Phi-4 14B in Q4_K_M with an 8k context takes 10.8 GB of GPU memory. The cheapest option right now is RTX 3060 12 GB, from $0.084/h.
The weight file and the total GPU memory with the context cache. One GPU, one request.
| Quantisation | File | 8K context | Fits on |
|---|---|---|---|
| BF16 | 27.3 | 29.8 | Tesla V100 Super 32 GB |
| Q8_0 | 14.5 | 17.0 | RTX 4000 Ada 20 GB |
| Q6_K | 11.2 | 13.7 | RTX A4000 16 GB |
| Q5_K_M | 9.7 | 12.2 | RTX A4000 16 GB |
| Q4_K_M | 8.3 | 10.8 | RTX 3060 12 GB |
| Q3_K_M | 6.7 | 9.2 | RTX 3080 10 GB |
| Q2_K | 5.2 | 7.7 | RTX 4060 8 GB |
File sizes checked against Hugging Face on 2026-10-19. 1 GB here is 1024³ bytes, like GPU memory. Hugging Face lists decimal gigabytes, so its numbers are 7% larger.
How much memory the cache takes per request in FP16. A Q8 cache is about half. 200 KB per token.
GPUs from the catalog the model fits on, by price per hour.
The price is the cheapest GPU of the model × the GPU count. A 2–8 GPU server can cost a little more per GPU.
In the Ollama template the model is one command away:
ollama run phi4:14bFor an API under load there is the vLLM template with an OpenAI-compatible API.