Phi-4 14B at Q4_K_M needs about 11.2 GB, more than the 10 GB of the RTX 3080: part of the model runs from system RAM, which slows generation. At Q3_K_M it fits entirely.
Q4_K_M · 8K context
The model needs about 11.2 GB but only 9.7 GB of VRAM is usable: 1.6 GB must be loaded into system RAM. It runs, 86% on the GPU, but more slowly.
~11.2 GB (10.8 GB–11.8 GB) needed · estimate
1.6 GB must be offloaded to system RAM
EstimateData updated 24 days agoSource: Run LLM Locally engine, Hugging Face config.json + model card
What makes this estimate less reliable
| Phi-4 14B Q8_08K~17.6 GB · 2.5–5 tok/s estimate | CPU/GPU offload |
|---|---|
| Phi-4 14B Q6_K8K~14.2 GB · 4.5–8.5 tok/s estimate | CPU/GPU offload |
| Phi-4 14B Q5_K_M8K~12.8 GB · 6–12 tok/s estimate | CPU/GPU offload |
| Phi-4 14B Q4_K_M8K~11.2 GB · 11–21 tok/s estimate | CPU/GPU offload |
| Phi-4 14B IQ4_XS8K~10.2 GB · 23–43 tok/s estimate | CPU/GPU offload |
| Phi-4 14B Q3_K_M8K~9.6 GB · 61–85 tok/s estimate | Compatible with limits |
Prices and links follow the market you choose.
NVIDIA
ExcellentRuns Phi-4 14B Q4_K_M entirely in memory
AMD
ExcellentRuns Phi-4 14B Q4_K_M entirely in memory
Apple Silicon
CompatibleRuns Phi-4 14B Q4_K_M entirely in memory
Unified memory PC
ExcellentRuns Phi-4 14B Q4_K_M entirely in memory
No current offer imported for this market: the link opens a marketplace search.
