Independent guide · Updated November 2026

Every GeForce RTX card. Every variant. Every spec.

From the RTX 2060 to the RTX 5090: 35 desktop cards and 23 laptop GPUs across four generations, compared side by side. Plus which one to buy for gaming, creative work and running AI on your own PC.

Compare all models Help me choose

4generations: Turing, Ampere, Ada Lovelace, Blackwell

35desktop models and variants

23laptop GPUs

32 GBmost VRAM on a GeForce card (RTX 5090)

Current generation

GeForce RTX 50 Series

Built on NVIDIA's Blackwell architecture, with fifth-generation Tensor Cores, fourth-generation RT Cores, GDDR7 memory on every model except the RTX 5050, and support for DLSS 5 and Multi Frame Generation. Prices are NVIDIA's UK recommended retail prices in November 2026; shop prices vary.

Blackwell

RTX 5090

VRAM: 32 GB GDDR7
CUDA cores: 21,760
AI TOPS: 3,352
Bandwidth: 1,792 GB/s
Boost clock: 2.41 GHz
Board power: 575 W

4K and 8K gaming, the largest local AI models on a desktop

UK recommended price £1,799

Blackwell

RTX 5080

VRAM: 16 GB GDDR7
CUDA cores: 10,752
AI TOPS: 1,801
Bandwidth: 960 GB/s
Boost clock: 2.62 GHz
Board power: 360 W

High-refresh 4K gaming, 3D and video work

UK recommended price £909

Blackwell

RTX 5070 Ti

VRAM: 16 GB GDDR7
CUDA cores: 8,960
AI TOPS: 1,406
Bandwidth: 896 GB/s
Boost clock: 2.45 GHz
Board power: 300 W

1440p at high refresh, entry 4K, 16 GB for AI

UK recommended price £729

Blackwell

RTX 5070

VRAM: 12 GB GDDR7
CUDA cores: 6,144
AI TOPS: 988
Bandwidth: 672 GB/s
Boost clock: 2.51 GHz
Board power: 250 W

1440p gaming

UK recommended price £499

Blackwell

RTX 5060 Ti

16 GB and 8 GB

VRAM: 16 GB / 8 GB GDDR7
CUDA cores: 4,608
AI TOPS: 759
Bandwidth: 448 GB/s
Boost clock: 2.57 GHz
Board power: 180 W

1080p to 1440p; choose 16 GB for AI and future games

UK recommended price from £349

Blackwell

RTX 5060

VRAM: 8 GB GDDR7
CUDA cores: 3,840
AI TOPS: 614
Bandwidth: 448 GB/s
Boost clock: 2.50 GHz
Board power: 145 W

1080p gaming

UK recommended price £275

Blackwell

RTX 5050

VRAM: 8 GB GDDR6
CUDA cores: 2,560
AI TOPS: 421
Bandwidth: 320 GB/s
Boost clock: 2.57 GHz
Board power: 130 W

Entry 1080p gaming

UK recommended price £249

Specifications

Every desktop GeForce RTX card

Filter by generation, search, or click a column heading to sort. Figures are NVIDIA's reference specifications; partner cards often ship with higher clocks and power limits.

Model	Architecture	CUDA cores	VRAM	Memory	Bus	Bandwidth	Boost clock	Board power
RTX 5090	Blackwell	21,760	32 GB	GDDR7	512-bit	1,792 GB/s	2.41 GHz	575 W
RTX 5080	Blackwell	10,752	16 GB	GDDR7	256-bit	960 GB/s	2.62 GHz	360 W
RTX 5070 Ti	Blackwell	8,960	16 GB	GDDR7	256-bit	896 GB/s	2.45 GHz	300 W
RTX 5070	Blackwell	6,144	12 GB	GDDR7	192-bit	672 GB/s	2.51 GHz	250 W
RTX 5060 Ti16 GB and 8 GB	Blackwell	4,608	16 GB / 8 GB	GDDR7	128-bit	448 GB/s	2.57 GHz	180 W
RTX 5060	Blackwell	3,840	8 GB	GDDR7	128-bit	448 GB/s	2.50 GHz	145 W
RTX 5050	Blackwell	2,560	8 GB	GDDR6	128-bit	320 GB/s	2.57 GHz	130 W
RTX 4090	Ada Lovelace	16,384	24 GB	GDDR6X	384-bit	1,008 GB/s	2.52 GHz	450 W
RTX 4080 SUPER	Ada Lovelace	10,240	16 GB	GDDR6X	256-bit	736 GB/s	2.55 GHz	320 W
RTX 4080	Ada Lovelace	9,728	16 GB	GDDR6X	256-bit	717 GB/s	2.51 GHz	320 W
RTX 4070 Ti SUPER	Ada Lovelace	8,448	16 GB	GDDR6X	256-bit	672 GB/s	2.61 GHz	285 W
RTX 4070 Ti	Ada Lovelace	7,680	12 GB	GDDR6X	192-bit	504 GB/s	2.61 GHz	285 W
RTX 4070 SUPER	Ada Lovelace	7,168	12 GB	GDDR6X	192-bit	504 GB/s	2.48 GHz	220 W
RTX 4070GDDR6X and GDDR6	Ada Lovelace	5,888	12 GB	GDDR6X / GDDR6	192-bit	504 / 480 GB/s	2.48 GHz	200 W
RTX 4060 Ti16 GB and 8 GB	Ada Lovelace	4,352	16 GB / 8 GB	GDDR6	128-bit	288 GB/s	2.54 GHz	165 / 160 W
RTX 4060	Ada Lovelace	3,072	8 GB	GDDR6	128-bit	272 GB/s	2.46 GHz	115 W
RTX 3090 Ti	Ampere	10,752	24 GB	GDDR6X	384-bit	1,008 GB/s	1.86 GHz	450 W
RTX 3090	Ampere	10,496	24 GB	GDDR6X	384-bit	936 GB/s	1.70 GHz	350 W
RTX 3080 Ti	Ampere	10,240	12 GB	GDDR6X	384-bit	912 GB/s	1.67 GHz	350 W
RTX 3080 12 GB12 GB	Ampere	8,960	12 GB	GDDR6X	384-bit	912 GB/s	1.71 GHz	350 W
RTX 308010 GB	Ampere	8,704	10 GB	GDDR6X	320-bit	760 GB/s	1.71 GHz	320 W
RTX 3070 Ti	Ampere	6,144	8 GB	GDDR6X	256-bit	608 GB/s	1.77 GHz	290 W
RTX 3070	Ampere	5,888	8 GB	GDDR6	256-bit	448 GB/s	1.73 GHz	220 W
RTX 3060 TiGDDR6 and GDDR6X	Ampere	4,864	8 GB	GDDR6 / GDDR6X	256-bit	448 / 608 GB/s	1.67 GHz	200 / 225 W
RTX 306012 GB and 8 GB	Ampere	3,584	12 GB / 8 GB	GDDR6	192 / 128-bit	360 / 240 GB/s	1.78 GHz	170 W
RTX 30508 GB	Ampere	2,560	8 GB	GDDR6	128-bit	224 GB/s	1.78 GHz	130 W
RTX 3050 6 GB6 GB, low power	Ampere	2,304	6 GB	GDDR6	96-bit	168 GB/s	1.47 GHz	70 W
RTX 2080 Ti	Turing	4,352	11 GB	GDDR6	352-bit	616 GB/s	1.55 GHz	250 W
RTX 2080 SUPER	Turing	3,072	8 GB	GDDR6	256-bit	496 GB/s	1.82 GHz	250 W
RTX 2080	Turing	2,944	8 GB	GDDR6	256-bit	448 GB/s	1.71 GHz	215 W
RTX 2070 SUPER	Turing	2,560	8 GB	GDDR6	256-bit	448 GB/s	1.77 GHz	215 W
RTX 2070	Turing	2,304	8 GB	GDDR6	256-bit	448 GB/s	1.62 GHz	175 W
RTX 2060 SUPER	Turing	2,176	8 GB	GDDR6	256-bit	448 GB/s	1.65 GHz	175 W
RTX 2060 12 GB12 GB	Turing	2,176	12 GB	GDDR6	192-bit	336 GB/s	1.65 GHz	185 W
RTX 20606 GB	Turing	1,920	6 GB	GDDR6	192-bit	336 GB/s	1.68 GHz	160 W

Where a model has variants (memory size or type), both values are shown, larger first. Board power is NVIDIA's total graphics power for the reference design.

Laptops

GeForce RTX laptop GPUs

Laptop GPUs share names with desktop cards but have fewer cores, less memory and much lower power. Each laptop maker sets the power limit (TGP), so the same GPU can perform very differently in a thin laptop and a thick one: check the TGP in the laptop's specifications.

Laptop GPU	Architecture	CUDA cores	VRAM	Memory	Bus
RTX 5090 Laptop GPU	Blackwell	10,496	24 GB	GDDR7	256-bit
RTX 5080 Laptop GPU	Blackwell	7,680	16 GB	GDDR7	256-bit
RTX 5070 Ti Laptop GPU	Blackwell	5,888	12 GB	GDDR7	192-bit
RTX 5070 Laptop GPU8 GB, and 12 GB since April 2026	Blackwell	4,608	8 GB / 12 GB	GDDR7	128-bit
RTX 5060 Laptop GPU	Blackwell	3,328	8 GB	GDDR7	128-bit
RTX 5050 Laptop GPU	Blackwell	2,560	8 GB	GDDR7	128-bit
RTX 4090 Laptop GPU	Ada Lovelace	9,728	16 GB	GDDR6	256-bit
RTX 4080 Laptop GPU	Ada Lovelace	7,424	12 GB	GDDR6	192-bit
RTX 4070 Laptop GPU	Ada Lovelace	4,608	8 GB	GDDR6	128-bit
RTX 4060 Laptop GPU	Ada Lovelace	3,072	8 GB	GDDR6	128-bit
RTX 4050 Laptop GPU	Ada Lovelace	2,560	6 GB	GDDR6	96-bit
RTX 3080 Ti Laptop GPU	Ampere	7,424	16 GB	GDDR6	256-bit
RTX 3080 Laptop GPU16 GB and 8 GB	Ampere	6,144	16 GB / 8 GB	GDDR6	256-bit
RTX 3070 Ti Laptop GPU	Ampere	5,888	8 GB	GDDR6	256-bit
RTX 3070 Laptop GPU	Ampere	5,120	8 GB	GDDR6	256-bit
RTX 3060 Laptop GPU	Ampere	3,840	6 GB	GDDR6	192-bit
RTX 3050 Ti Laptop GPU	Ampere	2,560	4 GB	GDDR6	128-bit
RTX 3050 Laptop GPU4 GB and 6 GB	Ampere	2,048	4 GB / 6 GB	GDDR6	128 / 96-bit
RTX 2080 SUPER Laptop GPU	Turing	3,072	8 GB	GDDR6	256-bit
RTX 2080 Laptop GPU	Turing	2,944	8 GB	GDDR6	256-bit
RTX 2070 SUPER Laptop GPU	Turing	2,560	8 GB	GDDR6	256-bit
RTX 2070 Laptop GPU	Turing	2,304	8 GB	GDDR6	256-bit
RTX 2060 Laptop GPU	Turing	1,920	6 GB	GDDR6	192-bit

Variants explained

Same name, different card

Memory variants

Several models come in two memory sizes: RTX 5060 Ti (16 GB or 8 GB), RTX 4060 Ti (16 GB or 8 GB), RTX 3060 (12 GB or 8 GB), RTX 3080 (12 GB or 10 GB) and RTX 2060 (12 GB or 6 GB). The larger version is the safer buy for new games and essential for AI. Some differ in memory type too: RTX 4070 (GDDR6X or GDDR6) and RTX 3060 Ti (GDDR6 or GDDR6X).

Ti and SUPER

Ti and SUPER mark faster versions of a model. SUPER cards usually arrive part-way through a generation as a refresh, like the RTX 4080 SUPER, 4070 Ti SUPER and 4070 SUPER in 2024. RTX 50 SUPER cards have been widely reported but have not launched as of November 2026.

Founders Edition and partner cards

Founders Edition cards are NVIDIA's own design. Partner cards from ASUS, MSI, Gigabyte, Zotac, PNY and others use the same GPU with their own coolers, clocks and power limits. They can be faster, quieter or smaller, so compare size, power connectors and noise as well as price.

Desktop and laptop

An "RTX 5090 Laptop GPU" is not a desktop RTX 5090 in a laptop. It has under half the cores of the desktop card and runs at a fraction of the power. Compare laptop GPUs with each other, not with desktop cards.

Four generations

What each generation added

Series	Architecture	Launched	Process	Tensor Cores	RT Cores	Encoder	Memory	DLSS features
GeForce RTX 50 Series	Blackwell	2025	TSMC 4N	5th gen	4th gen	9th gen	GDDR7 (GDDR6 on RTX 5050)	DLSS 5, Multi Frame Generation, Frame Generation, Super Resolution, Ray Reconstruction
GeForce RTX 40 Series	Ada Lovelace	2022–2024	TSMC 4N	4th gen	3rd gen	8th gen (adds AV1 encode)	GDDR6X / GDDR6	Frame Generation, Super Resolution, Ray Reconstruction
GeForce RTX 30 Series	Ampere	2020–2022	Samsung 8 nm	3rd gen	2nd gen	7th gen	GDDR6X / GDDR6	Super Resolution, Ray Reconstruction
GeForce RTX 20 Series	Turing	2018–2019	TSMC 12 nm	2nd gen	1st gen	7th gen	GDDR6	Super Resolution, Ray Reconstruction

DLSS Super Resolution and Ray Reconstruction work on every RTX card. Frame Generation needs an RTX 40 or 50 Series card; Multi Frame Generation and DLSS 5's neural rendering need an RTX 50 Series card.

Buying guide

Which card for what

Our starting points for new RTX 50 Series cards. Older cards at a good price can be excellent value too: compare the specs above.

You want	Start with	Why
1080p gaming	RTX 5060 or RTX 5060 Ti 16 GB	High frame rates at 1080p; the 16 GB Ti has room for the next few years of games.
1440p gaming	RTX 5070 or RTX 5070 Ti	12 to 16 GB of fast GDDR7 for high refresh rates at 1440p.
4K gaming	RTX 5080 or RTX 5090	The bandwidth and memory for 4K with ray tracing; the 5090 for the highest settings.
Video editing and streaming	RTX 5070 Ti or above	Two ninth-generation encoders from the 5070 Ti up (three on the 5090) speed up exports.
3D and rendering	RTX 5080 or RTX 5090	Scene size is limited by VRAM; the 5090's 32 GB fits the biggest scenes.
Local AI and development	RTX 5060 Ti 16 GB, RTX 5070 Ti or RTX 5090	VRAM decides which models you can run; see the guide below.

Check your power supply and case: the RTX 5090 is rated at 575 W and NVIDIA recommends a 1000 W power supply for it. Many 50 Series cards use the 16-pin (12V-2x6) power connector.

AI on RTX

Running AI models on a GeForce card

GeForce RTX cards run AI models locally with tools such as Ollama, LM Studio and ComfyUI, keeping your data on your own PC. For language models, the amount of VRAM decides the largest model you can run at a usable speed. These are rough guides for 4-bit quantised models with a modest context; longer contexts need more memory.

VRAM	Example cards	Language models that fit (approx.)
8 GB	RTX 5060, 5050, 4060	Up to about 8 billion parameters
12 GB	RTX 5070, 4070 SUPER, 3060 12 GB	Up to about 14 billion parameters
16 GB	RTX 5060 Ti 16 GB, 5070 Ti, 5080, 4080 SUPER	Up to about 20 to 24 billion parameters
24 GB	RTX 4090, 3090 Ti, 3090	Up to about 32 billion parameters
32 GB	RTX 5090	About 32 billion parameters with long contexts or higher precision

AI TOPS is NVIDIA's measure of peak AI throughput; Blackwell's figures use FP4, so they aren't directly comparable with older generations. Bigger models need more memory than any GeForce card has: see the options below.

From Workstation

When a GeForce card isn't enough

Workstation builds and supports AI workstations and GPU servers in the UK. When you need more memory than a GeForce card has, 24/7 duty, certified drivers or several GPUs in one system, we can help: from DGX Spark with 128 GB of unified memory on a desk, to RTX PRO 6000 servers with 96 GB per GPU, to full GPU clusters.

Visit the Workstation shop AI workstations GPU leasing

Questions

FAQ

Is this site run by NVIDIA?

No. GeForce RTX Guide is an independent guide written and published by Workstation, a UK company that builds and supports AI workstations and GPU servers. It is not affiliated with, sponsored or endorsed by NVIDIA. NVIDIA, GeForce, GeForce RTX, RTX and DLSS are trademarks of NVIDIA Corporation.

Where do the specifications come from?

From NVIDIA's published specifications for each card at launch. Cards from board partners such as ASUS, MSI and Gigabyte often run higher clocks and power limits than these reference figures, so check the exact card you are buying.

Are there RTX 50 SUPER cards?

Not as of November 2026. RTX 50 SUPER models with more memory were widely reported but NVIDIA has not launched them. We will add them here if it does.

How much VRAM do I need?

For 1080p gaming, 8 GB is the minimum and 12 GB is more comfortable for new games. For 1440p and 4K, 12 to 16 GB. For running AI models locally, memory matters more than anything else: see the AI section above.

Why do laptop GPUs with the same name perform differently?

Each laptop maker sets the GPU's power limit (TGP) to suit its cooling. The same RTX 5070 Laptop GPU can be much faster in a thick laptop than in a thin one, so compare the TGP as well as the name.

What is the difference between Ti, SUPER and the plain model?

Ti and SUPER mark faster versions of a model, usually with more cores and sometimes more memory. A SUPER refresh normally arrives part-way through a generation, as the RTX 40 SUPER cards did in 2024.

Should I buy a GeForce RTX or an RTX PRO card for work?

GeForce RTX is excellent value for gaming, creative work and AI development on one desk. Choose RTX PRO workstation or server cards when you need more memory (up to 96 GB per card), certified drivers, 24/7 duty in a server, or virtualisation. DGX Spark gives 128 GB of unified memory for larger AI models on a desk.