

When the design has passed its pre-production checks, the final layout data is sent for mask preparation and wafer fabrication. This milestone is traditionally called tape-out.

Tape-out does not mean the product is ready. Initial silicon may reveal problems that require firmware, driver, manufacturing or even design changes.

What Machines Are Required to Manufacture GPU Chips?

An advanced GPU fabrication facility costs billions of dollars and contains hundreds of sophisticated systems. No single machine manufactures a GPU from start to finish.

Main Categories of Semiconductor Equipment

Machine or system	Purpose
Crystal-growth equipment	Produces high-purity silicon crystal ingots
Wafer slicing and polishing equipment	Produces extremely flat silicon wafers
Oxidation and diffusion furnaces	Grow or introduce materials into wafer layers
Deposition equipment	Adds thin films of conducting or insulating material
Photoresist coaters and developers	Apply and process light-sensitive material
DUV/EUV lithography scanners	Project circuit patterns onto the wafer
Plasma etching equipment	Removes selected material
Ion implantation systems	Introduce controlled impurities into silicon
Annealing equipment	Repairs and activates implanted material
Chemical-mechanical planarization equipment	Flattens wafer surfaces
Wet-cleaning systems	Remove particles and chemical residue
Metrology equipment	Measures dimensions, alignment and film thickness
Inspection equipment	Detects particles and pattern defects
Wafer probers	Electrically test chips before cutting
Wafer grinders	Reduce wafer thickness when required
Dicing saws or laser systems	Separate the wafer into individual dies
Die-bonding equipment	Places a die onto a substrate or package
Flip-chip bonding equipment	Connects microscopic solder bumps
Packaging and underfill systems	Protect and mechanically support the chip
X-ray inspection	Examines hidden package connections
Automated test equipment	Performs electrical and functional testing
Burn-in equipment	Operates devices under elevated stress
PCB assembly equipment	Builds complete graphics cards
Reflow ovens	Solder components to the PCB
Optical inspection equipment	Checks PCB placement and soldering

ASML explains that chip manufacturing involves hundreds of steps, can take months and builds many aligned layers on a silicon wafer. The process occurs in tightly controlled cleanrooms, where contamination can destroy microscopic circuit structures. ASML: How microchips are made

How Is a GPU Chip Manufactured?

Step 1: Produce the Silicon Wafer

Highly purified silicon is melted and grown into a large single-crystal ingot. The ingot is sliced into thin circular wafers, which are polished to create an extremely flat surface.

Modern high-volume logic fabs commonly process 300 mm wafers, although other wafer sizes remain in use for different semiconductor products.

Step 2: Deposit Material

Machines deposit extremely thin layers of insulating, semiconducting or conducting materials. Processes may include:

Chemical Vapour Deposition
Physical Vapour Deposition
Atomic Layer Deposition
Epitaxial growth

Step 3: Apply Photoresist

A light-sensitive chemical called photoresist is applied to the spinning wafer, producing a thin and uniform coating.

Step 4: Lithography

A lithography system projects a circuit pattern onto the photoresist. Advanced processes use combinations of DUV and EUV lithography, depending on the layer and required feature dimensions.

A process-node name such as “3 nm” or “5 nm” should not be interpreted as the exact measurement of every transistor feature. Modern node names primarily identify a technology generation.

Step 5: Develop and Etch

The exposed photoresist is developed, leaving selected areas protected. Etching equipment removes material from unprotected regions to reproduce the pattern in the underlying layer.

Step 6: Ion Implantation

Ion implantation introduces controlled impurities into silicon to modify its electrical characteristics and create transistor regions.

Step 7: Remove, Clean and Repeat

Remaining photoresist and residue are removed. Deposition, lithography, etching, implantation, polishing and cleaning are repeated many times to create the transistor and interconnect layers.

Step 8: Wafer Inspection and Testing

Inspection tools search for contamination and pattern defects. Metrology systems measure line dimensions, film thickness, overlay accuracy and other parameters.

Electrical wafer probing identifies dies that meet basic specifications and those that are defective.

Step 9: Wafer Dicing

The completed wafer is cut into individual pieces called dies. A large GPU die occupies more wafer area than a small chip, so fewer copies fit on each wafer and defects can have a greater cost impact.

Step 10: Packaging

A usable GPU requires a protective and electrically connected package. Depending on the product, packaging may involve:

Organic substrates
Flip-chip connections
Solder bumps
Interposers
Chiplets
Silicon bridges
HBM stacks
Heat spreaders
Underfill materials

Advanced AI accelerators may integrate multiple logic dies and memory stacks in a single package.

Step 11: Final Testing and Classification

Packaged chips are tested again. Functional devices may be classified according to:

Stable operating frequency
Voltage requirements
Power consumption
Working compute units
Memory-interface functionality
Thermal characteristics

This classification process is commonly known as binning.

Step 12: Graphics-Card Assembly

For a consumer graphics card, machines place the GPU package, memory chips and electronic components onto a multilayer PCB. The board passes through solder-paste printing, component placement and a reflow oven.

The manufacturer then installs the cooler, fans, thermal pads, heat spreader and exterior enclosure.

Why Are Some GPU Models Based on the Same Chip?

Not every die produced from a wafer operates identically. One die may support every processing block, while another may contain a small defective region.

Manufacturers can sometimes disable defective or unnecessary sections and sell the remaining functional chip as a lower-tier product. This improves manufacturing yield and reduces waste.

Product segmentation may also intentionally vary:

Enabled processing units
Clock frequency
Memory capacity
Memory-bus width
Power limit
Cache configuration
Media or display features

A lower model is therefore not always made from a physically smaller design.

What Raw Materials Are Used?

GPU production and graphics-card manufacturing may involve:

Electronic-grade silicon
Copper
Aluminium
Cobalt
Tungsten
Tantalum
Tin
Nickel
Gold
Silver
Palladium
Fibreglass
Epoxy resin
Ceramic materials
Solder alloys
Polyimide
Photoresist chemicals
Process gases
Cooling metals
Thermal interface compounds
Plastics

Semiconductor manufacturing also consumes considerable quantities of ultra-pure water, electricity and specialty chemicals.

How Many Companies Manufacture GPUs?

There is no authoritative fixed total because “manufacture” can refer to several different activities. Companies enter and leave markets, products may be regional, and many organizations design graphics logic without selling a separate PC graphics card.

Major PC Discrete-GPU Designers

For mainstream Windows desktop and laptop discrete graphics, the three major designers are:

NVIDIA
AMD
Intel

These companies design and sell GPU products, but this does not necessarily mean that they fabricate every silicon die in their own factories.

Other Companies Designing GPU Technology

The broader GPU and graphics-IP industry includes companies such as:

Apple
ARM
Qualcomm
Imagination Technologies
Samsung
MediaTek
Broadcom
VeriSilicon
Huawei
Biren Technology
Moore Threads
MetaX
Zhaoxin
Jingjia Micro

Some create GPUs integrated into mobile processors or system-on-chips. Some license GPU intellectual property. Others develop accelerators for regional, embedded or specialist markets.

Because “GPU” overlaps with graphics IP, AI accelerators, console processors and integrated system-on-chip designs, claiming an exact global count would be misleading. A defensible description is:

Three major companies currently design mainstream PC discrete GPUs.
Several more design integrated mobile, console or embedded GPUs.
Dozens of companies participate if licensed GPU IP and specialist accelerators are included.

This count can change and should be dated whenever it is published.

Semiconductor Foundries

Foundries physically fabricate chips designed by other companies. Important foundry businesses include:

TSMC
Samsung Foundry
Intel Foundry
GlobalFoundries
UMC
SMIC

Not all foundries manufacture the latest high-performance GPUs. A foundry must offer a suitable process, capacity, packaging ecosystem, yield, performance and commercial agreement.

ASML distinguishes between integrated device manufacturers such as Intel and Samsung, foundries such as TSMC, and fabless chip designers such as NVIDIA and AMD. ASML’s explanation of chipmaker types

Graphics-Card Manufacturers

Companies assembling complete add-in graphics cards include, depending on brand authorization, product generation and region:

ASUS
MSI
Gigabyte
Zotac
Palit
Gainward
PNY
Inno3D
Colorful
Galax
Sapphire
PowerColor
XFX
ASRock
Acer
Biostar
Sparkle

These companies normally do not design the underlying GPU architecture or fabricate its silicon. They develop or manufacture PCBs, power systems, cooling systems, firmware configurations and finished cards using GPU chips supplied under commercial agreements.

Memory Manufacturers

Major companies involved in graphics and high-bandwidth memory include:

Samsung Electronics
SK hynix
Micron Technology

A GPU designer may qualify memory from one or more suppliers for a particular product.

Packaging and Testing Companies

Semiconductor packaging and testing may be performed by the foundry, GPU supplier or an outsourced semiconductor assembly and test company. Examples of large OSAT providers include:

ASE Technology
Amkor Technology
JCET Group
Powertech Technology
Tongfu Microelectronics

The exact supply chain for a particular GPU may be confidential or may change between products.

How Are GPU Chips Tested?

Testing occurs throughout design, wafer fabrication, packaging and final system production.

Pre-Silicon Verification

Before the physical chip exists, engineers test the design using:

Software simulation
Hardware emulation
Formal verification
Static timing analysis
Power analysis
Signal-integrity analysis
Design-rule checking
Layout-versus-schematic checking
Fault simulation

Pre-silicon verification cannot guarantee that the first manufactured chip will be perfect, but it greatly reduces risk.

Process Monitoring

Test structures on the wafer help engineers monitor:

Transistor behaviour
Resistance
Leakage
Layer thickness
Critical dimensions
Overlay alignment
Defect density
Process uniformity

Wafer-Probe Testing

Before the wafer is cut, a probe card makes electrical contact with each die. Automated Test Equipment can check:

Power connections
Short circuits
Leakage current
Basic logic
Embedded memories
Interfaces
Built-in self-test results
Operating voltage
Preliminary speed characteristics

A wafer map records which dies passed, failed or require further evaluation.

Package Testing

After packaging, the chip is tested again because dicing, bonding and packaging can introduce defects.

Tests may include:

Functional instruction tests
Clock and voltage tests
Memory-interface tests
PCIe link tests
Display-engine tests
Media-engine tests
Compute tests
Thermal-sensor tests
Power-state tests

Built-In Self-Test

Modern chips contain Design-for-Test circuitry, including techniques such as:

Scan chains
Memory Built-In Self-Test
Logic Built-In Self-Test
Boundary scan
Error-detection circuits
On-chip sensors

These features improve manufacturing fault detection and later diagnostics.

Burn-In and Stress Testing

Selected devices may be operated at elevated temperature, voltage or workload to expose early-life failures.

Stress testing must remain within a controlled qualification procedure. Applying excessive voltage or temperature outside approved conditions can damage the device and does not represent responsible testing.

Reliability Qualification

Qualification programs may assess resistance to:

High operating temperature
Temperature cycling
Thermal shock
Humidity
Mechanical stress
Electrostatic discharge
Voltage stress
Package warpage
Solder fatigue
Electromigration

Testing standards and conditions depend on product class. Automotive, industrial and data-centre components may require different qualification programs from ordinary consumer products.

Failure Analysis

Failed devices can be investigated using:

X-ray imaging
Scanning acoustic microscopy
Optical microscopy
Electron microscopy
Emission microscopy
Electrical probing
Cross-section analysis
Package decapsulation
Material analysis

The objective is to identify the physical root cause and prevent its recurrence.

How Are Complete Graphics Cards Tested?

A card manufacturer may perform the following tests:

Automated Optical Inspection

Cameras check component orientation, placement, missing parts and visible solder defects.

X-Ray Inspection

X-rays examine hidden solder joints beneath the GPU, memory or other ball-grid-array packages.

Electrical Tests

Manufacturers verify:

Input resistance
Power rails
Voltage-regulator operation
Current consumption
Short circuits
Firmware communication

Power-On and Functional Tests

The card is installed in a test system and checked for:

Successful startup
PCIe detection
Driver operation
VRAM availability
Display output
Supported resolutions
Fan control
Sensor readings
Video encoding and decoding
3D rendering
Compute execution

VRAM Testing

Memory tests write and read different patterns to identify:

Defective memory cells
Data corruption
Addressing failures
Timing instability
Temperature-related errors

A card can display the correct memory capacity yet still contain unstable memory, so pattern-based workload testing is important.

Thermal Testing

Engineers monitor:

GPU temperature
Memory temperature
Power-stage temperature
Fan speed
Hotspots
Clock stability
Thermal throttling
Cooler mounting pressure
Noise

Performance Validation

Benchmark results are compared with expected ranges. A score slightly different from another computer is not automatically evidence of a fault because CPU, drivers, operating system, cooling and test settings affect results.

Extended Load Testing

Sustained workloads reveal problems that short tests can miss, including:

Overheating
Clock instability
Driver crashes
Visual artifacts
Power-limit problems
Fan failure
Intermittent memory errors

Port Testing

Each required HDMI or DisplayPort output may be checked. Premium or professional products may undergo additional multi-monitor, colour, refresh-rate and compatibility testing.

Common GPU Defects and Symptoms

Symptom	Possible causes
Coloured blocks or artifacts	VRAM errors, GPU instability, overheating or software problems
Black screen	Power, cable, display, firmware, driver or hardware failure
Driver timeout	Software bug, unstable clock, heat, power or defective hardware
System restarts under load	PSU, card power delivery, heat or system instability
Low performance	Thermal throttling, wrong slot, driver settings, CPU limit or power mode
Excessive fan noise	Dust, restrictive airflow, fan curve, worn fan or high power use
Screen flickering	Cable, monitor, refresh-rate setting, driver or output problem
GPU not detected	Improper installation, BIOS settings, power connection or failed card

These symptoms do not prove that the GPU chip itself is defective. Diagnosis should include the power supply, motherboard, memory, display cable, drivers and operating system.

Safe Ways to Test a GPU

A technician can test a graphics card using:

Visual inspection for damage, dust or loose connectors.
Confirmation that all power connectors are installed correctly.
Official or trusted driver installation.
Device Manager and system-event review.
Temperature and clock monitoring.
Repeatable 3D benchmark testing.
Dedicated VRAM error testing.
Video encode and decode testing where relevant.
Multi-monitor and port testing.
A sustained workload while monitoring temperature and errors.
Testing the card in a known-good computer when necessary.

Safety Warning

Do not open a power supply or dismantle a graphics card while it is powered. Avoid modifying firmware, increasing voltage or removing cooling components unless you understand the risks. These actions can cause permanent damage, electric shock, data loss or warranty cancellation.

GPU Performance Terms Explained

Specification	What it indicates
Core or shader count	Quantity of certain arithmetic resources within that architecture
Clock speed	Approximate operating frequency under specified conditions
Boost clock	Dynamic target frequency, not a guaranteed constant
VRAM capacity	Amount of graphics memory available
Memory bandwidth	Potential data-transfer rate between GPU and memory
Memory-bus width	Number of data paths connecting the GPU and memory
Cache	Fast on-chip storage that reduces external memory traffic
TGP or board power	Product-defined power measurement; terminology varies
Process node	Manufacturing technology generation
ROP or TMU count	Resources associated with pixel output and texture processing
Compute performance	Theoretical arithmetic throughput under defined precision
Ray-tracing performance	Depends on architecture, software and workload
AI performance	Depends heavily on numeric format, sparsity rules and software

Theoretical figures from different architectures, data types or test methods should not be compared without context.

Integrated vs Discrete GPU

Feature	Integrated GPU	Discrete GPU
Location	Inside or beside the CPU/SoC	Separate processor or card
Memory	Usually shares system memory	Usually has dedicated VRAM
Power consumption	Generally lower	Generally higher
Performance	Suitable for display and lighter workloads	Suitable for demanding workloads
Upgradeability	Usually not separately replaceable	Desktop card may be replaceable
Cooling	Shares system cooling	Often has dedicated cooling

Modern integrated GPUs can handle office work, media playback and some creative or gaming workloads. A discrete GPU remains preferable for demanding 3D, AI, rendering and high-resolution professional work.

Consumer, Professional and Data-Centre GPUs

Consumer GPUs

Designed mainly for gaming, home content creation and general computing.

Professional GPUs

May provide:

Certified application drivers
Extended support
Larger memory
Error-correcting features
Specialized synchronization
Professional visualization capabilities
Longer product availability

Data-Centre GPUs

Designed for servers, artificial intelligence and high-performance computing. They may use HBM, advanced packaging and high-speed accelerator interconnects. Some have no physical display connectors.

The most expensive GPU is not automatically the best choice. Software certification, memory capacity, precision support, power, cooling and workload are often more important.

Common GPU Buying Mistakes

Comparing core counts across different architectures
Assuming more VRAM always means more speed
Ignoring power-supply capacity and connector requirements
Forgetting physical card dimensions
Ignoring case airflow
Choosing a gaming GPU for software requiring certified drivers
Buying powerful hardware for a CPU-limited workload
Selecting a card without checking application compatibility
Judging performance only by a single benchmark
Using an unsuitable low-quality power supply
Confusing a GPU designer with a graphics-card brand
Assuming every model with the same GPU has identical cooling and power design

Best Practices When Selecting a GPU

Identify the applications and resolution first.
Check the software vendor’s supported GPU list.
Confirm the required VRAM capacity.
Review independent application-specific testing.
Check PSU wattage and power connectors.
Verify cabinet clearance and slot thickness.
Consider heat, noise and room conditions.
Check driver and operating-system support.
For business use, consider warranty and service availability.
For AI workloads, confirm framework, data-type and software compatibility.
For professional work, examine certification rather than gaming performance alone.

Frequently Asked Questions

Is GPU another name for a graphics card?

Technically, no. The GPU is the processing chip, while the graphics card is the complete board containing the GPU, memory, power circuitry and cooling. The terms are commonly used interchangeably.

Does every computer have a GPU?

Most modern computers have some form of graphics-processing hardware, either integrated into the CPU or system-on-chip or installed as a discrete GPU.

Can a computer run without a discrete graphics card?

Yes, if its processor or motherboard platform provides integrated graphics and the required display output. Some CPUs contain no active integrated GPU.

Who actually manufactures NVIDIA and AMD GPU chips?

NVIDIA and AMD primarily design GPUs and contract suitable semiconductor foundries to fabricate them. The chosen foundry and manufacturing process can vary by product generation. Packaging, memory and board assembly may involve additional suppliers.

Does Intel manufacture its own GPUs?

Intel is an integrated device manufacturer with internal manufacturing capability, but a particular Intel product may use internal manufacturing, external foundry services or a combination of technologies. Product-specific documentation should be checked rather than assuming one supply arrangement for every GPU.

Are ASUS, MSI and Gigabyte GPU manufacturers?

They manufacture or market complete graphics cards, but they generally use GPU processors designed and supplied by companies such as NVIDIA, AMD or Intel.

What is a GPU made from?

The central die is primarily made from silicon containing microscopic transistors and metal interconnects. A complete package or graphics card also uses copper, substrate materials, solder, memory chips, power components, cooling metals, plastics and thermal compounds.

How long does it take to manufacture a GPU?

Wafer fabrication involves hundreds of repeated processes and can take several months. Packaging, testing, card assembly, logistics and final validation add more time. There is no universal duration for every product.

Why are advanced GPUs expensive?

Cost is affected by research and design, software development, mask sets, wafer fabrication, die size, manufacturing yield, advanced packaging, memory, circuit-board complexity, cooling, testing, logistics and market conditions.

What is GPU binning?

Binning is the testing and classification of chips according to working resources, stable frequency, voltage, power and other characteristics.

Why do GPUs become hot?

Millions or billions of switching electronic operations consume electrical power and generate heat. The cooler transfers that heat from the chip to the surrounding air or a liquid-cooling system.

Can overheating permanently damage a GPU?

Modern GPUs normally reduce frequency or shut down when temperature limits are reached. Nevertheless, defective cooling, excessive voltage, sustained unsafe conditions or damaged components can shorten life or cause failure.

Is an AI accelerator always a GPU?

No. GPUs are widely used for AI, but dedicated neural-processing units, tensor processors, custom ASICs and other accelerators can also execute AI workloads.

Can a damaged graphics card be repaired?

Some board-level failures, such as damaged fans, connectors or particular power components, may be repairable by qualified technicians. Internal silicon, multilayer PCB or advanced package failures are often impractical to repair economically.

How can I tell whether a GPU is genuine?

Check the exact device identifier, firmware, memory configuration, physical components, warranty status and performance against reliable specifications. Software information alone may be falsified on counterfeit or modified cards.

Conclusion

A GPU is a highly parallel processor created to handle graphics and other data-intensive calculations. Manufacturing one requires far more than a GPU brand and a single factory. It involves architecture design, electronic verification, silicon wafer production, lithography, deposition, etching, inspection, packaging, memory manufacturing, circuit-board assembly, cooling and several stages of testing.

The GPU industry is therefore an international supply chain. NVIDIA, AMD, Intel and other organizations may design graphics processors; advanced foundries fabricate the silicon; memory companies produce GDDR or HBM; packaging providers connect the components; and board partners manufacture complete graphics cards.

Understanding these separate roles explains why the question “How many companies manufacture GPUs?” has no single numerical answer. There are three major mainstream PC discrete-GPU designers, several integrated and specialist GPU designers, multiple foundries and memory suppliers, and many graphics-card manufacturers worldwide.

#GPU #GraphicsProcessingUnit #GraphicsCard #GPUTechnology #GPUArchitecture #GPUManufacturing #ChipManufacturing #Semiconductor #SemiconductorManufacturing #SiliconChip #GPUComponents #GraphicsCardComponents #HowGPUWorks #ParallelProcessing #CPUvsGPU #IntegratedGPU #DiscreteGPU #GamingGPU #ProfessionalGPU #DataCentreGPU #AIGPU #VRAM #GraphicsMemory #GDDR6 #GDDR7 #HBM #MemoryBandwidth #ShaderCores #CUDACores #ComputeUnits #TensorCores #RayTracing #GraphicsPipeline #ChipFabrication #SiliconWafer #EUVLithography #AdvancedPackaging #GPUFoundry #TSMC #NVIDIA #AMD #IntelGPU #GPUTesting #GraphicsCardTesting #GPUStressTest #GPUCooling #GPUPerformance #GPUBuyingGuide #ComputerHardware #TechnologyExplained