How do I calculate the cost of a training run?

Multiply the number of GPUs by the hours the run takes and by the price per GPU-hour. A 90-day run on 1,000 GPUs at $2 per GPU-hour is 1,000 × 2,160 × 2 = $4.32M.

How much is a 1% reduction in training compute worth?

One percent of the run's cost. On a 100,000-GPU, 90-day run at $2 per GPU-hour that is about $4.3M.

Guide

2026-09-27

Training cost calculator

What a training run costs in GPUs, hours and dollars, and what a measured reduction in compute is worth on your cluster. Projection, labelled as one.

Set the cluster, the price and the length of the run. The reduction defaults to about half, which is the middle of what we measured across real text domains; the slider runs to 63%, the best case we have measured, on PubMed.

GPUsPrice per GPU-hour, $Length of the run, daysCompute reduction, %

50%

The run: $4.3M; 1,000 GPUs × 2,160 hours × $2
Saved at 50%: $2.2M; and about 6 weeks back on the calendar
Every 1%: $43K

A projection: cluster cost is arithmetic, and the reduction is carried from runs measured on models up to 41.7M parameters. A pilot on your own job verifies it on yours.

Questions this answers

How do I calculate the cost of a training run?: Multiply the number of GPUs by the hours the run takes and by the price per GPU-hour. A 90-day run on 1,000 GPUs at $2 per GPU-hour is 1,000 × 2,160 × 2 = $4.32M.
How much is a 1% reduction in training compute worth?: One percent of the run's cost. On a 100,000-GPU, 90-day run at $2 per GPU-hour that is about $4.3M.