With 32 GB of VRAM, the RTX 5090 runs Qwen2.5 14B Instruct at Q4_K_M entirely on the GPU with an 8K context (about 11.2 GB needed). At Q8_0 it fits entirely.
Q4_K_M · 8K context
The model needs about 11.2 GB and fits in the 31 GB available, with 19.8 GB to spare.
~11.2 GB (10.7 GB–11.8 GB) needed · estimate
Fits entirely in VRAM · 19.8 GB left
EstimateData updated 24 days agoSource: Run LLM Locally engine, Hugging Face config.json + model card
What makes this estimate less reliable
| Qwen2.5 14B Instruct Q8_08K~17.8 GB · 68–97 tok/s estimate | Excellent |
|---|---|
| Qwen2.5 14B Instruct Q6_K8K~14.3 GB · 82–121 tok/s estimate | Excellent |
| Qwen2.5 14B Instruct Q5_K_M8K~12.7 GB · 91–136 tok/s estimate | Excellent |
| Qwen2.5 14B Instruct Q4_K_M8K~11.2 GB · 101–154 tok/s estimate | Excellent |
| Qwen2.5 14B Instruct IQ4_XS8K~10.4 GB · 107–167 tok/s estimate | Excellent |
| Qwen2.5 14B Instruct Q3_K_M8K~9.6 GB · 114–180 tok/s estimate | Excellent |
Prices and links follow the market you choose.
32 GB GDDR7 · 1,792 GB/s · 575 W
ExcellentRuns Qwen2.5 14B Instruct Q4_K_M entirely in memory
No current offer imported for this market: the link opens a marketplace search.
