Can it run?
Runs well
Using Q4_K_M in Ollama
Fit status
Runs well
Decode
130.4 tok/s
TTFT
1484 ms
Safe context
86K
Memory
34.2 GB / 184.3 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 130.4 tok/s | 2159 ms | 167K |
| Chat | C | Runs well | 130.4 tok/s | 810 ms | 43K |
| Coding | C | Runs well | 130.4 tok/s | 1484 ms | 86K |
| RAG | C | Runs well | 130.4 tok/s | 2699 ms | 167K |
| Reasoning | C | Runs well | 130.4 tok/s | 1754 ms | 86K |
How Llama 2 7B Chat (7B params) fits at each quantization level on Mac Studio M3 Ultra 256GB (184.3 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 2.7 GB | Low | D30 |
Q3_K_S | 3 | 3.4 GB | Low | D30 |
NVFP4 | 4 | 3.9 GB | Medium | D30 |
Q4_K_M | 4 | 4.3 GB | Medium | D30 |
Q5_K_M | 5 | 5.0 GB | High | D30 |
Q6_K | 6 | 5.7 GB | High | D30 |
Q8_0 | 8 | 7.5 GB | Very High | D30 |
F16Best for your GPU | 16 | 14.3 GB | Maximum | D31 |
