Phi-4 14B: VRAM needs and speed by GPU
At Q4_K_M, Phi-4 14B needs about 10.1 GB of VRAM. We have 1 measured speed results for it.
14.66B parameters · license: mit · GGUF files on Hugging Face
Measured speeds
| Chip | Generation tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Intel Arc Pro B70 | 43.6 | 1,424 | Q4_K_M | SYCL FA on | tg128 | PMZFX intel-arc-pro-b70-benchmarks (MIT) details |
Memory needed per quantization
| Quant | File size | Needs (estimate) | Smallest card it fits |
|---|---|---|---|
| Q2_K | 5.6 GB | 6.8 GB | 8 GB cards |
| Q2_K_L | 5.7 GB | 6.9 GB | 8 GB cards |
| Q3_K_M | 7.2 GB | 8.4 GB | 12 GB cards |
| Q4_K_M | 8.9 GB | 10.1 GB | 12 GB cards |
| Q5_K_M | 10.4 GB | 11.6 GB | 16 GB cards |
| Q6_K | 12.0 GB | 13.2 GB | 16 GB cards |
| Q8_0 | 15.6 GB | 16.8 GB | 24 GB cards |
| F16 | 29.3 GB | 30.5 GB | 32 GB cards |
Expected speed on popular hardware
| Hardware | Best quant that fits | Expected tok/s |
|---|---|---|
| NVIDIA RTX 3060 | Q4_K_M | 34.7 estimate |
| NVIDIA RTX 4060 Ti | Q2_K | 43.6 estimate |
| NVIDIA RTX 3090 | Q8_0 | 42.1 estimate |
| NVIDIA RTX 4090 | Q8_0 | 55.9 estimate |
| NVIDIA RTX 5090 | Q8_0 | 73.7 estimate |
| AMD Radeon RX 7900 XTX | Q8_0 | 46.8 estimate |
| Apple M4 Pro (20-core GPU) (64 GB) | Q8_0 | 14.9 estimate |
| Apple M3 Max (40-core GPU) (64 GB) | Q8_0 | 21.7 estimate |
| AMD Ryzen AI Max+ 395 (Radeon 8060S) | Q8_0 | 13.3 estimate |
Too big for your card?
Run Phi-4 14B on a rented 24-80 GB GPU by the hour.
Rent a bigger GPU on RunPodAffiliate link: if you sign up, RigBench may earn a referral credit. It never affects rankings or numbers.