---
id: throttle-test-length
type: concept
title: How long should a throttle test be?
summary: Five minutes for the CPU and fifteen for the GPU. In the developer's 20-minute tests on an iPhone 17 Pro, iPad Air and iPad mini (5 November 2026), CPU speed settled within about 4 minutes and GPU speed within 10 to 15. Running to 20 minutes added nothing.
section: kb
url: "https://irrhuh.app/therbie/kb/throttle-test-length.html"
related: [benchmarks, device-link, thermal-states, glossary]
platform: [iphone, ipad, mac]
applies_to: 1.0
updated: 2026-11-16
---

# How long should a throttle test be?

Five minutes for the CPU and fifteen for the GPU. Those are the two lengths Therbie offers on iPhone and iPad. They come from 20-minute tests the developer ran on three devices overnight, and again the next afternoon.

## How the tests were run

- **When:** 5 November 2026.
- **Devices:** an iPad Air (A14), an iPad mini (A17 Pro) and an iPhone 17 Pro.
- **Method:** a run plan from Therbie for Mac. Each device ran a CPU test, a GPU test and a repeat CPU test, 20 minutes each, with rests in between.
- **Builds:** debug builds, which understate raw CPU speed. The % kept is relative to each run's own peak, so it isn't affected.

These are one set of measurements on three devices. Your device and room will give different numbers.

## The overnight runs

The table shows how much of its peak speed each device was still holding at each point. "Settled" is when the run stayed within ±5% of its final level.

| Run | Settled (±5%) | 5 min | 10 min | 15 min | 20 min |
|---|---|---|---|---|---|
| iPad Air (A14) · CPU | 1 min 10 s | 85% | 84% | 83% | 84% |
| iPad mini (A17 Pro) · CPU | 30 s | 75% | 75% | 75% | 75% |
| iPhone 17 Pro · CPU | 3 min 10 s | 79% | 78% | 79% | 78% |
| iPad Air (A14) · GPU | From the start | 100% | 100% | 100% | 100% |
| iPad mini (A17 Pro) · GPU | 11 min 9 s | 68% | 64% | 68% | 67% |
| iPhone 17 Pro · GPU | 10 min 17 s | 72% | 68% | 73% | 73% |

- **CPU:** every device settled within 4 minutes. The 5-minute figure matched the 20-minute one to within 1 to 4 points, and the repeat runs agreed.
- **GPU:** the two newer chips dipped and recovered for 10 to 15 minutes before settling. The older iPad Air's GPU never slowed at all.

## The room mattered more than the case

The same tests ran again in the afternoon with cases removed, in a room about 2 °C warmer (25 °C instead of 23 °C). Every device kept less of its peak:

| Device | CPU at 5 min, overnight → afternoon | GPU at 15 min, overnight → afternoon |
|---|---|---|
| iPad Air (A14) | 85% → 80–83% | 100% → 89% |
| iPhone 17 Pro | 79–81% → 72–74% | 73% → 63% |
| iPad mini (A17 Pro) | 75–76% → 66–69% | 68% → 65% |

Lying flat on a desk also traps heat against the back of the device, where these devices shed most of it. A keyboard case or stand that lets air reach the back helps.

> **Tip** For fair comparisons, test in the same room, at the same time of day, in the same position, and label each run with its setup.

## Cooling time

Between tests, the iPhone took 18 to 31 minutes to return to Normal; the iPads took about a minute. Plan runs on iPhones with long rests, or let the run plan's **Wait Until Normal** step handle it. See [Testing devices from your Mac](https://irrhuh.app/therbie/guide/device-link.html).

## Other lengths on the Mac

Therbie for Mac's **Throttle Test…** sheet also offers 10 and 20 minutes, and a run plan's throttle step can run from 1 to 60 minutes.
