Running a powerful Claude like LLM on a dedicated server is achievable by matching model scale to NVIDIA GPU VRAM