Phi-3.5 Mini 3.8B hardware requirements
3.82B parameters. Needs about 6.1 GB at Q4_K_M with 8K context. Model card.
Memory needed at each quantization
| Quant | Quality | 8K context | 32K context |
|---|---|---|---|
| FP16 | Full quality | 11.5 GB | 21.1 GB |
| Q8_0 | Near-lossless | 7.9 GB | 17.5 GB |
| Q6_K | Excellent | 7 GB | 16.6 GB |
| Q5_K_M | Very good | 6.5 GB | 16.2 GB |
| Q4_K_M | Good, most popular | 6.1 GB | 15.8 GB |
| Q3_K_M | Noticeably degraded | 5.7 GB | 15.4 GB |
| Q2_K | Heavily degraded | 5.3 GB | 14.9 GB |
Run it with Ollama: ollama run phi3.5:3.8b (downloads Q4_K_M by default).
Run it in the cloud
Don't want to buy hardware? Run Phi-3.5 Mini 3.8B on a rented GPU today.
Rent a cloud GPU by the hour. No upfront cost, cancel any time.
Every GPU and Mac checked
| GPU | Runs? | Best fit |
|---|---|---|
| NVIDIA RTX 3060 12GB | Yes | FP16 · ~11.5 GB · ~28 tok/s |
| NVIDIA RTX 3070 | Yes | Q8_0 · ~7.9 GB · ~66 tok/s |
| NVIDIA RTX 3080 10GB | Yes | Q8_0 · ~7.9 GB · ~112 tok/s |
| NVIDIA RTX 3090 | Yes | FP16 · ~11.5 GB · ~74 tok/s |
| NVIDIA RTX 4060 | Yes | Q8_0 · ~7.9 GB · ~40 tok/s |
| NVIDIA RTX 4060 Ti 16GB | Yes | FP16 · ~11.5 GB · ~23 tok/s |
| NVIDIA RTX 5060 Ti 16GB | Yes | FP16 · ~11.5 GB · ~35 tok/s |
| NVIDIA RTX 4070 | Yes | FP16 · ~11.5 GB · ~40 tok/s |
| NVIDIA RTX 4070 Ti Super | Yes | FP16 · ~11.5 GB · ~53 tok/s |
| NVIDIA RTX 4080 | Yes | FP16 · ~11.5 GB · ~56 tok/s |
| NVIDIA RTX 4090 | Yes | FP16 · ~11.5 GB · ~79 tok/s |
| NVIDIA RTX 5070 | Yes | FP16 · ~11.5 GB · ~53 tok/s |
| NVIDIA RTX 5070 Ti | Yes | FP16 · ~11.5 GB · ~70 tok/s |
| NVIDIA RTX 5080 | Yes | FP16 · ~11.5 GB · ~75 tok/s |
| NVIDIA RTX 5090 | Yes | FP16 · ~11.5 GB · ~141 tok/s |
| AMD Radeon RX 7800 XT | Yes | FP16 · ~11.5 GB · ~49 tok/s |
| AMD Radeon RX 7900 XTX | Yes | FP16 · ~11.5 GB · ~75 tok/s |
| Apple M4 (16GB) | Yes | Q8_0 · ~7.9 GB · ~18 tok/s |
| Apple M1 (16GB) | Yes | Q8_0 · ~7.9 GB · ~10 tok/s |
| Apple M2 Pro (32GB) | Yes | FP16 · ~11.5 GB · ~16 tok/s |
| Apple M3 Max (64GB) | Yes | FP16 · ~11.5 GB · ~31 tok/s |
| Apple M4 Pro (48GB) | Yes | FP16 · ~11.5 GB · ~21 tok/s |
| Apple M4 Max (128GB) | Yes | FP16 · ~11.5 GB · ~43 tok/s |
| Apple M2 Ultra (192GB) | Yes | FP16 · ~11.5 GB · ~63 tok/s |