GPUs & Macs › NVIDIA RTX A6000
NVIDIA RTX A6000: local LLM speed
Llama 2 7B (Q4_0) generates 145 tokens/s on the NVIDIA RTX A6000 (CUDA), and processes prompts at 5,662 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 145 | 5,662 | Q4_0 | CUDA FA on | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 139 | 4,914 | Q4_0 | CUDA FA off | tg128 | llama.cpp #15013 details |
| Llama 3.2 1B Instruct | 302 | 12,696 | Q4_K_M | llamafile | LocalScore avg · n=2 | LocalScore LocalScore details |
| Llama 3.1 8B Instruct | 90.5 | 3,447 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
| Qwen2.5 14B Instruct | 52.0 | 1,883 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
What fits on the NVIDIA RTX A6000
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 145 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 | Q8_0 | 29.0 GB | 19.1 estimate |
| gemma-4-12B-it | Q8_0 | 12.7 GB | 43.7 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 58.1 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 124 estimate |
| Qwen3.6-35B-A3B | Q8_0 | 36.9 GB | 87.6 estimate |
| gemma-4-26B-A4B-it | Q8_0 | 26.9 GB | 67.0 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 58.0 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 67.6 estimate |
| Muse-Glimmer-30B | Q8_0 | 29.6 GB | 18.7 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 129 estimate |
| gemma-4-31B-it | Q8_0 | 32.6 GB | 17.0 estimate |
| gemma-4-26B-A4B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 14.2 GB | 126 estimate |
| gemma-4-12B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 6.7 GB | 82.5 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 462 estimate |
Similar hardware
NVIDIA RTX 6000 Ada · NVIDIA Quadro RTX 8000 · NVIDIA RTX PRO 4500 Blackwell · NVIDIA RTX PRO 4000 Blackwell · NVIDIA RTX A5000 · NVIDIA Quadro RTX 6000 · NVIDIA RTX A4500 · NVIDIA RTX 4000 Ada