RTX PRO 6000 Blackwell Max-Q: Proxmox Passthrough for Local AI
··6437 words·31 mins
Running a 96 GB RTX PRO 6000 Blackwell Max-Q on Proxmox and K3s: PCIe topology, VFIO, NVIDIA drivers, vLLM and real onboarding failures.
Homelab AI
This series follows the AI side of the homelab: GPU lanes, local models, OpenAI-compatible gateways, agent workflows, observability integrations, and safety boundaries.
A Read-Only SRE Agent for My Kubernetes Homelab
·6829 words·33 mins
How a homelab SRE agent uses HolmesGPT, Grafana MCP, Qdrant, and a local LLM for bounded Kubernetes triage without cluster write access.
One GPU, One LLM Lane: Operating an RTX PRO 4000 SFF in k3s
··6503 words·31 mins
A practical operating model for a single 24 GB GPU in k3s: one active LLM workload, explicit scheduling rules, preflight-gated model switches, and measured Qwen3.6 performance