Can the NVIDIA RTX 4090 run Phi-4 14B?
Yes
QuantQ8_0Near-lossless
Memory17.9 GBof 24 GB
Speed~39tokens/s
Yes. The NVIDIA RTX 4090 can run Phi-4 14B fully in VRAM at Q8_0 (near-lossless), using about 17.9 GB of the 24 GB available, at roughly 39 tokens/second.
Try other context lengths in the calculator →Memory needed by quantization
At 8,192 tokens of context.
| Quant | Quality | Memory | Fits? | Speed |
|---|---|---|---|---|
| FP16 | Full quality | 31.7 GB | Too big | — |
| Q8_0 | Near-lossless | 17.9 GB | Fits | ~39 tok/s |
| Q6_K | Excellent | 14.3 GB | Fits | ~50 tok/s |
| Q5_K_M | Very good | 12.7 GB | Fits | ~58 tok/s |
| Q4_K_M | Good, most popular | 11.2 GB | Fits | ~68 tok/s |
| Q3_K_M | Noticeably degraded | 9.5 GB | Fits | ~84 tok/s |
| Q2_K | Heavily degraded | 7.8 GB | Fits | ~110 tok/s |
How to run Phi-4 14B on the NVIDIA RTX 4090
With Ollama installed, one command downloads and starts it:
ollama run phi4:14bOllama's default download is Q4_K_M (about 11.2 GB here). You have room for Q8_0, which is near-lossless: pick that GGUF in LM Studio or from Hugging Face.
GPUs that run Phi-4 14B comfortably
- NVIDIA RTX 3060 12GB
- NVIDIA RTX 4070 (12 GB)
- NVIDIA RTX 5070 (12 GB)
- NVIDIA RTX 4060 Ti 16GB
- NVIDIA RTX 5060 Ti 16GB
Models that run well on the NVIDIA RTX 4090
Run it in the cloud
Don't want to buy hardware? Run Phi-4 14B on a rented GPU today.
Rent a cloud GPU by the hour. No upfront cost, cancel any time.