How to use this glossary
Graphics cards come with their own language, and reviews, shop pages and forum posts assume you already know it. This glossary is a map of that language for people who are starting from zero. It is organised by topic rather than alphabetically, because related words make more sense together.
Reading order
If you are brand new, read the performance and hardware groups first, because most other terms build on them. If you are shopping, read the memory, power and buying groups next. If you are tuning a computer you already own, jump to the software and building groups.
How each entry is written
Every entry gives a plain definition, then a short note about why it matters. Where a longer guide exists on this site, the entry links to it. The glossary describes RTX 50 series cards, but most terms apply to any modern graphics card.
A note on accuracy
Definitions here are simplified so that a beginner can follow them. Engineers use more precise versions, and a few terms vary slightly between companies. When a figure matters, check the manufacturer's page for the exact product. The longer guide on reading a graphics card spec sheet shows how to do that.
Performance terms
These words describe how fast and how smooth a game feels.
Speed and smoothness
FPS (frames per second)
FPS is the number of pictures the game draws each second. Higher FPS usually looks smoother and responds faster. Many players aim for 60 FPS as a comfortable minimum and higher for fast shooters. Average FPS alone does not tell the whole story, which is why the next terms exist.
Frame time
Frame time is how long each individual picture takes to draw, measured in milliseconds. Smooth play needs frame times that stay even. A game can average a high FPS and still feel rough if some frames take much longer than others.
1% lows
The 1% lows figure describes the slowest one percent of frames. It reveals stutter that an average hides. When reviews quote 1% lows, they are telling you how bad the worst moments feel.
Stutter
Stutter is a visible hitch or pause in motion, often caused by uneven frame times, running out of video memory, or loading data. It feels worse than a steady lower frame rate.
Delay and limits
Latency (input lag)
Latency is the delay between pressing a button or moving the mouse and seeing the result. Lower is better, especially in competitive games. Features such as NVIDIA Reflex aim to reduce it.
Bottleneck
A bottleneck is one part of your computer limiting another. A slow processor holding back a fast graphics card is the classic case. The site's article on CPU and system bottlenecks explains how to spot one.
Benchmark
A benchmark is a repeatable test used to measure performance, either a built-in game test or a dedicated programme. Benchmarks help compare cards fairly, provided the settings and test machine are stated.
Hardware terms
These words name the physical and logical parts of a graphics card.
The main parts
GPU (graphics processing unit)
The GPU is the processor on the graphics card that draws pictures. People often use "GPU" to mean the whole card, but strictly it is the chip. Our guide to GPU chips explains the difference.
Graphics card
The graphics card is the complete product you buy: the GPU chip, memory, circuit board, cooler and ports. Two cards with the same GPU can differ in cooling and size.
CPU (central processing unit)
The CPU is the main processor of the computer. It runs the operating system and game logic. A fast graphics card still needs a capable CPU to feed it work.
Die
The die is the piece of silicon that forms the chip. A larger die holds more circuits and is more expensive to make.
Inside the chip
CUDA core
A CUDA core is a small calculator inside an NVIDIA GPU. Thousands run in parallel. More cores generally mean more capacity within one generation.
Streaming multiprocessor (SM)
An SM is a block of cores and helper units. Counting SMs is a neat way to compare chips from one family.
RT core
An RT core accelerates ray tracing calculations. Cards with stronger RT cores handle ray-traced games better.
Tensor core
A Tensor core accelerates AI maths, which powers DLSS and other AI features.
TMU and ROP
A texture mapping unit (TMU) applies textures to surfaces. A render output unit (ROP) writes finished pixels to memory. Their counts appear on spec sheets.
Cache
Cache is a small amount of very fast memory on the chip that stores frequently used data. A bigger cache can help cards with narrower memory buses.
Memory terms
Memory words matter more every year because games ask for more of it.
Capacity and type
VRAM (video memory)
VRAM is the memory built into the graphics card. It stores textures, frames and other game data. If a game needs more than the card has, stutter and lower quality follow. Read how much VRAM you need for guidance.
GDDR7
GDDR7 is the newest type of graphics memory, used on most RTX 50 cards. It sends more data per second than earlier types. The article on GDDR7 memory covers the details.
GDDR6
GDDR6 is the previous mainstream graphics memory type. The RTX 5050 uses it. It is slower than GDDR7 but well understood and cheaper.
Clamshell mode
Clamshell mode places memory chips on both sides of the board so that two chips share one connection, doubling capacity without widening the bus. It is how some cards offer a larger memory version.
Speed
Bandwidth
Bandwidth is how much data moves between the GPU and its memory each second. It is calculated from memory speed and bus width. Higher bandwidth helps at high resolutions.
Bus width
Bus width is the size of the data path between the GPU and its memory, in bits. A wider bus moves more data per tick. It usually rises with the tier of the card.
PCIe (PCI Express)
PCIe is the slot connection between the graphics card and the motherboard. Newer generations are faster. The guide on whether you need a PCIe 5.0 motherboard explains the details.
Power and heat terms
Power and temperature words decide whether a card fits and runs well in your computer.
Power
TGP (total graphics power)
TGP is the rated power of the graphics card in watts. It guides the choice of power supply and cooling. A higher TGP usually means a more powerful card.
TDP (thermal design power)
TDP is a similar rating used by many companies to describe the heat a component is designed to give off. Some sheets use TDP where others use TGP.
PSU (power supply unit)
The PSU converts mains electricity into the voltages your computer needs. It must provide enough watts and the right connectors for the graphics card.
12V-2x6
12V-2x6 is the 16-pin power connector used by high-end graphics cards. It must be inserted fully and handled carefully. See the guide to dedicated GPU cables for safe practice.
Heat and tuning
Thermal throttling
Thermal throttling is when a chip slows itself down to avoid overheating. If your card runs hot, performance can drop. Better cooling and airflow help.
Hotspot temperature
The hotspot is the hottest point measured on the chip, usually higher than the average core temperature. Some monitoring tools show both.
Fan curve
A fan curve sets how fast the fans spin at each temperature. A custom curve can make a card quieter or cooler.
Overclocking
Overclocking means running a component faster than its default settings. Gains are usually modest and carry some risk. The site's RTX 50 overclocking guide explains safe steps.
Undervolting
Undervolting lowers the voltage a chip uses, reducing heat and power while often keeping most of the speed. It is popular among owners who want a quieter card.
Power limit
The power limit caps how many watts the card may use. Raising or lowering it changes speed, heat and noise.
Upscaling and AI terms
These words describe NVIDIA's AI-based features.
Making pictures bigger and smoother
DLSS (Deep Learning Super Sampling)
DLSS renders a game at a lower resolution and uses AI to rebuild a sharper picture, which raises frame rates. It is a core RTX feature.
Upscaling
Upscaling means drawing at a lower resolution and enlarging the image to fit your screen. DLSS, FSR and XeSS are different upscalers.
Frame generation
Frame generation creates extra frames between real ones to make motion look smoother. It does not make the game respond faster.
Multi Frame Generation
Multi Frame Generation, exclusive to RTX 50 cards, can create up to three extra frames for each real frame.
DLAA
DLAA applies DLSS image quality improvements at native resolution without upscaling, for the best-looking picture when you have spare performance.
Related technologies
Ray Reconstruction
Ray Reconstruction uses AI to clean up the noisy images that ray tracing produces, giving clearer lighting.
NVIDIA Reflex
Reflex reduces system latency by keeping the processor and graphics card in step. It is supported in many competitive games.
FSR and XeSS
FSR is AMD's upscaler and XeSS is Intel's. They offer similar ideas, and some work on NVIDIA cards.
AI TOPS
TOPS means trillions of operations per second, a measure of AI calculation capacity. It is a marketing figure that does not predict game frame rates.
Lighting terms
Lighting words explain why some games are so demanding.
How light is drawn
Rasterisation
Rasterisation is the traditional way of drawing 3D scenes, turning shapes into pixels quickly using tricks to fake lighting. It is fast and used by all games.
Ray tracing
Ray tracing simulates how light bounces, producing more realistic reflections, shadows and lighting at a high performance cost.
Path tracing
Path tracing is a more complete form of ray tracing that models many bounces of light. It looks very realistic and is extremely demanding.
Colour and brightness
HDR (high dynamic range)
HDR lets a screen show brighter highlights and deeper darks with richer colour. It needs a screen and a game that support it.
Global illumination
Global illumination describes how light from one surface lights up others. Ray tracing can calculate it more accurately than older methods.
Display terms
These words relate to the screen your card connects to.
Sharpness and speed
Resolution
Resolution is the number of pixels on the screen. Common examples are 1080p, 1440p and 4K. Higher resolutions look sharper and demand more from the card.
Refresh rate
Refresh rate is how many times per second the screen updates, measured in hertz. A 144 Hz screen can show up to 144 pictures per second.
VRR (variable refresh rate)
VRR lets the screen match its refresh rate to the game's frame rate, reducing tearing and stutter. NVIDIA's version is called G-Sync, and AMD's is FreeSync. The site's article on VRR, G-Sync, FreeSync and V-Sync explains the options.
V-Sync
V-Sync locks the frame rate to the screen's refresh rate to prevent tearing, sometimes at the cost of added delay.
Connections
HDMI
HDMI is a common video connection found on TVs and monitors. RTX 50 cards offer HDMI 2.1b on supported models.
DisplayPort
DisplayPort is a connection favoured by monitors. RTX 50 cards offer DisplayPort 2.1b on supported models, allowing very high resolutions and refresh rates.
Software terms
Software words cover the programmes that control the card.
Drivers and apps
Driver
A driver is software that lets Windows and games communicate with the graphics card. Keeping it up to date helps performance and stability. The guide to Game Ready and Studio drivers explains the two types.
NVIDIA App
The NVIDIA App manages drivers, game settings and the overlay.
Overlay
An overlay is information drawn on top of a game, such as frame rate and temperatures. It helps you monitor performance.
Tuning tools
MSI Afterburner
MSI Afterburner is a free tool for monitoring, fan control and overclocking that works with many brands of card.
Shader cache
The shader cache stores compiled graphics instructions so that games load and run smoothly. Clearing it can cause temporary stutter while it rebuilds.
Buying and naming terms
These terms appear on shop pages.
Names and labels
Ti
Ti marks a stronger variant between two base tiers, such as the RTX 5070 Ti. See the model names guide.
OC
OC means factory overclocked. The extra speed is usually small.
Founders Edition
The Founders Edition is NVIDIA's own reference card.
Board partner
A board partner is a company such as ASUS, MSI or Gigabyte that builds cards using NVIDIA's chips.
Prices
MSRP
MSRP stands for manufacturer's suggested retail price, the official launch price. Shop prices can differ.
Warranty
A warranty is the manufacturer's promise to repair or replace a faulty product for a set period. Check the terms before buying.
Building terms
These words matter if you install a card yourself.
Inside the case
Airflow
Airflow is the movement of cool air in and hot air out of a computer case. Good airflow keeps temperatures down.
Motherboard and BIOS
The motherboard connects all parts. The BIOS is its built-in software, where settings such as Resizable BAR live.
Resizable BAR
Resizable BAR lets the processor access the card's whole memory at once, which can help in some games.
GPU sag
GPU sag is when a heavy card bends downward in its slot. A support bracket prevents it.
Riser cable
A riser cable lets a card be mounted vertically or away from the slot. Quality matters at high speeds.
Image quality terms
These words describe how the picture looks and the common problems people notice.
Sharpness and smoothness of edges
Native resolution
Native resolution is the resolution your screen is built to display, such as 2560 by 1440 on a 1440p monitor. Running a game at native resolution gives the sharpest picture without scaling blur.
Render resolution
Render resolution is the resolution the game actually draws at before any upscaling. With DLSS enabled it is lower than native, and the output is enlarged to fit the screen.
Anti-aliasing
Anti-aliasing smooths the jagged stair-step edges that appear on diagonal lines. Methods differ in quality and cost, and DLSS and DLAA include AI-based anti-aliasing.
Texture filtering
Texture filtering keeps textures sharp when viewed at an angle or in the distance. Higher settings improve clarity on floors and roads for a small performance cost.
Common visual problems
Screen tearing
Tearing is when the screen shows parts of two different frames at once, creating a horizontal split. VRR and V-Sync reduce it.
Ghosting
Ghosting leaves faint trails behind moving objects. It can come from a slow monitor panel or from certain upscaling or frame generation artefacts.
Artefacts
Artefacts are visual glitches such as shimmering, flickering or odd shapes. They can come from a game bug, an unstable overclock, a faulty cable or an overheating card.
Shader
A shader is a small programme that tells the GPU how to colour and light surfaces. Games contain thousands of them, and compiling them can cause temporary stutter the first time they run.
Sharpening
Sharpening is a filter that increases edge contrast to make the picture look crisper. Too much creates harsh halos around objects.
Storage, cooling and noise terms
A few more words you will meet when building or upgrading.
Storage
SSD (solid-state drive)
An SSD stores data on flash memory and is far faster than a traditional hard drive. Games load faster and stream textures more smoothly from an SSD.
NVMe
NVMe is a fast connection standard for SSDs that plug into small M.2 slots on the motherboard. It uses PCIe lanes, which is why lane sharing appears in motherboard manuals.
Cooling and noise
Thermal paste and thermal pads
Thermal paste and pads carry heat from the chip and memory into the cooler. Factory applications are usually fine, and replacing them can void a warranty.
Zero RPM mode
Zero RPM mode stops the fans at low temperatures, so the card is silent at idle. The fans start when the card warms up. It is normal behaviour on many modern cards.
Coil whine
Coil whine is a high-pitched buzz from electrical components under load. It is usually harmless and varies from card to card.
Backplate
A backplate is a metal plate on the rear of the card that adds strength and sometimes helps cooling. It also holds the weight of heavy cards.
Heatpipe and vapour chamber
Heatpipes and vapour chambers are cooler parts that spread heat quickly from the chip to the fins. Bigger or better versions often mean a cooler, quieter card.
Pairs of terms people often confuse
Some words sound alike or get swapped in everyday speech. This table separates the commonest pairs.
| Pair | What differs |
|---|---|
| GPU and graphics card | Chip versus finished product |
| VRAM and RAM | Card memory versus computer memory |
| Upscaling and frame generation | Fewer pixels versus extra frames |
| Resolution and refresh rate | Sharpness versus update speed |
| FPS and frame time | Count per second versus time per picture |
| TGP and recommended PSU | Card power versus total supply needed |
| Ray tracing and path tracing | Partial light simulation versus fuller simulation |
| Overclocking and undervolting | More speed versus less voltage |
| Driver and NVIDIA App | Core software versus the helper programme |
How to use the table
When a review or forum post uses a term, check whether it means the left or the right side of a pair. Many arguments online come from people using the same word for different things.
Why the pairs matter when you shop
A product page may say "16 GB" without saying whether it means VRAM or system memory. Knowing the difference stops you misreading a listing.
An annotated sentence from a review
Seeing the terms in action helps them stick. Here is an invented sentence in the style of a review, followed by a translation.
> "At 1440p with DLSS Quality and ray tracing on, the card averaged a strong frame rate with good 1% lows, though the 8 GB of VRAM was close to its limit in the busiest scenes."
Translation
"At 1440p" is the screen resolution. "DLSS Quality" is the lightest upscaling mode, which renders at a lower internal resolution. "Ray tracing on" means realistic lighting is enabled, which costs performance. "Averaged a strong frame rate" refers to FPS. "Good 1% lows" means the slowest frames were not much worse than the average, so the game felt smooth. "8 GB of VRAM was close to its limit" warns that heavier settings might cause stutter.
What you can conclude
The card handles the tested setting well, but a buyer who wants higher texture settings should consider a card with more memory. Notice how six glossary terms were needed to understand one sentence.
Abbreviations in one place
- ASUS, MSI, Gigabyte, Palit, Zotac, PNY: brands that build graphics cards with NVIDIA chips.
- AIB: add-in board, another phrase for a graphics card sold by a partner.
- API: application programming interface, the set of rules games use to talk to graphics hardware, such as DirectX or Vulkan.
- DX12: DirectX 12, a common graphics API on Windows.
- NVENC: NVIDIA's hardware video encoder.
- RMA: return merchandise authorisation, the process for returning a faulty product.
- SFF: small form factor, meaning a compact computer.
- SSD: solid-state drive, the fast storage that helps games load.
Where to find more
If a term is not here, search the manufacturer's page or a trustworthy review site. The site's graphics cards comparison page is a good place to see many terms used in real product tables.
How the terms fit together: a short story
Words are easier to remember as a story. Imagine you start a demanding game on a 1440p monitor.
- The CPU runs the game's logic and sends instructions to the driver.
- The driver passes them to the GPU, which uses its CUDA cores to calculate colours and its RT cores to calculate ray tracing.
- Textures and finished frames sit in VRAM, moving at the card's bandwidth.
- DLSS renders at a lower resolution and rebuilds a sharper image, raising the FPS.
- Frame generation adds extra frames, and Reflex keeps latency low.
- The card sends the picture through DisplayPort to the monitor, which refreshes at its refresh rate, with VRR keeping things in step.
- All this uses power measured as TGP and creates heat that the cooler and airflow must remove, or the chip will throttle.
What the story shows
Every term in the glossary has a place in that chain. If something goes wrong, you can ask where in the chain the problem lies: logic, drivers, calculation, memory, upscaling, display or cooling. That is exactly how enthusiasts troubleshoot.
Where to go next
For a gentle overview of how the whole family fits together, read the plain-English RTX 50 series guide. For technical depth, continue with Blackwell architecture.
Suggested study plan for absolute beginners
You do not have to memorise ninety terms. A short plan builds understanding in layers.
Week one: the survival vocabulary
Learn FPS, resolution, refresh rate, VRAM, driver and bottleneck. With these six, you can follow most basic discussions and shop listings.
Week two: the buying vocabulary
Add TGP, PSU, bandwidth, bus width, Ti, OC and warranty. These help you compare cards and avoid compatibility mistakes.
Week three: the feature vocabulary
Learn DLSS, frame generation, Reflex, ray tracing and VRR. These explain what makes RTX cards special and how to set up your games.
Keep a personal list
Write down any word you had to look up. Review the list now and then. Within a month, most readers find that reviews and videos suddenly make sense.
Frequently asked questions about graphics card terms
What is the difference between a GPU and a graphics card?
The GPU is the processor chip that draws the pictures, while the graphics card is the complete product containing the GPU, memory, circuit board, cooler and ports. People often use the terms interchangeably, but two graphics cards can share the same GPU and still differ in cooling, size and price.
What does FPS mean in gaming?
FPS means frames per second, the number of pictures a game draws each second. Higher FPS usually looks smoother and responds faster. Many players aim for at least 60 FPS, while competitive players on fast monitors often aim higher. Frame time and 1% lows give a fuller picture of smoothness.
What is the difference between VRAM and RAM?
VRAM is the memory on the graphics card, used for textures and frames, while RAM is the computer's main memory used by the processor and programmes. They are separate, so adding RAM does not give your graphics card more VRAM. Games need enough of both.
What is TGP on a graphics card?
TGP, or total graphics power, is the rated power of the card in watts. It helps you choose a suitable power supply and estimate heat and noise. Higher TGP usually means a faster card that needs stronger cooling and a more capable power supply.
Is frame generation the same as upscaling?
No. Upscaling draws fewer pixels and rebuilds a larger image, which raises the frame rate. Frame generation creates extra frames between real ones to make motion look smoother. Many games let you use both. Upscaling can improve responsiveness, while frame generation mainly improves visual smoothness.
What is a bottleneck in a gaming computer?
A bottleneck is when one component limits another, such as a slow processor stopping a fast graphics card from reaching its full speed. You can spot one by watching usage: if the graphics card sits well below full load while the frame rate is low, something else is holding it back.