GPUs & Macs › NVIDIA RTX A4500
NVIDIA RTX A4500: local LLM speed
Llama 2 7B (Q4_0) generates 103 tokens/s on the NVIDIA RTX A4500 (CUDA), and processes prompts at 3,453 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 103 | 3,453 | Q4_0 | CUDA FA on | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 97.3 | 2,827 | Q4_0 | CUDA FA off | tg128 | llama.cpp #15013 details |
| Llama 3.2 1B Instruct | 286 | 10,422 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
| Llama 3.1 8B Instruct | 77.9 | 2,498 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
| Qwen2.5 14B Instruct | 44.0 | 1,331 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
What fits on the NVIDIA RTX A4500
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 103 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-Q4_K_M | 16.5 GB | 23.9 estimate |
| gemma-4-12B-it | Q8_0 | 12.7 GB | 31.1 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 41.3 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 87.8 estimate |
| Qwen3.6-35B-A3B | UD-IQ4_XS | 17.7 GB | 130 estimate |
| gemma-4-26B-A4B-it | UD-Q4_K_M | 16.9 GB | 75.5 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 41.2 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 48.1 estimate |
| Muse-Glimmer-30B | UD-IQ3_M | 14.1 GB | 27.9 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 92.0 estimate |
| gemma-4-31B-it | Q4_K_S | 17.4 GB | 22.6 estimate |
| gemma-4-26B-A4B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 14.2 GB | 89.8 estimate |
| gemma-4-12B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 6.7 GB | 58.6 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 328 estimate |
Similar hardware
NVIDIA RTX 4000 Ada · NVIDIA RTX 4000 SFF Ada · NVIDIA RTX PRO 4000 Blackwell · NVIDIA RTX A5000 · NVIDIA Quadro RTX 6000 · NVIDIA RTX A4000 · NVIDIA RTX 2000 Ada · NVIDIA RTX A2000 12GB