Showing 631 Graphics Cards found
- GPU Brand
- NVIDIA
- Series
- GeForce RTX 50
- Tier
- Upper Mid-Range
$759.00
- GPU Brand
- AMD
- Series
- Radeon RX 9000
- Tier
- Upper Mid-Range
$719.00
- GPU Brand
- NVIDIA
- Series
- GeForce RTX 40
- Tier
- Mainstream
$499.00
Hardware finder
Compare graphics cards by architecture, series, tier and board partner.
Updated October 10, 2026
631 graphics cards from 5 brands. Prices from $55.00 to $8,565.00.
Consumercarts is reader-supported. Learn more
Showing 631 Graphics Cards found
$759.00
$719.00
$499.00
Compare graphics cards by architecture, series, tier and board partner.
Graphics cards listed with the specifications that decide whether one works in your build: card length, slot width, board power, the power connectors it needs, and the supply wattage the maker recommends.
Length against your case clearance is the check that catches most people. Our case clearance guide covers the arithmetic, and GPU clearance with a radiator handles the case where a front-mounted AIO takes the space first.
The catalog holds 631 cards. Where a card has enough public Blender Open Data results to be trustworthy we publish a measured render score with its sample size, which covers 191 cards today. Where it does not, we say so rather than substituting a marketing number.
Note that board partner cards of the same GPU differ in length, cooler thickness and power connector, sometimes by enough to change whether the card fits. Compare the specific card you are buying, not the chip.
For picks by use: desktop GPU tier list, best GPU for 1440p, budget cards under $300 and best GPU for local LLMs. On VRAM, bus width and 1440p ray tracing explains why capacity alone is not the number to read.
Start from the resolution and refresh rate of the monitor you already own. A card that is overkill at 1080p is mid range at 1440p and entry level at 4K, so the panel sets the budget, not the other way round.
8GB is workable at 1080p and already tight at 1440p in recent titles with high textures. 12GB is the sensible floor for 1440p, 16GB for 4K or for local AI work. A card with a fast GPU and too little memory falls off a cliff rather than degrading gracefully, which is why VRAM decides how long the card stays usable.
Three measurements, all of which sink builds: card length against your case clearance, slot thickness (many cards are 3 or 3.5 slots and block the slot below), and the power connectors your PSU actually has. A 12V-2x6 card on a supply with only 8 pin connectors means an adapter, which is where most melted connector reports start.
A synthetic score ranks cards against each other under an identical load. Real games vary by engine, by API and by how much the CPU is holding the card back. Use the synthetic figure for ordering, and a frame rate in a game you actually play for the decision.
Upscaling has become the deciding feature at the low end, because it is often the difference between playable and not. Judge the quality and availability of the upscaler in the games you play, not the marketing tier.
Compare the card length against the GPU clearance figure on your case, and leave 20 to 30mm beyond it for the power connector and cable bend. If you run a front radiator, use the case second, shorter figure.
Start with the maker recommended wattage on the card, then confirm your supply physically has the connectors it needs, whether that is multiple 8-pin or a 12V-2x6.
Only up to what your resolution and settings use. Beyond that, memory bandwidth and the GPU itself decide performance, and extra unused VRAM does nothing.
How many expansion slots the cooler physically occupies. A 3.5 slot card mounts in a three slot space but blocks the slot below and may press against the side panel.
No current card is limited by PCIe 4.0 x16. A PCIe 5.0 card works in a 4.0 slot at 4.0 speeds with no meaningful loss.