How to Measure Laptop and Phone Performance Accurately: A Power, Heat and Background App Checklist

If you have ever run a benchmark on your laptop or phone and got a different score today than yesterday, or a different number from a friend with the same model, the device is almost certainly not broken. More likely, the test conditions were different. A browser benchmark reports how fast your device actually performed at that moment, so the power state, the temperature and whatever else was running all show up in the score. This guide walks through what makes scores wobble when you measure with Web Bench, what to check before you start, when to use the quick test, the standard test or the stress test, and how to read the results sensibly. Keep one thing in mind throughout: every score is a reference value that depends on the environment.

Why the same device gives different scores

A benchmark score is not calculated from a chip name or a spec sheet. The test runs real work and measures how long it takes, so the result depends on what state the device was in while the work ran. Four factors cause most of the variation.

• Power state and power mode: whether a laptop is on battery or plugged in, and whether the operating system is set to a battery saver, balanced or performance profile, changes how much power the CPU and GPU may use. A phone's low power mode is the same kind of limit.
• Heat and throttling: when a device gets too hot, it slows itself down to protect itself. This is called thermal throttling. Many devices are fast at first and slower once a heavy load has gone on for a while.
• Background work: other tabs, other apps, file downloads, cloud sync, video playback and security scans all share the CPU and memory, which lowers the score.
• Screen and tab state: if the test tab is hidden or the screen turns off, the browser and the operating system try to cut back that tab's work. A number produced in that state is not a normal measurement.

On top of that come software differences such as the browser and its version, hardware acceleration settings and graphics drivers. So the basic rule for comparing scores is simple: make the conditions of the results you compare as identical as you can. The Web Bench result screen says the same thing: the same device can score differently depending on browser, power and heat, so compare results taken under the same conditions.

How power mode, heat and background apps affect your score

Power is the easiest factor to overlook and the one that moves scores most readily. When a laptop runs on battery, the operating system often limits CPU and GPU behavior to save energy. The Web Bench stress screen even carries a note for this case: when you are on battery, plugging in reduces the score swings caused by battery drain and power limits. Unless you are deliberately testing battery behavior, plug the laptop in before you measure.

The name and location of power settings differ between operating systems, but what you need to check is the same everywhere. Look at whether the current setting leans toward saving power, balance or performance, and write it down. The next time you measure, you need to match that setting for the comparison to mean anything. If you are curious how much a power mode changes your score, change only that setting and run two measurements. Keep everything else the same so the one change is the only variable.

Heat matters more the longer a measurement runs. Thin laptops and phones cannot push heat out quickly, so if you run several tests back to back, the later ones often score lower than the earlier ones. The same happens if you test right after a heavy game or video edit, because the device starts out already warm. Let it rest for a few minutes to cool down before you begin.

Background work is not limited to apps you can see. A video playing in another browser tab, an automatic update, cloud file sync and a file transfer in a messenger all count. Pausing or finishing these before you start narrows the spread between runs.

Screen sleep and hidden tabs matter too. If the tab is hidden during a standard or quick test, Web Bench tells you the tab was hidden, throws away the round in progress and measures that round again when you return. In the stress test, hiding the tab or turning off the screen pauses the run, discards the 15 second segment in progress and measures it again on return. This keeps bad values out of the result, but it also stretches the test, and the device may cool down while it is paused. The best approach is not to hide the tab in the first place. Keep the test tab in front, keep the screen awake and do not wander off to other windows.

A checklist to run before every measurement

The preparation screen in Web Bench includes a pre-start checklist with three items: connect the charger, close other apps and tabs, and keep the screen brightness steady. Adding what we covered above gives you this longer list.

1. Connect the charger. On a laptop, plug it in. On a phone, decide whether to charge and keep it the same every time.
2. Check the power mode and write it down: battery saver, balanced or performance.
3. Close other apps and tabs. Finish anything that uses the CPU, such as downloads, sync, video playback and video calls.
4. Set the screen brightness to a fixed level. If brightness adjusts automatically, lock it for the test.
5. Let the device cool. If you just did something heavy, wait a few minutes before you start.
6. Choose your browser. When comparing, use the same browser each time, and update it if needed.
7. Keep the test tab in front and do not switch windows until the run finishes.
8. Silence notifications. A notification or a phone call in the middle of a run can disturb the result.

Quick, standard or stress test: which one and when

Web Bench offers three kinds of measurement with different purposes. What differs is not only the time they take but also what each one tells you.

• Quick test (about 30 seconds, representative items): runs a selected set of representative tests instead of everything. It suits a light check of roughly where a device stands, or seeing the effect of a setting change right away.
• Standard test (about 2 to 3 minutes, all items): runs CPU, memory, GPU, browser and storage tests broadly. For each detail test, the score comes from the median of three measurements taken after a warm-up, and the category and overall scores are averages of those. This is the baseline measurement for understanding a device or comparing it with another device or setting.
• Stress test (5 or 10 minutes): holds a load for a long time and records a score every 15 seconds, so you can see how performance changes over time. It tells you the speed a device can sustain, not its momentary peak.

If you are new, start with the quick test to see the rough level, then run the standard test when you want more detail. Add the stress test only when you care about heat or sustained performance. The standard test also offers an optional AI measurement, but because it involves a model download and inference, it takes longer and uses more memory and power. If the other items are enough for you, leave it off. One more thing to remember: quick results and standard results use different methods, so do not compare them directly. Always compare like with like.

Reading heat-related slowdown with the stress test

For the stress test you pick the load target, which is CPU multi, GPU or both, and the loaded time, which is 5 or 10 minutes. Five minutes corresponds to 20 rounds of 15 seconds and ten minutes to 40 rounds. Each round's score is recorded and drawn on a time graph. The stress score uses a formula separate from the standard test, so never mix the two when comparing.

There are two things to look at in the result. The first is stability: the lowest recorded round score divided by the highest, as a percentage. The closer it is to 100%, the better the device held its speed over time. The second is the throttling start time. When two rounds in a row come in at least 10% below the best score so far, the first of them is estimated to be where throttling began. A graph that starts high, steps down and settles at a lower level can be a typical sign of a device lowering its speed because of heat. A nearly flat line suggests the device handles a long load well.

A few cautions apply. These values are not a direct reading of a temperature sensor; they are inferred from changes in performance. In browsers that support CPU Compute Pressure, the state changes are shown as well, but those are not temperature readings either. Background tasks or power-saving settings can cause a similar drop, so do not conclude from the curve alone that your device is faulty. A running stress test can make the device hot, so if it feels too hot, stop it right away with the stop button.

Measure more than once and compare with recent results

One score cannot describe a device. Run the test two or three times under the same conditions and look at the range. If the values cluster within a few percent, you can trust them. If one run is unusually low or high, it is reasonable to suspect a variable such as background work or heat in that run. Rather than recording only the best score, use the range from runs in your normal environment as your baseline, because that makes later changes easier to compare.

Web Bench keeps the latest 20 completed results in this browser only and does not send them to a server. On the result screen, the recent results comparison lets you pick a baseline result and a comparison result and see the change. Only results that share the same scoring version, measurement method, GPU path, AI path and valid items can be compared. If the compare button is blocked because conditions differ, that is deliberate: it prevents an unfair comparison. Because storage is inside the browser, clearing site data removes the history, and if the browser cannot use its storage, the result disappears when you close the page.

The most accurate experiments change one thing at a time. To see the effect of a power mode, keep the charger, the browser and the brightness the same, change only the power mode, and compare the two runs side by side in the recent results. Use the same method when you compare before and after a driver or browser update.

Tips for measuring a smartphone

A phone has a small, thin body that is hard to cool, and its operating system tends to apply strong power-saving policies. That makes controlling conditions even more important than on a laptop.

• Decide whether to charge and do the same every time. While charging, the battery and the body produce more heat, which can change results.
• Check low power mode. When it is on, performance is limited. Measuring once in normal mode and once in low power mode shows how much the battery-saving setting costs you.
• Take off the case and test in a cool place. A thick case traps heat.
• Keep the screen on with the test tab in front. If the screen turns off or the tab is hidden, the segment in progress is discarded and the run pauses.
• Turn on Do Not Disturb. A call or notification in the middle of a run can disturb the result.
• Start with the quick test. If heat worries you, check briefly first and only then move on to the stress test.

During a stress test the phone may get hotter than usual, so keep an eye on it, and stop right away if it becomes uncomfortable. If you download the result as a PNG card to share, check the device model information on the card first.

Cautions when interpreting results

• Scores are reference values. They vary with browser, power mode and heat, and they do not certify a device's grade or quality.
• Do not read meaning into small differences. A gap of a few percent may be within measurement noise. Focus on the items that differ widely.
• Compare only under the same conditions. A different browser, method, power state or GPU path makes it hard to say which result is better. Placing WebGPU and WebGL2 results on the same line is an especially easy way to be misled.
• Do not convert a multi-core score directly into a core count or clock speed. It depends on the number of Workers the browser uses and how they synchronize.
• If a result is lower than expected, do not suspect a defect right away. Check your settings and environment first, then measure again under matching conditions.
• The throttling time and stability figures are inferred from performance changes, not measured temperatures.

In the end, the best use of a benchmark is not bragging about a number but comparing your own device before and after a change under the same conditions. Control the conditions, measure several times, and the numbers become far more trustworthy.

When you are ready, start a measurement from the link below. No installation or login is needed, and everything runs inside your browser.

Start a performance test on Web Bench