How we got this number
Each generated token reads the whole model from memory once: 2.2 GB. The NVIDIA DGX Spark (128GB) moves 273 GB/s, so the ceiling is 120+ tokens/sec. Real machines reach a fraction of that ceiling; we use 70%, an assumption until we get measured runs for this kind of machine. Send us your real numbers and we'll replace the estimate.
Faster machines for Phi 3 3.8B
Other models on the NVIDIA DGX Spark (128GB)
| DeepSeek-R1 1.5B | Fast | 120+ tok/s |
| Llama 3.2 1B | Fast | 120+ tok/s |
| Qwen 3 1.7B | Fast | 120+ tok/s |
| Qwen3.5 0.8B | Fast | 120+ tok/s |
| Qwen 1.8B | Fast | 120+ tok/s |
| Granite 3.1 moe 1B | Fast | 120+ tok/s |