256 GPUs · 204.8 Tb/s InfiniBand NDR800 · 1:1 Non-Blocking Fat-Tree · Live Topology Visualization
Comprehensive 15-section reference — power audit, rail topology, switch specs, BOM, bandwidth summary, topology diagrams, design decisions & risk matrix.
Detailed rack-level layout for compute zone (C01–C32) and network tower (N1–N2). Covers IB fabric, Spectrum-4 Ethernet, UFM Appliance & OOB management.
Per-server deep-dive: ASUS XA NB3I-E12 BOM, port-by-port connectivity, CX8 rail mapping, BF3 DPU wiring, X710 OOB and power audit per component.
Complete planning overview consolidating the 20 kW wall-output envelope, zone layout, networking decisions, switch selection rationale and deployment checklist.
Cooling (4× CRAC, N+1 with 3 active + 1 standby · 650+ kWth), dual-feed power distribution (80 rack PDUs · Row ATS · UPS), and IB-attached shared storage cluster (~590 TB usable · NVMe-oF / Lustre). Separate infrastructure layer covering all 40 rack positions.