readygpu

NVIDIA GeForce RTX 3090 vs NVIDIA GeForce RTX 4090

For running models locally, priced in the UK. At 8k context the first runs 51 models fully in memory, the second 51.

In short

Both have 24 GB of memory for models; the NVIDIA GeForce RTX 3090 has 936 GB/s of bandwidth and the NVIDIA GeForce RTX 4090 1,008 GB/s. At 8k context both run the same 51 models we track fully in memory. On the models both run fully in memory, the NVIDIA GeForce RTX 4090 is about 1.1× as fast as the NVIDIA GeForce RTX 3090 (estimated, median across models). In the UK the NVIDIA GeForce RTX 3090 costs £1,707 at Amazon and the NVIDIA GeForce RTX 4090 £6,015 at Amazon: the NVIDIA GeForce RTX 3090 is £4,308 cheaper.

Updated . Speeds are estimates; how they are worked out.

NVIDIA GeForce RTX 3090

£1,707

Amazon

Add to build
NVIDIA GeForce RTX 4090

£6,015

Amazon

Add to build

Specifications

SpecNVIDIA GeForce RTX 3090NVIDIA GeForce RTX 4090
GPUGeForce RTX 3090GeForce RTX 4090
Memory24 GB GDDR6X24 GB GDDR6X
Memory bus384-bit384-bit
Memory bandwidth936 GB/s1,008 GB/s
FP16 compute71 TFLOPS165 TFLOPS
SoftwareCUDACUDA
Board power350 W450 W
PCIe generationGen 4Gen 4
PCIe lanesx16x16

What each runs

Estimated decode speed with the most precise quant that fits each one est.

ModelNVIDIA GeForce RTX 3090NVIDIA GeForce RTX 4090
Qwen 3.8 27B26tok/sUD-Q6_K28tok/sUD-Q6_K
Qwen 3.6 27B25tok/sQ6_K27tok/sQ6_K
Gemma 4 31B25tok/sQ5_K_M27tok/sQ5_K_M
Qwen 3.6 35B-A3B167tok/sQ4_K_M179tok/sQ4_K_M
Qwen 3.5 27B25tok/sQ6_K27tok/sQ6_K
Gemma 4 26B-A4B88tok/sUD-Q6_K95tok/sUD-Q6_K
Qwen 3.5 35B-A3B168tok/sQ4_K_M181tok/sQ4_K_M
GLM 4.7 Flash87tok/sQ5_K_M94tok/sQ5_K_M
Devstral Small 2 24B28tok/sQ6_K30tok/sQ6_K
gpt-oss 20B132tok/sMXFP4142tok/sMXFP4
Qwen3 Coder 30B-A3B118tok/sQ5_K_M128tok/sQ5_K_M
Qwen3 32B27tok/sQ4_K_M29tok/sQ4_K_M
Gemma 4 12B44tok/sQ8_048tok/sQ8_0
Qwen 3.5 9B59tok/sQ8_064tok/sQ8_0
Mistral Small 3.2 24B28tok/sQ6_K30tok/sQ6_K
DeepSeek R1 Distill Qwen 32B26tok/sQ4_K_M28tok/sQ4_K_M
Gemma 3 27B25tok/sQ6_K27tok/sQ6_K
Ministral 3 14B37tok/sQ8_040tok/sQ8_0
Qwen3 14B34tok/sQ8_037tok/sQ8_0
DeepSeek R1 0528 Qwen3 8B59tok/sQ8_063tok/sQ8_0
Qwen3 8B59tok/sQ8_063tok/sQ8_0
Phi-4 14B34tok/sQ8_036tok/sQ8_0
Qwen 3.5 4B121tok/sQ8_0130tok/sQ8_0
Ministral 3 8B57tok/sQ8_062tok/sQ8_0
Granite 4.2 8B54tok/sQ8_059tok/sQ8_0
Gemma 4 E4B69tok/sQ8_074tok/sQ8_0
LFM2.5 8B-A1B237tok/sQ8_0256tok/sQ8_0
Llama 3.1 8B76tok/sQ6_K82tok/sQ6_K
Qwen2.5 0.5B919tok/sQ8_0989tok/sQ8_0
Qwen3.5 0.8B623tok/sQ8_0671tok/sQ8_0
Qwen3.5 2B373tok/sQ5_K_M402tok/sQ5_K_M
DeepSeek R1 Distill Qwen 1.5B273tok/sQ8_0294tok/sQ8_0
MiniCPM5 2B191tok/sQ8_0206tok/sQ8_0
LFM2.5 2.6B172tok/sQ8_0185tok/sQ8_0
LFM2.5 VL 3B172tok/sQ8_0185tok/sQ8_0
LFM2.5 24B A2B143tok/sQ6_K154tok/sQ6_K
granite vision 4.1 4b135tok/sQ8_0146tok/sQ8_0
granite 4.2 3b127tok/sQ8_0137tok/sQ8_0
Mellum2.1 12B A2.5B Thinking123tok/sQ8_0133tok/sQ8_0
Kimi VL A3B99tok/smlx-8bit107tok/smlx-8bit
DeepSeek R1 Distill Qwen 7B68tok/sQ8_073tok/sQ8_0
DeepSeek R1 Distill Llama 8B60tok/sQ8_065tok/sQ8_0
Ministral 3 8B Reasoning 251257tok/sQ8_062tok/sQ8_0
GLM 4.6V Flash56tok/sQ8_061tok/sQ8_0
GLM 4.1V 9B Thinking48tok/smlx-8bit52tok/smlx-8bit
DeepSeek R1 Distill Qwen 14B34tok/sQ8_036tok/sQ8_0
Devstral Small 250732tok/sQ5_K_M35tok/sQ5_K_M
Magistral Small 250732tok/sQ5_K_M35tok/sQ5_K_M
Magistral Small 250932tok/sQ5_K_M35tok/sQ5_K_M
Magistral Small 250628tok/sQ6_K30tok/sQ6_K
granite 4.2 30b25tok/sQ5_K_M27tok/sQ5_K_M
Qwen 3.5 122B-A10Boffloadoffload
Qwen3 Coder Nextoffloadoffload
Mistral Small 4 119Boffloadoffload
GLM 4.5 Airoffloadoffload
Qwen3 Next 80B-A3Boffloadoffload
Nemotron 3.5 Lightning 30B-A3Boffloadoffload
Kimi Linear 48B-A3Boffloadoffload
Llama 3.3 70Boffloadoffload
DeepSeek R1 Distill Llama 70Boffloadoffload
DeepSeek V4 Flashwon't runwon't run
GLM 5.3 Flashwon't runwon't run
Qwen 3.8 Flash Nextwon't runwon't run
MiniMax M2.7won't runwon't run
Nemotron 3 Super 120B-A12Bwon't runwon't run
gpt-oss 120Bwon't runwon't run
DeepSeek R1won't runwon't run
DeepSeek R1 0528won't runwon't run
DeepSeek V3 0324won't runwon't run
DeepSeek V3.1won't runwon't run
DeepSeek V3.2won't runwon't run
DeepSeek V4 Flash 0731won't runwon't run
DeepSeek V4 Flash Vision Expwon't runwon't run
Devstral 2 123B Instruct 2512won't runwon't run
GLM 4.5Vwon't runwon't run
GLM 4.6Vwon't runwon't run
GLM 4.7won't runwon't run
GLM 5.1won't runwon't run
GLM 5.2won't runwon't run
GLM 5.3won't runwon't run
Kimi K2won't runwon't run
Kimi K2 Instruct 0905won't runwon't run
Kimi K2 Thinkingwon't runwon't run
Kimi K2.5won't runwon't run
MiniMax M2won't runwon't run
MiniMax M2.1won't runwon't run
MiniMax M2.5won't runwon't run
MiniMax M3won't runwon't run
Mistral Medium 3.5 128Bwon't runwon't run

Questions

Which is better for local AI, the NVIDIA GeForce RTX 3090 or the NVIDIA GeForce RTX 4090?
Choose the NVIDIA GeForce RTX 4090 for speed. Choose the NVIDIA GeForce RTX 3090 to spend less.
Which can run bigger models, the NVIDIA GeForce RTX 3090 or the NVIDIA GeForce RTX 4090?
At 8k context both run the same 51 models we track fully in memory.
Which is faster for local AI, the NVIDIA GeForce RTX 3090 or the NVIDIA GeForce RTX 4090?
On the models both run fully in memory, the NVIDIA GeForce RTX 4090 is about 1.1× as fast as the NVIDIA GeForce RTX 3090 (estimated, median across models).
What is the price difference between the NVIDIA GeForce RTX 3090 and the NVIDIA GeForce RTX 4090?
In the UK the NVIDIA GeForce RTX 3090 costs £1,707 at Amazon and the NVIDIA GeForce RTX 4090 £6,015 at Amazon: the NVIDIA GeForce RTX 3090 is £4,308 cheaper.

More comparisons

NVIDIA GeForce RTX 3090 vs NVIDIA GeForce RTX 4090 for local AI (UK) · readygpu