Poor man’s guide to servicing a used RTX 3090 for local LLM inference

Wrote up the whole process with disassembly photos and HWiNFO before/after data. Hope it saves someone some headaches.

https://preview.redd.it/glwl9427hgyg1.png?width=1678&format=png&auto=webp&s=8e0755ccd9be456e2e65a96613edb0bce573991b

https://github.com/cubebecu/writeups/tree/main/gpu-service