The Bench
9 checks for the hardware in front of you: screen, mouse, keyboard, microphone, camera and connection. Everything runs in your browser, nothing installs, and this page measures nothing until you open a test.
The tests
Every test, by group
2 of them ask for a device and say so on the card. The other 7 need nothing but the tab they open in.
9 tests
Display
Panel faults, motion and colour. Run the monitor test fullscreen, in a dark room, at native resolution.
Monitor Test
One stage, seven modes: 15 test patterns, dead-pixel colour fields, an OLED burn-in wash, PWM flicker detection, a ghosting sweep, blank screen and a measured refresh rate.
Display and Scaling Calculator
Screen size plus resolution gives PPI, the Windows scaling percentage that actually fits, effective desktop space at each step, and a viewing-distance check.
Input
Mice and keyboards: the parts that wear out and start lying to you.
Mouse Test
One stage, four modes: true DPI against the advertised number, real polling Hz, a chattering double-click switch, and clicks per second.
Keyboard Tester
Press every key to verify it fires, test NKRO and ghosting with live and max simultaneous-key counts, catch stuck keys, read event.key and event.code.
Audio and video
The two tests that ask permission, and the only two. Nothing recorded leaves the tab.
Network
Latency first, throughput second. Jitter is what you actually feel.
System
What the machine says it is, read locally and copyable in one block.
What The Bench measures, and what it does not
This page is a directory. It runs no probe, asks for no permission and reads nothing about your machine. Each test measures when you open it, and only then.
- Refresh rate is measured, not looked up. Your browser repaints once per display frame, so timing about forty consecutive frames and taking the median gap gives your real rate. The median shrugs off the one-off hiccups that would wreck an average. A result within 2% of a common rate (60, 75, 90, 120, 144, 165, 240 or 360) snaps to it, and the raw figure stays on screen. That measurement lives in the Monitor Test and in System Info.
- Resolution and scaling come from
screen.width,screen.heightanddevicePixelRatio. A reading of 2560x1440 at 1.25x means your system reports a 2560x1440 desktop drawn at 1.25 density, so the physical panel is 3200x1800. The Display and Scaling Calculator works the same numbers from the other end. - Colour gamut and HDR are CSS media queries. They report what the browser is willing to render, which is what actually reaches your eyes.
- Audio comes from an
AudioContextthat is opened, read for sample rate and channel count, and closed again in the same breath. No microphone is touched until you press start in the Mic Test. - Cameras and microphones are counted, not identified. Device names are the part that triggers a permission prompt, so a count is all you get before you grant access. Some browsers collapse the list to one entry per kind until then, so treat the counts as a floor rather than a total.
- The GPU string comes from WebGL, and some browsers deliberately mask it. If yours reads
masked by browser, that is privacy protection working, not a fault. - Nothing is stored on a server. No account, no upload, no analytics on any measurement. The numbers exist only in the tab you are looking at.
Before you trust a reading
Five things that make a browser measurement wrong, in the order they catch people out.
- Background tabs are throttled. Browsers slow a hidden tab to roughly one frame a second. Keep the test in front while it measures.
- Battery saver caps frame rate. Laptops and phones in low power mode routinely pin the browser at 30 to 60 fps regardless of the panel. Plug in and run it again.
- The browser follows the display its window is on. Drag the window to the monitor you want to measure before you start.
- The
downlinkfigure is deliberately coarse. Browsers round it to protect you from fingerprinting. For a real number, run the Speed Test. - A browser cannot read everything. SMART data, temperatures, fan speeds and voltages need a native tool. Everything you can see, hear or click, a browser can test.
How to run a full check
Six steps, in this order. Each one rules out the faults that would make the next reading a lie.
- Start with the screen. Dead pixels and burn-in are the only faults that are permanent, so find them first, especially inside a return window.
- Then motion. Refresh rate, ghosting and flicker only mean something once you know the panel itself is clean. All three are modes of the Monitor Test.
- Then input. A double-clicking switch and a stuck key are the two commonest reasons a mouse or a keyboard gets replaced, and both take under a minute to prove.
- Then audio and video. Microphone first, then camera. Both count your devices before they ask for one, so you know how many the browser can even see.
- Finish on the network. Ping and jitter matter far more than raw throughput for games and calls. The download figure is the last thing to worry about.
- Copy the readout. System Info prints one block of plain text. If you are filing a return or asking for help, it answers the first five questions anyone will ask you.
When to run which test
Five things that send people here, and what to open for each.
Just bought a monitor
Run the Monitor Test in the first week, fullscreen, in a dark room. Dead pixel policies have a time limit and a pixel count, and a photograph of a stuck pixel is much easier to take than to argue about later. If the panel is OLED, run the burn-in wash as well. The Display and Scaling Calculator tells you whether the scaling you are running is the one that fits.
Games feel smooth for everyone else but not you
Check the measured refresh rate first. A large share of 144 Hz monitors are still running at 60 Hz because nobody changed the setting. Then the ghosting sweep for pixel response, and the flicker mode if you get eye strain, which is often PWM backlight dimming rather than the game.
Your mouse clicks twice
The Mouse Test counts unintended second clicks. A switch that double fires on more than a few percent of clicks is failing, and it only gets worse. It is the single commonest mouse fault there is. Run the DPI mode as well if the cursor also feels wrong at a sensitivity that used to be right, and the Keyboard Tester if a key is repeating or dropping.
Nobody can hear you
Run the Mic Test and watch your own waveform before blaming the app. If the meter moves, the mic works and the problem is which input device the software picked. The Webcam Test does the same for video, including the resolution you are actually being given rather than the one the box claims.
The game is laggy but the speed test is fine
That is the normal case. Throughput is rarely the problem. Run the ping mode and watch the jitter figure rather than the average: a stable 60 ms beats a 20 ms average that spikes to 300. What Is My IP fills in the rest, including whether you are behind WARP or a VPN.
Questions
Does this page measure my hardware?
Not any more. The hub is a list of links and it reads nothing about your machine. The readout it used to run lives in full at System Info, which uses standard browser APIs that need no permission: the refresh rate by timing about forty consecutive frames and taking the median gap, resolution and pixel ratio from screen and devicePixelRatio, the core count from hardwareConcurrency, the GPU name from WebGL, and the camera and microphone figures from a device count that returns no names. There is no lookup, no database, and no request to a server.
Does anything here need a permission prompt?
The hub itself never asks for one, and neither do 7 of the 9 tests. Only the Mic Test and the Webcam Test do. They ask when you press start, the browser prompt is the gate, and you can revoke access at any time. Screen, mouse, keyboard, network and system tests need nothing.
Why do the cards not show live values?
They used to. The old hub measured your machine while it loaded and printed twenty-five readings before you clicked anything. It does not now, because a directory that quietly probes your hardware is not a directory. Every one of those readings survives at System Info, which measures when you open it and says what it is doing.
Why does it say I have no camera when I do?
Some browsers hide device counts until a page has been granted access once, and privacy modes, including Firefox resist fingerprinting, Safari Lockdown Mode and most content blockers, can zero the count entirely. Open the Webcam Test and allow access. If the picture appears, the camera is fine and only the count was hidden.
Is a browser test as good as a real diagnostic?
For faults you can see or hear, yes. A dead pixel, a double-clicking switch, a stuck key, a flickering backlight and a microphone that never moves the meter all show up identically in a browser. What a browser cannot do is read SMART data, temperatures, fan speeds or voltages. For those you need a native tool.
Are these results good enough for a warranty claim?
For visible faults, yes. Dead pixels, stuck subpixels, burn-in, ghosting and a mouse that double-clicks on one press all photograph or record clearly off these tests, and manufacturers accept that evidence. For timing claims, such as a 144 Hz monitor running at 60, the measurement is accurate but check the obvious causes first: the operating system display setting, battery saver, and which monitor the window is actually on.
Does anything I test get uploaded?
No. Every test runs entirely in your browser. Microphone audio and webcam video never leave the machine, and are played back to you from memory. The network traffic The Bench makes is all its own: the Speed Test downloads a file from this site to time it, repeats a small timed request to the same host to measure ping and jitter, and sends one 8 KB POST probe to find out whether this host would accept an upload at all. What Is My IP reads your connection details back from our own edge. Every one of those goes to vorla.dev and none of them carries a thing you recorded, typed or measured.