Patrick from ServeTheHome just built an absolute beast: an 8x NVIDIA GB10 Blackwell cluster with 1TB of unified memory, 160 Arm cores, and 1.7Tbps networking via MikroTik. The craziest part?
They didn't even configure it manually. After setting up initial access, they let autonomous AI agents handle the entire multi-node networking and RDMA setup. The video benchmarks local giants like Qwen3.5-397B and Kimi K2.6, while covering raw power consumption.
This completely changes local infrastructure deployment.
Eight GB10 nodes, RDMA fabric, and 1TB of unified memory make a serious inference cluster. If autonomous agents can reliably automate provisioning and network tuning, that's a genuine DevOps upgrade.