GPUs & Macs › NVIDIA RTX 4070
NVIDIA RTX 4070: local LLM speed
Llama 2 7B (Q4_0) generates 92.3 tokens/s on the NVIDIA RTX 4070 (Vulkan), and processes prompts at 3,179 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 92.3 | 3,179 | Q4_0 | Vulkan FA off | tg128 | llama.cpp #10879 details |
| Llama 2 7B | 91.5 | 4,294 | Q4_0 | Vulkan FA on | tg128 | llama.cpp #10879 details |
| Llama 3.2 1B Instruct | 283 | 12,717 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
| Llama 3.1 8B Instruct | 76.3 | 3,192 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
| Qwen2.5 14B Instruct | 37.9 | 1,692 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
What fits on the NVIDIA RTX 4070
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 92.3 tok/s (Vulkan, 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-IQ2_XXS | 7.3 GB | 48.6 estimate |
| gemma-4-12B-it | Q6_K | 9.8 GB | 36.1 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 37.0 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 78.7 estimate |
| gemma-4-26B-A4B-it | UD-IQ2_M | 10.0 GB | 115 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 37.0 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 43.1 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 82.4 estimate |
| gemma-4-31B-it | UD-IQ2_XXS | 8.5 GB | 41.3 estimate |
| gemma-4-12B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 6.7 GB | 52.5 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 294 estimate |
| gemma-4-E2B-it | Q8_0 | 5.0 GB | 71.0 estimate |
| DeepSeek-R1-0528-Qwen3-8B | Q8_0 | 8.7 GB | 40.5 estimate |
| Llama-3.2-3B-Instruct | Q8_0 | 3.4 GB | 103 estimate |
Similar hardware
NVIDIA RTX 5070 · NVIDIA TITAN V · NVIDIA TITAN Xp · NVIDIA RTX 3060 · NVIDIA RTX 4070 Ti · NVIDIA RTX 4070 SUPER · NVIDIA RTX 3080 Ti · NVIDIA RTX 2080 Ti