GeForce RTX 5060 · 8GB GDDR7
Can the GeForce RTX 5060 run Phi-4 (14B)?
Only with CPU offload at Q4_K_M.
Phi-4 (14B) is 14.7B parameters. At Q4_K_M the weights are about 8.2 GiB; with the 1.5 GiB runtime reserve that is 9.7 GiB against the card's 8GB, about 1.7 GiB more than the card has.
All three quants
8GB card| Quant | Weights | With reserve | Verdict | Headroom |
|---|---|---|---|---|
| Q4_K_Mthe everyday quant | 8.2 GiB | 9.7 GiB | Only with CPU offload | short 1.7 GiB |
| Q8_0near-lossless | 14.5 GiB | 16.0 GiB | Does not fit | short 8.0 GiB |
| FP16full weights | 27.4 GiB | 28.9 GiB | Does not fit | short 20.9 GiB |
Verdict rules: fits comfortably when weights plus reserve sit within 85% of VRAM; tight when they fit with little left for context; CPU offload when they exceed VRAM by up to 2x (runs, slowly, with layers in system RAM); no beyond that. The KV cache per token depends on the model's architecture and is not modelled here.
Not this card. These fit it comfortably at Q4.
- GeForce RTX 5090 32GB · headroom 22.3 GiB
- GeForce RTX 3090 (used) 24GB · headroom 14.3 GiB
- GeForce RTX 4090 (used) 24GB · headroom 14.3 GiB
- Radeon RX 7900 XTX 24GB · headroom 14.3 GiB
What the GeForce RTX 5060 does run well → · Phi-4 (14B) on every card →
About the model
Phi-4 (14B) by Microsoft: 16,384-token native context, text, released 2024 under the MIT. Model card → · Hosted pricing on PicksByModel →