Skip to content
ENExplore Calculators

Compute Budget (Chinchilla) Calculator

Work out compute budget (chinchilla) instantly with clear inputs, formula shown and shareable results.

About Compute Budget (Chinchilla) Calculator

The Chinchilla scaling study found that for a fixed compute budget, parameters and training tokens should grow together at roughly 20 tokens per parameter — far more data than earlier models used. Multiplying by the 6ND rule gives the FLOPs, and dividing by device throughput times model FLOPs utilisation converts that into device time you can actually schedule.

Formula

Chinchilla-optimal budget

tokens = 20 x parameters; FLOPs = 6 x parameters x tokens; device time = FLOPs / (peak throughput x utilisation)

Worked example

With the default values below, Compute Budget (Chinchilla) Calculator returns Compute-optimal training tokens of 140,000,000,000. Change any input above and every figure here is recalculated in your browser.

Inputs used

Model parametersminimum 1
7,000,000,000
Training tokens per parameteraccepts 1–200
20
Peak device throughputminimum 1 TFLOP/s
312 TFLOP/s
Sustained utilisation (MFU)accepts 1–100 %
45 %

Results

Compute-optimal training tokens
140,000,000,000
Training FLOPs required
5.880e+21
Device-days of compute
484.73 device-days
Device-hours of compute
11,633.4 device-h

Step by step

  1. 1Values used
  2. 2Chinchilla-optimal budget
  3. 3Compute-optimal training tokens= 140,000,000,000
  4. 4Training FLOPs required= 5.880e+21
  5. 5Device-days of compute= 484.73 device-days
  6. 6Device-hours of compute= 11,633.4 device-h

Assumptions and limits

  • The result is derived only from the 4 inputs listed above. Anything not among them is not part of the calculation.
  • In particular, this calculator does not take inflation, tax, fees or charges or compounding frequency into account.
  • Figures are rounded for display to 2 decimal places, so totals built from several rounded parts can differ slightly from the unrounded arithmetic.

How these calculators are built and tested is described on the Methodology page. If a figure looks wrong, please report it.

Frequently Asked Questions

Why do modern models train on far more than 20 tokens per parameter?

Because inference cost matters. Overtraining a smaller model past the compute-optimal point costs more to train but is permanently cheaper to serve, which is why ratios of 100 or more are now common.

What is a realistic MFU?

Large well-tuned runs reach 40 to 55 percent model FLOPs utilisation. Below 30 percent, look at data loading, sequence packing and communication overlap.

Related Calculators

Compute Budget (Chinchilla) Calculator | Calcrivo