Cluster Power Calculator
Estimate the electrical footprint of a GPU cluster: pick a chip, set the count, and see IT load, facility load and an illustrative energy bill. Arithmetic only — every assumption is yours to change.
TDP is each accelerator's maximum design power (vendor spec). Host overhead covers CPUs, networking, fans and power conversion around the GPUs. PUE multiplies IT load into facility load (1.1 = efficient hyperscale, 1.5 = ordinary data center).
—IT load (chips + hosts)accelerator TDP × count × (1 + overhead)
—Facility loadIT load × PUE — what the grid must supply
—Energy per monthat the chosen utilization
—Illustrative power bill / monthenergy × your price — illustrative only
—US homes equivalentat ~1.2 kW average household draw (EIA-derived rule of thumb)
Worked example
1,024 × NVIDIA H100 (700 W vendor spec), +60% host overhead, PUE 1.50, 70% utilization, $0.08/kWh: IT load ≈ 1.15 MW; facility load ≈ 1.72 MW; energy ≈ 879 MWh/month; illustrative bill ≈ $70,000/month. A 100,000-GPU campus at the same assumptions lands near 168 MW — utility scale, which is why the Power section of this database exists.
All outputs are arithmetic on your inputs plus vendor-spec TDP; real deployments vary with workload, binning, cooling and tariffs. Research and education only.