Head to head

RTX 3090 vs RTX 5080

The RTX 3090 holds more — 24 GB against 16 GB — which decides what you can load at all. The RTX 5080 has more bandwidth at 960 GB/s, which decides how fast tokens come out once a model fits.

Model (Q4_K_M, 8K)RTX 3090tok/sRTX 5080tok/s
Llama 3.1 8B Runs comfortably 112 Runs comfortably 121
Llama 3.1 70B Won't fit 1.67 Won't fit 1.35
Llama 3.2 3B Runs comfortably 246 Runs comfortably 266
Llama 3.2 1B Runs comfortably 650 Runs comfortably 700
Llama 4 Scout 109B-A17B Won't fit 5.39 Won't fit 4.91
Qwen3 8B Runs comfortably 108 Runs comfortably 117
Qwen3 14B Runs comfortably 63.5 Runs comfortably 68.7
Qwen3 32B Fits, but tight 29.5 Won't fit 4.98
Qwen3 4B Runs comfortably 196 Runs comfortably 212
Qwen3 30B-A3B Runs comfortably 117 Won't fit 46.2
Qwen3 235B-A22B Won't fit 3.36 Won't fit 3.36
Qwen2.5-Coder 32B Fits, but tight 29.5 Won't fit 4.98
Qwen2.5 7B Runs comfortably 125 Runs comfortably 135
Qwen2.5 72B Won't fit 1.59 Won't fit 1.31
Gemma 3 4B Runs comfortably 209 Runs comfortably 226
Gemma 3 12B Runs comfortably 76.1 Runs comfortably 82.3

Specifications

RTX 3090RTX 5080
Memory24 GB16 GB
Bandwidth936 GB/s960 GB/s
FP16 compute71 TFLOPS112 TFLOPS
Launch price$1,499$999
ArchitectureAmpereBlackwell