RTX 5090 vs RTX 4090 for 4K Gaming and Local AI
Updated 20 November 2026 · 4 min read
The RTX 5090 is the faster card, with 32 GB of GDDR7 and about 75% more memory bandwidth than the RTX 4090. The 4090 runs the same model sizes on 24 GB, with less room for long contexts, draws 450 W instead of 575 W and asks for an 850 W power supply instead of 1,000 W. Choose the 5090 for the most speed and memory on one card, the 4090 if 24 GB covers your models and your power supply.
Side by side#
| NVIDIA GeForce RTX 5090 | NVIDIA GeForce RTX 4090 | |
|---|---|---|
| Architecture | Blackwell (GB202) | Ada Lovelace (AD102) |
| CUDA cores | 21,760 | 16,384 |
| Memory | 32 GB GDDR7, 512-bit | 24 GB GDDR6X, 384-bit |
| Memory bandwidth | 1,792 GB/s | 1,008 GB/s |
| Graphics card power | 575 W | 450 W |
| Recommended PSU | 1,000 W | 850 W |
| Interface | PCIe 5.0 x16 | PCIe 4.0 x16 |
| Upscaling | DLSS 4 with Multi Frame Generation | DLSS 3.5 with Frame Generation |
| Outputs | 3 × DisplayPort 2.1b, 1 × HDMI 2.1b | 3 × DisplayPort 1.4a, 1 × HDMI 2.1 |

NVIDIA
NVIDIA GeForce RTX 5090
- 32 GB
- NVIDIA Blackwell (GB202)
Runs up to 30–34B at 4-bit
32 GB runs 30B-class models comfortably at 4-bit, but a 70B model will not fit on one card. For 70B, plan for two cards or a large unified-memory system.
£1,669
NVIDIA
NVIDIA GeForce RTX 4090
- 24 GB
- NVIDIA Ada Lovelace (AD102)
Runs up to 30–34B at 4-bit
Push the 16-pin power cable fully into the card and avoid sharp bends close to the plug. A loose connector is the most common cause of trouble with this card.
£1,3394K gaming#
Both are 4K cards. The RTX 5090 adds more cores, more bandwidth and DLSS 4 Multi Frame Generation, which helps most on 4K screens at 240 Hz and with path tracing. The RTX 4090 still plays at 4K with high settings, ray tracing and DLSS 3.5 frame generation. At 1440p or 1080p, much of either card's power goes unused.
Local AI#
- Model sizes: about 90% of the memory is usable, so both run 30B-class models at 4-bit and 14B at 8-bit. The 5090's extra 8 GB leaves room for longer contexts.
- Speed: text generation follows memory bandwidth, and the 5090 has about 75% more (1,792 against 1,008 GB/s).
- 70B models: neither holds one alone. Two cards, a 48 GB or 96 GB workstation card or a unified-memory machine do; see how much memory a local LLM needs.
- FP4: the 5090's fifth-generation Tensor Cores support FP4; the 4090's do not, though it handles 8-bit inference well.
Power, size and fit#
The 5090 draws up to 575 W through one 16-pin 12V-2x6 connector and NVIDIA recommends 1,000 W; the 4090 draws up to 450 W and NVIDIA recommends 850 W. Both Founders Editions are 304 mm long; the 5090's is two slots thick, the 4090's three. Partner cards differ, so check the exact model. Our guide to fitting a graphics card covers connectors, cases and airflow.
Which to choose#
- RTX 5090: the most speed and memory on one GeForce card, for 4K at high refresh rates and for 30B-class models with long contexts.
- RTX 4090: 24 GB of CUDA memory covers your models, or your power supply is 850 W.
- Neither: for 4K gaming alone, the 16 GB RTX 5080 is a realistic choice; for 70B models, look at the RTX PRO 6000 or a 128 GB mini PC.
Questions people ask
How much faster is the RTX 5090 than the RTX 4090?
On paper it has about a third more CUDA cores and about 75% more memory bandwidth. How much faster it runs depends on the game, the resolution and the model; the gap is largest at 4K and on memory-bound AI work.
Can the RTX 4090 run 70B models?
Not on one card. Two RTX 4090s give 48 GB, enough for a 70B model at 4-bit split across both.
Will an RTX 5090 work in a PCIe 4.0 slot?
Yes, at PCIe 4.0 speed. For a model held in the card's memory, the slot speed mostly affects loading time.
Do I need a new power supply for the RTX 5090?
NVIDIA recommends 1,000 W and a 16-pin 12V-2x6 cable or the adapter in the box. Many 850 W units fall short, so check before you order.
Shop this topic: Graphics cards · Graphics cards for 4K gaming · Hardware for local AI
Related guides
Graphics cards · 5 min read
Choosing a GPU for Local AI: RTX 5090, R9700, B60
Choose a GPU for local AI by memory, software and bandwidth: RTX 5090, Radeon AI PRO R9700, Arc Pro B60 and the 48 GB and 96 GB workstation cards.
Read the guide →Graphics cards · 4 min read
Fitting a High-End Graphics Card: Power, Case, Cooling
Before you order a high-end graphics card: power supply wattage, the 16-pin connector, case size, PCIe slots, airflow and running two cards.
Read the guide →Graphics cards · 4 min read
RTX 5080 vs RTX 5070 Ti: Is the Step Up Worth It?
RTX 5080 vs RTX 5070 Ti: both have 16 GB of GDDR7. What the 5080's extra cores and bandwidth give you at 4K, what they don't change for AI, and power.
Read the guide →