Today I pulled my “Super Epic” box out of the rack and did what any sane homelab tinkerer does: I shoved more hardware into it and created brand new problems. The mission was simple—install a second GeForce RTX 3060 12GB (because one is not enough, and if I had three, I’d run three). I’m using these cards for local AI, and after getting burned by a previous GPU that would crash any time I tried to run models, I grabbed this tested 3060 off eBay for about 200 bucks and went back at it. The install itself was a bit of a fine dance. My PCIe NVMe x4 adapter card (the one that holds four NVMe drives) leans just enough that it wanted to rub the fan on the second 3060. My “solution” was a very classy paper shim wedged right where it needed to be—cheap, effective, and yes, I know it’s not the proper way. I also dropped a screw into the case (because of course I did), had to tear it back down, and then reassemble once I found it. After confirming both GPUs show up in Proxmox (lspci), I made the mistake of upgrading Proxmox mid-project, which meant a new kernel and redoing NVIDIA drivers/DKMS. Once it was back, I passed both GPUs into my Ollama/Open WebUI VM and stress-tested with larger models (like Qwen/Gwen 32B and Gemma 3). It works, temps hit around 90–92°C under load, and now I’m debating more airflow or a PCIe extension cable to physically move a card for better cooling.