GPUs & Macs › NVIDIA RTX PRO 4000 Blackwell
NVIDIA RTX PRO 4000 Blackwell: local LLM speed
Llama 2 7B (Q4_0) generates 136 tokens/s on the NVIDIA RTX PRO 4000 Blackwell (CUDA), and processes prompts at 5,674 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 136 | 5,674 | Q4_0 | CUDA FA on | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 132 | 4,993 | Q4_0 | CUDA FA off | tg128 | llama.cpp #15013 details |
What fits on the NVIDIA RTX PRO 4000 Blackwell
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 136 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-Q6_K | 22.0 GB | 23.7 estimate |
| gemma-4-12B-it | Q8_0 | 12.7 GB | 41.1 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 54.7 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 116 estimate |
| Qwen3.6-35B-A3B | UD-Q4_K_S | 20.9 GB | 146 estimate |
| gemma-4-26B-A4B-it | UD-Q5_K_M | 21.2 GB | 80.1 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 54.6 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 63.6 estimate |
| Muse-Glimmer-30B | UD-Q5_K_M | 19.2 GB | 27.2 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 122 estimate |
| gemma-4-31B-it | Q5_K_M | 21.7 GB | 24.1 estimate |
| gemma-4-26B-A4B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 14.2 GB | 119 estimate |
| gemma-4-12B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 6.7 GB | 77.6 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 434 estimate |
Similar hardware
NVIDIA RTX A5000 · NVIDIA Quadro RTX 6000 · NVIDIA RTX A4500 · NVIDIA RTX 4000 Ada · NVIDIA RTX 4000 SFF Ada · NVIDIA RTX A4000 · NVIDIA RTX 2000 Ada · NVIDIA RTX PRO 4500 Blackwell