GPUs & Macs › NVIDIA Quadro RTX 4000
NVIDIA Quadro RTX 4000: local LLM speed
Llama 2 7B (Q4_0) generates 67.6 tokens/s on the NVIDIA Quadro RTX 4000 (CUDA), and processes prompts at 1,663 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 67.6 | 1,663 | Q4_0 | CUDA FA on | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 65.6 | 1,537 | Q4_0 | CUDA FA off | tg128 | llama.cpp #15013 details |
| Llama 3.1 8B Instruct | 40.0 | 992 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
What fits on the NVIDIA Quadro RTX 4000
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 67.6 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 | 40.5 estimate |
| Qwen3.5-9B | Q5_K_S | 6.4 GB | 40.6 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 57.7 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q4_K_M | 5.8 GB | 44.3 estimate |
| gemma-4-E4B-it | Q5_K_M | 5.5 GB | 47.2 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 60.4 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 215 estimate |
| gemma-4-E2B-it | Q8_0 | 5.0 GB | 52.0 estimate |
| DeepSeek-R1-0528-Qwen3-8B | Q5_K_M | 5.9 GB | 44.2 estimate |
| Llama-3.2-3B-Instruct | Q8_0 | 3.4 GB | 75.5 estimate |
| Phi-4-mini-instruct | Q8_0 | 4.1 GB | 63.3 estimate |
| gemma-3-4b-it | Q8_0 | 4.1 GB | 62.6 estimate |
| Qwen3-14B | UD-IQ3_XXS | 6.0 GB | 43.0 estimate |
| gemma-3-12b-it | Q3_K_M | 6.0 GB | 43.0 estimate |
Similar hardware
NVIDIA RTX A2000 · NVIDIA Quadro P2000 · NVIDIA T1000 · NVIDIA Quadro P1000 · NVIDIA RTX A2000 12GB · NVIDIA RTX A4000 · NVIDIA RTX 2000 Ada · NVIDIA RTX A4500