readygpu

NVIDIA GeForce RTX 3090 vs NVIDIA GeForce RTX 5070

For running models locally, priced in the US. At 128k context the first runs 25 models fully in memory, the second 11.

In short

The NVIDIA GeForce RTX 3090 has more memory for models: 24 GB against 12 GB. At 128k context the NVIDIA GeForce RTX 3090 runs 25 of the models we track fully in memory and the NVIDIA GeForce RTX 5070 11; only the NVIDIA GeForce RTX 3090 holds Gemma 4 26B-A4B. On the models both run fully in memory, the NVIDIA GeForce RTX 3090 is about 1.4× as fast as the NVIDIA GeForce RTX 5070 (estimated, median across models). In the US the NVIDIA GeForce RTX 3090 costs $1,600 at eBay and the NVIDIA GeForce RTX 5070 $750 at eBay: the NVIDIA GeForce RTX 5070 is $850 cheaper.

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

NVIDIA GeForce RTX 3090

$1,600

eBay

Add to build
NVIDIA GeForce RTX 5070

$750

eBay

Add to build

Specifications

SpecNVIDIA GeForce RTX 3090NVIDIA GeForce RTX 5070
GPUGeForce RTX 3090GeForce RTX 5070
Memory24 GB GDDR6X12 GB GDDR7
Memory bus384-bit192-bit
Memory bandwidth936 GB/s672 GB/s
FP16 compute71 TFLOPS62 TFLOPS
SoftwareCUDACUDA
Board power350 W250 W
PCIe generationGen 4Gen 5
PCIe lanesx16x16

What each runs

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

ModelNVIDIA GeForce RTX 3090NVIDIA GeForce RTX 5070
Gemma 4 26B-A4B59tok/sQ5_K_Moffload
gpt-oss 20B62tok/sMXFP4offload
Gemma 4 12B38tok/sQ8_043tok/sQ4_K_M
Qwen 3.5 9B42tok/sQ8_042tok/sQ4_K_M
Qwen 3.5 4B66tok/sQ8_047tok/sQ8_0
Ministral 3 8B25tok/sQ4_K_Moffload
Gemma 4 E4B48tok/sQ8_045tok/sQ5_K_M
LFM2.5 8B-A1B46tok/sQ8_0offload
Llama 3.1 8B25tok/sQ5_K_Moffload
Qwen2.5 0.5B271tok/sQ8_0195tok/sQ8_0
Qwen3.5 0.8B238tok/sQ8_0171tok/sQ8_0
Qwen3.5 2B189tok/sQ5_K_M136tok/sQ5_K_M
DeepSeek R1 Distill Qwen 1.5B103tok/sQ8_074tok/sQ8_0
Mellum2.1 12B A2.5B Thinking76tok/sQ8_0offload
MiniCPM5 2B70tok/sQ8_050tok/sQ8_0
LFM2.5 2.6B53tok/sQ8_041tok/sQ6_K
LFM2.5 VL 3B53tok/sQ8_041tok/sQ6_K
Kimi VL A3B48tok/smlx-8bitoffload
granite vision 4.1 4b40tok/sQ8_0offload
granite 4.2 3b40tok/sQ8_0offload
GLM 4.6V Flash38tok/sQ8_0offload
DeepSeek R1 Distill Qwen 7B37tok/sQ8_0offload
GLM 4.1V 9B Thinking34tok/smlx-8bitoffload
DeepSeek R1 Distill Llama 8B25tok/sQ5_K_Moffload
Ministral 3 8B Reasoning 251225tok/sQ4_K_Moffload
Qwen 3.8 27Boffloadoffload
Qwen 3.5 122B-A10Boffloadwon't run
Qwen 3.6 27Boffloadoffload
Gemma 4 31Boffloadoffload
Mistral Small 4 119Boffloadwon't run
Qwen 3.6 35B-A3Boffloadoffload
Qwen 3.5 27Boffloadoffload
Qwen 3.5 35B-A3Boffloadoffload
GLM 4.7 Flashoffloadoffload
Devstral Small 2 24Boffloadoffload
Nemotron 3.5 Lightning 30B-A3Boffloadoffload
Kimi Linear 48B-A3Boffloadoffload
Qwen3 Coder 30B-A3Boffloadoffload
Gemma 3 27Boffloadoffload
Mistral Small 3.2 24Boffloadoffload
Ministral 3 14Boffloadoffload
Qwen3 14Boffloadoffload
DeepSeek R1 0528 Qwen3 8Boffloadoffload
Qwen3 8Boffloadoffload
Phi-4 14Boffloadoffload
Granite 4.2 8Boffloadoffload
LFM2.5 24B A2Boffloadoffload
DeepSeek R1 Distill Qwen 14Boffloadoffload
Devstral Small 2507offloadoffload
Magistral Small 2506offloadoffload
Magistral Small 2507offloadoffload
Magistral Small 2509offloadoffload
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
Qwen3 Coder Nextwon't runwon't run
gpt-oss 120Bwon't runwon't run
GLM 4.5 Airwon't runwon't run
Qwen3 Next 80B-A3Bwon't runwon't run
Qwen3 32Bwon't runwon't run
DeepSeek R1 Distill Qwen 32Bwon't runwon't run
Llama 3.3 70Bwon't runwon't run
DeepSeek R1won't runwon't run
DeepSeek R1 0528won't runwon't run
DeepSeek R1 Distill Llama 70Bwon'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
granite 4.2 30bwon't runwon't run

Questions

Which is better for local AI, the NVIDIA GeForce RTX 3090 or the NVIDIA GeForce RTX 5070?
Choose the NVIDIA GeForce RTX 3090 to run bigger models. Choose the NVIDIA GeForce RTX 3090 for speed. Choose the NVIDIA GeForce RTX 5070 to spend less.
Which can run bigger models, the NVIDIA GeForce RTX 3090 or the NVIDIA GeForce RTX 5070?
At 128k context the NVIDIA GeForce RTX 3090 runs 25 of the models we track fully in memory and the NVIDIA GeForce RTX 5070 11; only the NVIDIA GeForce RTX 3090 holds Gemma 4 26B-A4B.
Which is faster for local AI, the NVIDIA GeForce RTX 3090 or the NVIDIA GeForce RTX 5070?
On the models both run fully in memory, the NVIDIA GeForce RTX 3090 is about 1.4× as fast as the NVIDIA GeForce RTX 5070 (estimated, median across models).
What is the price difference between the NVIDIA GeForce RTX 3090 and the NVIDIA GeForce RTX 5070?
In the US the NVIDIA GeForce RTX 3090 costs $1,600 at eBay and the NVIDIA GeForce RTX 5070 $750 at eBay: the NVIDIA GeForce RTX 5070 is $850 cheaper.

More comparisons