Can it run?
Runs well
Using Q5_K_M in Ollama
Fit status
Runs well
Decode
219.1 tok/s
TTFT
884 ms
Safe context
67K
Memory
11.1 GB / 46.1 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 219.1 tok/s | 1285 ms | 132K |
| Chat | C | Runs well | 219.1 tok/s | 482 ms | 33K |
| Coding | C | Runs well | 219.1 tok/s | 884 ms | 67K |
| RAG | C | Runs well | 219.1 tok/s | 1606 ms | 132K |
| Reasoning | C | Runs well | 219.1 tok/s | 1044 ms | 67K |
How Llama 3.2 3B Instruct (3B params) fits at each quantization level on Mac Studio M2 Ultra 64GB (46.1 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 1.2 GB | Low | D30 |
Q3_K_S | 3 | 1.5 GB | Low | D30 |
NVFP4 | 4 | 1.7 GB | Medium | D30 |
Q4_K_M | 4 | 1.8 GB | Medium | D30 |
Q5_K_M | 5 | 2.2 GB | High | D31 |
Q6_K | 6 | 2.5 GB | High | D31 |
Q8_0 | 8 | 3.2 GB | Very High | D31 |
F16Best for your GPU | 16 | 6.1 GB | Maximum | D32 |
huggingface-cli download hf-bartowski--llama-3-2-3b-instruct-gguf