How we got this number
Each generated token reads the whole model from memory once: 9.9 GB. The NVIDIA DGX Spark (128GB) moves 273 GB/s, so the ceiling is 28 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 starCoder 2 15B
| Mac Studio M5 Ultra (96GB) | 99 tok/s | Check price → |
| Mac Studio M5 Ultra (256GB) | 99 tok/s | Check price → |
| Mac Studio M5 Max (40-core GPU, 64GB) | 50 tok/s | Check price → |
| Mac mini M5 Pro (24GB) | 25 tok/s | Check price → |
Other models on the NVIDIA DGX Spark (128GB)
| lfm 2.5 8B (A1B) | Fast | 120+ tok/s |
| Llama 3.2 3B | Fast | 96 tok/s |
| orca mini 3B | Fast | 96 tok/s |
| Granite 3.1 moe 3B | Fast | 96 tok/s |
| falcon 3 3B | Fast | 96 tok/s |
| Granite 4.1 3B | Fast | 91 tok/s |