GPUs & Macs › NVIDIA RTX 4080
NVIDIA RTX 4080: local LLM speed
Llama 2 7B (Q4_0) generates 143 tokens/s on the NVIDIA RTX 4080 (CUDA), and processes prompts at 9,206 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 143 | 9,206 | Q4_0 | CUDA FA on | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 142 | 8,032 | Q4_0 | CUDA FA off | tg128 | llama.cpp #15013 details |
| Llama 3.2 1B Instruct | 241 | 14,496 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
| Llama 3.1 8B Instruct | 88.3 | 4,792 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
| Qwen2.5 14B Instruct | 35.2 | 2,339 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
What fits on the NVIDIA RTX 4080
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 143 tok/s (CUDA, source). Fit assumes the whole model in GPU memory with ~8K context.
| Model | Best quant that fits | File | Expected tok/s |
|---|---|---|---|
| Qwen3.8-27B | UD-IQ3_XXS | 10.9 GB | 50.2 estimate |
| gemma-4-12B-it | Q8_0 | 12.7 GB | 43.3 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 57.6 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 122 estimate |
| Qwen3.6-35B-A3B | UD-IQ3_XXS | 13.2 GB | 242 estimate |
| gemma-4-26B-A4B-it | UD-IQ4_XS | 13.6 GB | 131 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 57.4 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 66.9 estimate |
| Muse-Glimmer-30B | UD-IQ3_XXS | 13.1 GB | 41.8 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 128 estimate |
| gemma-4-31B-it | UD-IQ3_XXS | 11.8 GB | 46.3 estimate |
| gemma-4-12B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 6.7 GB | 81.7 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 457 estimate |
| gemma-4-E2B-it | Q8_0 | 5.0 GB | 110 estimate |
Similar hardware
NVIDIA RTX 5080 · NVIDIA RTX 5070 Ti · NVIDIA RTX 4080 SUPER · NVIDIA RTX 4070 Ti SUPER · NVIDIA RTX 5060 Ti · NVIDIA RTX PRO 2000 Blackwell · NVIDIA Quadro P5000 · NVIDIA Quadro RTX 5000