Can RTX 4090 run Phi-4 14B?
Comfortable: GGUF Q5_K_M needs 11.8 GB at 4K, batch 1; headroom 12.2 GB.
Text input, FP16 KV cache, one card. Unified-memory capacity uses the GPU guide’s usable limit; negative headroom may require host offload. Speed source settings may differ from this 4K estimate.
Measured runs
| Framework | Quant | tok/s | VRAM used (GB) |
|---|---|---|---|
| llama.cpp | GGUF Q4_K_M | 88 | 9.1 |
Quant memory at 4K
| Quant | Total GB | Verdict | Weight sizing |
|---|---|---|---|
| GGUF Q4_K_M | 10.2 | Comfortable | Estimated weights |
| GGUF Q5_K_M | 11.8 | Comfortable | Estimated weights |
| AWQ INT4 | 8.5 | Comfortable | Estimated weights |
Context capacity · GGUF Q5_K_M
| Context | Total GB | Verdict |
|---|---|---|
| 4K | 11.8 | Comfortable |
| 16K | 14.3 | Comfortable |
Generation speed
Measured: 88 tok/s, GGUF Q4_K_M, llama.cpp
GGUF Q5_K_M: 68–89 tok/s · Derived range