GPU Servers & On-Prem AI Guide

Running LLM inference on hardware you own: which refurbished GPU servers make sense, the NVIDIA A40 48GB as a card or installed in a tested build, what fits in 48GB of VRAM, the parts Dell doesn't list for GPU builds, and when on-prem beats the cloud bill.