GPUs & Macs › NVIDIA RTX 4060 Ti
NVIDIA RTX 4060 Ti: local LLM speed
Llama 2 7B (Q4_0) generates 64.0 tokens/s on the NVIDIA RTX 4060 Ti (CUDA), and processes prompts at 3,803 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 64.0 | 3,803 | Q4_0 | CUDA FA on | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 63.9 | 3,395 | Q4_0 | CUDA FA off | tg128 | llama.cpp #15013 details |
| Llama 3.2 1B Instruct | 167 | 8,581 | Q4_K_M | llamafile | LocalScore avg · n=2 | LocalScore LocalScore details |
| Llama 3.1 8B Instruct | 43.3 | 2,160 | Q4_K_M | llamafile | LocalScore avg · n=2 | LocalScore LocalScore details |
| Qwen2.5 14B Instruct | 14.6 | 783 | Q4_K_M | llamafile | LocalScore avg · n=2 · part CPU | LocalScore LocalScore details |
What fits on the NVIDIA RTX 4060 Ti
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 64.0 tok/s (CUDA, source). Fit assumes the whole model in GPU memory with ~8K context.
| Model | Best quant that fits | File | Expected tok/s |
|---|---|---|---|
| gemma-4-12B-it | IQ4_XS | 6.4 GB | 38.4 estimate |
| Qwen3.5-9B | Q5_K_S | 6.4 GB | 38.5 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 54.6 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q4_K_M | 5.8 GB | 41.9 estimate |
| gemma-4-E4B-it | Q5_K_M | 5.5 GB | 44.7 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 57.2 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 204 estimate |
| gemma-4-E2B-it | Q8_0 | 5.0 GB | 49.3 estimate |
| DeepSeek-R1-0528-Qwen3-8B | Q5_K_M | 5.9 GB | 41.8 estimate |
| Llama-3.2-3B-Instruct | Q8_0 | 3.4 GB | 71.5 estimate |
| Phi-4-mini-instruct | Q8_0 | 4.1 GB | 59.9 estimate |
| gemma-3-4b-it | Q8_0 | 4.1 GB | 59.3 estimate |
| Qwen3-14B | UD-IQ3_XXS | 6.0 GB | 40.7 estimate |
| gemma-3-12b-it | Q3_K_M | 6.0 GB | 40.7 estimate |
Similar hardware
NVIDIA RTX 2070 SUPER · NVIDIA RTX 2060 SUPER · NVIDIA GTX 1070 Ti · NVIDIA RTX 3060 Ti · NVIDIA RTX 3070 · NVIDIA GTX 1070 · NVIDIA RTX 3070 Ti · NVIDIA RTX 2080 SUPER