Design
Architecture Designer
Choose a GPU and cluster topology. The designer estimates the bill of materials, aggregate capacity, facility power, racks, and cost — and generates a topology diagram. All figures are planning estimates.
Total GPUs
64
Aggregate FP16
7.7 PF
GPU memory
1,536 GB
Largest model
~614B
FP16 weights, aggregate
Facility power
15.4 kW
PUE 1.4 · 4.6 kW GPU
Racks
2
~15 kW/rack
Est. CapEx
$160K
GPUs only
Est. energy/mo
$1K
$0.12/kWh
Generated topology
Rendering diagram…
Bill of materialsEstimate
| L4 GPU | 64 | 8/node × 8 |
| GPU server node | 8 | 8-GPU chassis |
| InfiniBand (RDMA) fabric | 8 ports+ | RDMA / GPUDirect |
| Parallel file system | 1 tier | checkpoints + datasets, high burst throughput |
| Rack | 2 | ~15 kW usable/rack |
| Kubernetes control plane | 3 | HA control-plane nodes |
Power, rack, and cost figures are planning estimates from illustrative pricing — not quotes.
Bootstrap a cluster with L4 →