GPUs & Macs › NVIDIA RTX 2000 Ada
NVIDIA RTX 2000 Ada: local LLM speed
Llama 2 7B (Q4_0) generates 50.7 tokens/s on the NVIDIA RTX 2000 Ada (CUDA), and processes prompts at 2,250 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 50.7 | 2,250 | Q4_0 | CUDA FA on | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 50.6 | 1,956 | Q4_0 | CUDA FA off | tg128 | llama.cpp #15013 details |
| Llama 3.2 1B Instruct | 168 | 6,087 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
| Llama 3.1 8B Instruct | 37.5 | 1,357 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
| Qwen2.5 14B Instruct | 20.7 | 740 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
What fits on the NVIDIA RTX 2000 Ada
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 50.7 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-IQ3_XXS | 10.9 GB | 17.7 estimate |
| gemma-4-12B-it | Q8_0 | 12.7 GB | 15.3 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 20.3 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 43.2 estimate |
| Qwen3.6-35B-A3B | UD-IQ3_XXS | 13.2 GB | 85.6 estimate |
| gemma-4-26B-A4B-it | UD-IQ4_XS | 13.6 GB | 46.3 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 20.3 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 23.7 estimate |
| Muse-Glimmer-30B | UD-IQ3_XXS | 13.1 GB | 14.8 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 45.3 estimate |
| gemma-4-31B-it | UD-IQ3_XXS | 11.8 GB | 16.4 estimate |
| gemma-4-12B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 6.7 GB | 28.9 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 162 estimate |
| gemma-4-E2B-it | Q8_0 | 5.0 GB | 39.0 estimate |
Similar hardware
NVIDIA RTX A4000 · NVIDIA RTX A4500 · NVIDIA RTX 4000 Ada · NVIDIA RTX 4000 SFF Ada · NVIDIA RTX A2000 12GB · NVIDIA RTX PRO 4000 Blackwell · NVIDIA RTX A5000 · NVIDIA Quadro RTX 6000