Can it run?
Runs well
Using Q6_K in Ollama
Fit status
Runs well
Decode
559.0 tok/s
TTFT
350 ms
Safe context
87K
Memory
12.8 GB / 69.1 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 559.0 tok/s | 504 ms | 173K |
| Chat | C | Runs well | 584.8 tok/s | 350 ms | 43K |
| Coding | C | Runs well | 559.0 tok/s | 350 ms | 87K |
| RAG | C | Runs well | 584.8 tok/s | 602 ms | 173K |
| Reasoning | C | Runs well | 584.8 tok/s | 391 ms | 87K |
How SmolVLM 500M Instruct (0.5B params) fits at each quantization level on Mac Studio M3 Ultra 96GB (69.1 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 0.2 GB | Low | D30 |
Q3_K_S | 3 | 0.2 GB | Low | D30 |
NVFP4 | 4 |
huggingface-cli download hf-ggml-org--smolvlm-500m-instruct-gguf0.3 GB |
| Medium |
| D30 |
Q4_K_M | 4 | 0.3 GB | Medium | D30 |
Q5_K_M | 5 | 0.4 GB | High | D30 |
Q6_K | 6 | 0.4 GB | High | D30 |
Q8_0 | 8 | 0.5 GB | Very High | D30 |
F16Best for your GPU | 16 | 1.0 GB | Maximum | D30 |