The working files an architect needs to move from an AI ambition to a defensible cluster design.

What is inside:

• GPU capacity-planning workbook — sizing for training and inference, tokens/sec and throughput modelling, utilisation and headroom
• Accelerator selection matrix — NVIDIA B300, H200 and RTX PRO 6000-class systems compared on memory, interconnect, power draw and workload fit
• Reference architectures — single-rack pilot, multi-rack training pod and inference-serving tier
• Network design guide — InfiniBand vs RoCE, rail-optimised topology, oversubscription ratios and switch counts
• AI storage architecture sheet — throughput per GPU, checkpoint sizing, parallel filesystem vs object tiering
• Datacenter planning sheet — kW per rack, power distribution, air vs liquid cooling and rack elevations
• Kubernetes and GPU orchestration baseline — scheduling, MIG, node pools and multi-tenancy
• GPU-as-a-Service comparison model — build vs rent economics over a three-year horizon

Built for engineers who have to defend the numbers in a design review.