INTERACTIVE COMPUTE PLANNER
Make the assumptions visible.
Compare two allocation scenarios. Every result comes from the values you enter, with no provider pricing or availability feed.
What the numbers mean
Transparent formulas and assumptions
Allocated GPU-hours = GPUs × days × allocated hours per day.
Active-equivalent GPU-hours = allocated GPU-hours × utilisation / 100.
Compute rental = allocated GPU-hours × entered US$ rate.
Rental per active-equivalent GPU-hour = compute rental / active-equivalent GPU-hours.
Illustrative facility energy (MWh) = allocated GPU-hours × average GPU watts / 1,000,000 × other IT multiplier × PUE.
Energy is a separate planning envelope
Enter average electrical draw across the allocation, including idle time. Do not enter TDP unless deliberately modelling a full-power envelope. The IT multiplier estimates CPU, memory, storage and networking overhead. PUE adds facility overhead such as cooling. Utilisation does not automatically reduce this energy estimate. These factors are your assumptions, not measured provider data. No electricity charge is added to compute rental; cloud rates commonly include power.
What this planner does not predict
It does not estimate tokens, model training time, accuracy, scaling efficiency, CPU-only workloads, carbon emissions, GPU availability or full ownership cost. Rental excludes storage, software licences, data transfer, support, tax, minimum commitments and contract charges. Compare workload benchmarks and total quoted charges before buying.