Can it run?
Runs well
Using Q5_K_M in Ollama
Fit status
Runs well
Decode
122.7 tok/s
TTFT
1578 ms
Safe context
26K
Memory
5.0 GB / 8.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | B | Runs well | 122.7 tok/s | 2295 ms | 50K |
| Chat | B | Runs well | 122.7 tok/s | 861 ms | 13K |
| Coding | B | Runs well | 122.7 tok/s | 1578 ms | 26K |
| RAG | B | Runs well | 122.7 tok/s | 2869 ms | 50K |
| Reasoning | B | Runs well | 122.7 tok/s | 1865 ms | 26K |
How Llama 3.2 3B Instruct (3B params) fits at each quantization level on RTX 2060 Super 8GB (8.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 1.2 GB | Low | D33 |
Q3_K_S | 3 | 1.5 GB | Low | D34 |
NVFP4 | 4 | 1.7 GB | Medium | D34 |
Q4_K_M | 4 | 1.8 GB | Medium | D34 |
Q5_K_M | 5 | 2.2 GB | High | D35 |
Q6_K | 6 | 2.5 GB | High | D36 |
Q8_0Best for your GPU | 8 | 3.2 GB | Very High | D38 |
F16 | 16 | 6.1 GB | Maximum | C45 |
huggingface-cli download hf-bartowski--llama-3-2-3b-instruct-gguf