RTX 3090 vs RTX 5070

Best sourced output tok/s per model (single-stream generation) — engine and version, quantization, context and a source link on every row. Hover a value for the quant.

RTX 309018 sourced runs · best 709.5 tok/s

RTX 50705 verified runs · best 185.8 tok/s

Specs and evidence

	RTX 3090	RTX 5070
VRAM	24 GB	12 GB
Sourced runs on the board	18	5
Models measured	11	5
Best output tok/s	709.5 tok/s	185.8 tok/s
Highest measured peak VRAM	22.0 GB	9.0 GB
Street price (latest snapshot)	$1,720 used	—

Sourced rows only. Peak VRAM is recorded by the person who ran the benchmark, so it is a floor for what the card can hold, not a specification. Price snapshots.

Head-to-head

On the 2 models measured on both cards, RTX 3090 leads on 1, RTX 5070 on 1. The widest gap is on Qwen3.8-27B: RTX 3090 164.6 tok/s vs RTX 5070 48.3 tok/s (RTX 3090 241% faster).

The measured matrix

Model	RTX 3090	RTX 5070
LFM2.5-1.2B-Instruct	709.5 Q4_K_M	—
Qwen3.5-0.8B-MTP	476.1 Q6_K_XL	—
NVIDIA-Nemotron-3.5-Lightning-30B-A3B	353.7 NVFP4	—
LFM2.5-2.6B	305.6 Q4_K_M	—
Qwen3.5-4B-MTP	240.4 Q6_K_XL	—
NVIDIA-Nemotron-3-Nano-4B	—	185.8 Q4_K_M
Apodex-1.1-mini	—	182.3 IQ1_M
Qwen3.8-27B	164.6 EXL3-3.5bpw	48.3 UD-Q2_K_XL
Qwen3.5-9B-MTP	164.3 Q6_K_XL	—
Ornith-1.5-9B	—	102.0 Q4_K_M
Qwen3.8-Flash-Next	38.6 EXL3 2.50bpw	81.8 Q2_0
Ornstein3.6-27B-MTP-NSC-ACE-SABER	32.9 Q4_K_M	—
GLM-5.3-Flash	28.1 EXL3 3.05 bpw	—
DeepSeek-V4.1-Flash	26.0 EXL3 3.0 bpw	—

Head-to-head (models tested on both): RTX 3090 1 · RTX 5070 1 · tied 0 · data as of Oct 30, 2026. Full rules: methodology.

Related reading: RTX 5090 vs 3090 · quantization · Value index · Full board