Epoch Time Estimate Calculator

Work out epoch time estimate instantly with clear inputs, formula shown and shareable results.

About Epoch Time Estimate Calculator

One epoch is ceil(samples / batch size) optimisation steps, so epoch time is that count multiplied by measured step time. Step time is the honest way to model this because it already includes forward, backward, optimiser and any data-loading stall. Throughput in samples per second is the figure to compare when changing batch size, precision or hardware.

Formula

Epoch time

steps = ceil(samples / batch size); epoch time = steps x step time

Worked example

With the default values below, Epoch Time Estimate Calculator returns Time per epoch of 843.8. Change any input above and every figure here is recalculated in your browser.

Inputs used

Training samplesminimum 1: 1,200,000
Batch sizeminimum 1: 256
Time per optimisation stepminimum 0.1 ms: 180 ms

Results

Time per epoch: 843.8 s
Optimisation steps per epoch: 4,688
Training throughput: 1,422.1 samples/s
Time per epoch in hours: 0.234 h

Step by step

1Values used
2Epoch time
3Time per epoch= 843.8 s
4Optimisation steps per epoch= 4,688
5Training throughput= 1,422.1 samples/s
6Time per epoch in hours= 0.234 h

Assumptions and limits

The result is derived only from the 3 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 3 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 does doubling the batch size not halve epoch time?

Step time grows with batch size once the GPU is saturated. The gain comes from better utilisation at small batches and from fewer optimiser and communication overheads, not from free parallelism.

What if the data loader is the bottleneck?

Then step time is dominated by waiting for input and no GPU change helps. Measure GPU utilisation; below about 80 percent, fix prefetching, worker count and decoding first.