Written by Jakub Rusinowski · Last updated January 6, 2025
Yes — comfortably
Yes, comfortably — Phi-4 (14B) at Q8_0 needs about 17.4 GB of the 24 GB usable on RTX 4090 Desktop (24 GB VRAM, 64 GB RAM), leaving ~6.6 GB spare and running at ~43.4 tok/s (estimated), with room for about 16,384 tokens of context.
Confidence: high · Recommended quantization: Q8_0 · Estimated speed: ~43.4 tok/s
| Usable memory for models | 24 GB |
| Memory bandwidth | 1008 GB/s |
| Form factor | Desktop |
| Operating system | Windows or Linux |
| Memory upgradeable | Yes |
| Quant | Memory needed | Fits 24 GB? | Max context | Est. speed | Download |
|---|---|---|---|---|---|
| F16 | 30.5 GB | ✗ No | — | — | 28 GB |
| Q8_0 | 17.4 GB | ✓ Yes | 16K | ~43.4 tok/s | 14.9 GB |
| Q6_K | 14 GB | ✓ Yes | 16K | ~53.9 tok/s | 11.5 GB |
| Q5_K_M | 12.4 GB | ✓ Yes | 16K | ~60.7 tok/s | 9.9 GB |
| Q4_K_M | 10.9 GB | ✓ Yes | 16K | ~68.8 tok/s | 8.5 GB |
| Q3_K_M | 8.4 GB | ✓ Yes | 16K | ~88.9 tok/s | 6 GB |
| Q2_K | 7.1 GB | ✓ Yes | 16K | ~105.9 tok/s | 4.6 GB |
Ollama or llama.cpp (CUDA/ROCm) — vLLM if you need concurrent serving
Yes, comfortably — Phi-4 (14B) at Q8_0 needs about 17.4 GB of the 24 GB usable on RTX 4090 Desktop (24 GB VRAM, 64 GB RAM), leaving ~6.6 GB spare and running at ~43.4 tok/s (estimated), with room for about 16,384 tokens of context.
Q8_0 — it needs about 17.4 GB of the 24 GB available, downloads as roughly 14.9 GB, and runs at an estimated 43.4 tokens/sec with up to 16K of context.
Nothing binding — the model fits with headroom and generates at a usable speed on this hardware.
Ollama or llama.cpp (CUDA/ROCm) — vLLM if you need concurrent serving
← Can I Run It? | Phi-4 Family model page | Check your hardware