← Interactive LLM VRAM Calculator — try your own context length & quantization
How much VRAM to run Phi-4 14B? (FP16/Q8/Q6/Q4 requirements)
Architecture: 40 layers, GQA with 10 KV heads. Figures are approximations for weights + an 8k-token FP16 KV cache + ~1 GB runtime overhead, based on typical published GGUF sizes. Actual usage varies by runtime and settings.
| Quantization | Weights | Total VRAM (8k context) | Fits on (example) |
|---|---|---|---|
| FP16 full precision (16-bit) | ~29 GB | ~32 GB | 2×RTX 3090 or RTX A6000 (48 GB) |
| Q8_0 8-bit | ~16 GB | ~18 GB | RTX 3090 / RTX 4090 (24 GB) |
| Q6_K ~6.6-bit | ~12 GB | ~15 GB | RTX 4060 Ti 16 GB |
| Q5_K_M ~5.7-bit | ~10 GB | ~13 GB | RTX 4060 Ti 16 GB |
| Q4_K_M ~4.8-bit | ~9.0 GB | ~12 GB | RTX 3060 12 GB |
How much VRAM do I need to run Phi-4 14B?
You need roughly ~12 GB of VRAM to run Phi-4 14B at Q4_K_M with an 8k-token context — that fits on RTX 3060 12 GB. At Q8_0 plan for about ~18 GB, and full FP16 needs around ~32 GB. Phi-4 (14.7B parameters) is a strong mid-size option: Q4_K_M fits on a 12 GB RTX 3060, and Q6_K/Q8_0 fit easily on 16–24 GB cards.
What affects these numbers?
Three things: the quantization (bytes per parameter), the KV cache (grows linearly with context length — the table assumes 8k tokens), and runtime overhead. Longer contexts need more memory; quantizing the KV cache (e.g. q8_0 cache in llama.cpp) roughly halves that part. Use the interactive calculator to plug in your exact context length and quantization.