GPUs & Macs › NVIDIA Jetson AGX Orin 64GB
NVIDIA Jetson AGX Orin 64GB: local LLM speed
Llama 2 7B (Q4_0) generates 35.9 tokens/s on the NVIDIA Jetson AGX Orin 64GB (CUDA), and processes prompts at 1,172 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 35.9 | 1,172 | Q4_0 | CUDA FA on | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 33.6 | 991 | Q4_0 | CUDA FA off | tg128 | llama.cpp #15013 details |
| Llama 3.2 1B Instruct | 45.3 | 1,157 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
What fits on the NVIDIA Jetson AGX Orin 64GB
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 35.9 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 | 4.7 estimate |
| gemma-4-12B-it | Q8_0 | 12.7 GB | 10.8 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 14.4 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 30.6 estimate |
| Qwen3.6-35B-A3B | Q8_0 | 36.9 GB | 21.7 estimate |
| gemma-4-26B-A4B-it | Q8_0 | 26.9 GB | 16.6 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 14.4 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 16.7 estimate |
| Muse-Glimmer-30B | Q8_0 | 29.6 GB | 4.6 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 32.1 estimate |
| gemma-4-31B-it | Q8_0 | 32.6 GB | 4.2 estimate |
| gemma-4-26B-A4B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 14.2 GB | 31.3 estimate |
| gemma-4-12B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 6.7 GB | 20.4 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 114 estimate |