Qwen3 8B in Q4_K_M with an 8k context takes 6.8 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 | 32K context | Fits on |
|---|---|---|---|---|
| BF16 | 15.3 | 17.4 | 20.8 | RTX 4000 Ada 20 GB |
| FP8 | 8.8 | 11.0 | 14.3 | RTX 3060 12 GB |
| Q8_0 | 8.1 | 10.3 | 13.7 | RTX 3060 12 GB |
| Q6_K | 6.3 | 8.4 | 11.8 | RTX 3080 10 GB |
| Q5_K_M | 5.4 | 7.6 | 11.0 | RTX 4060 8 GB |
| Q4_K_M | 4.7 | 6.8 | 10.2 | RTX 4060 8 GB |
| Q3_K_M | 3.8 | 6.0 | 9.4 | RTX 4060 8 GB |
| Q2_K | 3.1 | 5.2 | 8.6 | RTX A2000 6 GB |
How much memory the cache takes per request in FP16. A Q8 cache is about half. 144 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 qwen3:8bFor an API under load there is the vLLM template with an OpenAI-compatible API.
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.