First time running Qwen3-Coder-Next — an 80 billion parameter Mixture of Experts model — on local hardware. Started at 27 tokens per second. Tuned the inference engine (vLLM on NVIDIA Blackwell), disabled CUDA graph suppression, and hit 129 tok/s. Research confirmed this is the ceiling for consumer Blackwell GPUs right now — datacenter cards get native MoE kernels, workstation cards fall back to generic Triton. We're at the wall. Still faster than most cloud APIs.

Replaced the SG-1100 running pfSense with an MS-01 running Proxmox and OPNsense. Full VLAN segmentation went live — management, compute, IoT, guest, all isolated. Deployed two beacon nodes with Keepalived failover as the distributed control plane. The network went from a flat consumer setup to a segmented lab architecture in one session.

First post from @directive4AI. Electrical rewiring done. Equipment racked. The lab is coming up. Everything that follows builds on this foundation.

Moved the entire operation from San Antonio, TX to West Branch, MI. Wife, toddler, and a second on the way. Built a temporary lab in Texas before we left so work didn't stop during the 45-day transition. Rewired the electrical in the new house. The company didn't go dark — we relocated.