Research / Engineering tools
Make the assumptions visible.
Estimate model weight memory, facility electricity and cost per million completed tokens. These tools expose formulas and measurement boundaries; inputs are illustrative, not Aguasto capacity, pricing or investment forecasts.
Weight memory estimator
Estimate the arithmetic size of model weights from the total parameter count. For mixture-of-experts models, use all parameters, including inactive experts.
Weights GB = parameters (billions) × bits ÷ 8
Weights GiB = GB × 10⁹ ÷ 2³⁰Illustrative default: 70B parameters at 16 bits. GB uses decimal bytes; GiB uses binary bytes. This excludes KV cache, activations, quantization scales, non-quantized layers, runtime overhead and duplicated weights. Bit selection does not verify an available checkpoint or supported kernel.
Weights only
130.385 GiBThis arithmetic estimate does not establish whether the model can run on a device.Facility electricity estimate
Translate a constant IT load and an assumed PUE into facility energy and electricity cost for a specified period.
Facility kW = IT kW × PUE
Energy kWh = facility kW × hours
Electricity USD = energy kWh × tariff USD/kWhAll inputs must be finite. IT load, hours and tariff may be zero; PUE must be at least 1. The defaults are illustrative assumptions, including a user-editable tariff of $0.10/kWh. They are not an Aguasto facility specification or a supplier quote.
Electricity cost for this period
28,800 kWh · 1,200 kW facility loadAssumes a constant average load and PUE. Excludes demand charges, taxes, network fees, hardware, staffing and capital costs.Cost per million completed tokens
Normalize an attributed cost over the completed tokens measured during the same period.
USD per million tokens = attributed USD ÷ completed tokens × 1,000,000Cost must be finite and non-negative. Tokens must be a positive safe integer; zero tokens cannot yield a meaningful rate. Defaults are illustrative arithmetic, not API pricing or an Aguasto performance result.