Run Qwen 3.6-27B on $800 of consumer GPUs: llama.cpp vs vLLM

The fact

Comparison between llama.cpp and vLLM shows respective costs and performance.

The test proves LLM inference can work without expensive cloud infrastructure.