EN
TestYourSetup / PC and printer / GPU stress test

GPU stress test online: check your graphics card for throttling and stutter

Load your graphics card for one to ten minutes and watch the frame rate second by second. A flat line means stable; a line that falls means the card slows down when it gets hot.

Duration
Intensity
More options: graphics API

Both run the same shader. WebGPU is newer and is available in recent Chrome, Edge and Safari; the test falls back to WebGL2 when it is missing.

Before you start: this loads your graphics card fully, so fans get loud and the PC gets warm. On a laptop, plug it in. Stop if you see flicker, artifacts or a crash. The test stops by itself when the time is up, when you press Stop or when you switch tabs.

Nothing runs until you press Start.

Runs only on your device, only after you press Start. Nothing is uploaded and nothing is installed.

A GPU stress test runs your graphics card at full load for several minutes to see whether it stays fast and stable. This one works in your browser, with no download: it draws a heavy fractal shader many times per frame through WebGPU or WebGL2, measures the frame rate every second, and plots it over time. Because the work per frame stays fixed during the run, a frame rate that falls means the card itself got slower, which is what throttling looks like from the outside. The test also counts stutters (frames that take far longer than usual) and notices when the graphics driver resets, a classic sign of an unstable overclock or a failing card. Use it as a quick GPU test after a new build, a driver update, cleaning a dusty cooler, or before you trust a used graphics card. It is lighter than native tools such as FurMark, 3DMark or OCCT, and a browser cannot read temperature or clock speed, so the guide below shows which free tools to keep open next to it.

How does this GPU stress test work?

The test runs in three phases, all on your device:

  • Refresh rate. With almost no load, it times 45 frames to learn your display's refresh rate, for example 60, 144 or 240 Hz. A browser never shows more frames than that.
  • Calibration. It draws the fractal as a stack of full-screen layers, one instanced draw per frame, and multiplies the number of layers by 1.6 until the frame rate falls below the refresh rate, then narrows the step down. The last layer count that still held the refresh rate is the most work your GPU finishes in one refresh.
  • The timed run. The intensity sets a fixed load from that: Light uses half of it, Medium 90%, and Full 1.7 times it, so the card can never catch up and stays busy all the time. The layer count then stays the same until the end, and the page records the time of every frame.

Each layer runs 40 steps of a fractal formula on every pixel, so the load is mostly shader arithmetic, the same kind of work that heats a card in games. At most two frames are queued at once (a fence for WebGL2, a work-done promise for WebGPU), so the frame times follow the GPU rather than the browser.

Worked example: a 3 minute run on Intel Iris Xe Graphics

This laptop also has a GeForce RTX 3050 Ti, but the browser ran on the built-in Iris Xe graphics, which is exactly the two-GPU trap described below. At Full intensity the calibration settled on 19 layers per frame at a 120 Hz refresh rate. Over the run the median frame rate was 49 fps. The first 30 seconds after warm-up had a median of 49 fps and the last 30 seconds 49 fps, a change of +0.2%. The frame rate held flat, so no throttling, but the test counted 29 stutters in 3 minutes, about one every 6 seconds, so the verdict flagged uneven frame pacing: typical for built-in graphics that also draw the Windows desktop.

Frame rate per second during a 3 minute GPU stress test on Intel Iris Xe Graphics02550751000:001:002:003:00Minutes into the testfps
Our own run of this test (Microsoft Edge 155, WebGPU, October 2026). One point per second; red ticks mark seconds with a stutter. A flat line at a fixed load means the GPU kept its speed.

How to stress test your GPU: step by step

Step 1: Close games and heavy apps, and plug in a laptop

Anything else that uses the graphics card changes the frame rate and spoils the comparison. On a laptop, plug in the charger and pick the best performance power mode, because on battery the GPU runs at a lower power limit.

Step 2: Open a temperature monitor next to the browser

A browser cannot read the GPU temperature. Open Task Manager (Performance, GPU), MSI Afterburner, GPU-Z or HWiNFO beside the browser window so you can watch temperature and clock speed while the test runs.

Step 3: Pick a duration and an intensity

Three minutes at Full is a good first check. Use 5 or 10 minutes to see whether the card slows down once it is fully warm. Light and Medium keep the frame rate at your refresh rate and are kinder to a small laptop.

Step 4: Press Start and leave the tab in front

The test first measures your refresh rate and how much work the GPU finishes per frame, then fixes the load and starts the clock. Keep the tab visible: switching tabs stops the test.

Step 5: Read the verdict and the chart

The sentence on top says whether the frame rate held, eased or fell. A flat line is stable; a line that slopes down at a fixed load means the GPU got slower, usually from heat or a power limit.

How to read the result: throttling, stutter and drops

The verdict compares the median frame rate of the first part of the run (after a 3 second warm-up) with the median of the last part. Each part is a fifth of the run, between 3 and 30 seconds. These are the thresholds the page uses:

What you seeRuleWhat it usually means
StableEnd within 4% of the startCooling and power keep up with a full load
Mostly stableEnd 4% to 10% lowerNormal boost behaviour as the card warms up
ThrottlingEnd 10% or more lowerA heat or power limit lowers the clock speed
RoseEnd 10% or more higherSomething else used the GPU at the start
StutterA frame over 2.5 times the usual frame timeA few are normal in a browser; many point to background apps, drivers or an unstable card
DropA second below 85% of the run's medianShort slowdowns, often another app or the browser
Driver resetThe graphics context was lostUnstable overclock, failing card, weak power supply or a driver fault

The 1% low is the frame rate that 99% of frames beat. It is the number gamers use for smoothness: a 1% low far below the average means uneven frame pacing, even when the average looks fine.

How long should you stress test a GPU?

For a health check, 3 to 5 minutes is enough: a graphics cooler reaches a steady temperature within a few minutes, and any throttling shows as a slope in the chart. Use 10 minutes for a laptop or a small case, where heat builds up more slowly in the chassis. For an overclock or undervolt, this browser test is a first check only. Run a native tool for 20 to 30 minutes, then play your most demanding game for an hour, because different games hit different parts of the chip.

Is a GPU stress test safe?

For a working graphics card, yes. Modern GPUs watch their own temperature and power and lower the clock speed before they reach their limits. NVIDIA, for example, lists a maximum GPU temperature of 90 °C for the GeForce RTX 4090 on its spec page, and the card slows itself down rather than go past it. A stress test puts the same load on the card that a demanding game does, only without breaks.

What a stress test can do is reveal a problem that already exists: a clogged cooler, a fan that does not spin, a power supply that cannot hold up, or an overclock that is too high. That is the point of running it. Stop the test if you see flickering, coloured artifacts, a black screen or a crash. This test caps the time at 10 minutes and stops by itself when the tab is hidden, so it never keeps running in the background.

How to check GPU temperature while you test

A web page cannot read sensors, so keep one of these free tools open beside the browser while the test runs:

  • Windows Task Manager. Press Ctrl + Shift + Esc, open Performance and select the GPU. On Windows 10 and 11 it shows GPU usage and, for most dedicated cards, the temperature.
  • MSI Afterburner shows temperature, clock speed, power and fan speed for most NVIDIA and AMD cards, with an on-screen graph.
  • GPU-Z logs the sensors and, under PerfCap Reason, tells you why the card is holding back: thermal, power or voltage.
  • HWiNFO shows every sensor in the PC, including the GPU hot spot and memory junction temperatures.
  • On Linux, nvidia-smi for NVIDIA cards, or the sensors command for AMD and Intel graphics.

Watch two numbers during the run. If the temperature climbs to its limit and the clock speed falls at the same moment the chart slopes down, the card is thermal throttling. If the clock falls while the temperature is fine, a power limit is the likely reason.

What is a normal GPU temperature under load?

There is no single safe number, because every chip has its own limit. Look up your card's maximum on its official spec page; the RTX 4090's listed maximum is 90 °C, for example. A card that settles clearly below its maximum under a full load is fine. A card that reaches its maximum and then lowers its clock is working as designed, but it is losing performance, and better cooling will win it back. Hot spot and memory temperatures, shown by HWiNFO and GPU-Z, normally run higher than the main GPU temperature.

GPU test vs GPU benchmark vs GPU stress test

A benchmark measures how fast a card is and turns it into a score you can compare. A stress test measures whether the card stays fast and stable over time. This page is a stress test: the layer count depends on your GPU and refresh rate, so it is not a score to compare between PCs. For a score, run the GPU benchmark; for a 25 second health check that the card renders correctly, the GPU test; for one very heavy shader with a live frame rate, the volume shader test; and if the test will not start, the WebGL test shows what your browser supports. Not sure which card you have? See what GPU do I have, and the GPU tier list for where it ranks.

Signs of an unstable or failing graphics card

  • The driver resets during the test or in games: the screen goes black for a moment and comes back, or the browser reports a lost graphics context.
  • Artifacts: coloured dots, lines, checkerboards or flickering blocks, especially once the card is warm.
  • A falling frame rate at a fixed load, with the temperature at its limit: cooling problem.
  • Crashes or reboots under load but not at idle: often the power supply or its cables, sometimes the card itself.
  • Loud fans with low clocks, or fans that never spin up: a fan or fan-curve problem.

When the test shows one of these, first undo any overclock, reinstall the graphics driver and clean the cooler, then test again. If it repeats, test with a native tool and, if you can, in another PC.

Laptop GPU stress test: what is different

Laptops share one small cooling system between the processor and the graphics chip, and they run at lower power limits on battery. Expect a laptop to throttle sooner than a desktop, and expect a big difference between plugged in and on battery. Plug in, choose the best performance power mode, keep the vents clear on a hard surface, and start with Light or 1 minute. On a laptop with two graphics chips, the browser may run on the built-in one: in Windows, open Settings, System, Display, Graphics, and set your browser to High performance before you test.

Why we built this GPU stress test

The question behind a GPU stress test is simple: does the card keep its speed once it gets hot? The well-known tools (FurMark, 3DMark, OCCT) answer it with a download and live graphs that you mostly read yourself. A browser test has a trap of its own: on a fast card the frame rate can sit at the monitor's refresh rate the whole time, so the card is never fully loaded and the flat line only looks like stability.

So this test measures how much work your GPU can do first, fixes the load above that, and plots the frame rate every second. The verdict follows thresholds published on this page, and it says so when the card was not fully loaded. It is open about the biggest limit too: a browser cannot read temperature. Nitin Raja, who builds this site, tried the usual GPU stress tools and browser tests before building this one, and kept only what people need: a fixed load, a chart over time and a verdict in plain words.

What this GPU stress test measures

The inputs are the duration (1, 3, 5 or 10 minutes), the intensity (Light, Medium or Full) and, under More options, the graphics API. The page draws our fractal shader through WebGPU, or WebGL2 where WebGPU is missing, times every frame with requestAnimationFrame and performance.now(), and records one frame rate per second. From those it computes the 1% low, counts stutters (frames over 2.5 times the usual frame time) and compares the start of the run with the end. A lost graphics context (webglcontextlost, or device.lost in WebGPU) is reported as a driver reset. The phases are in how this GPU stress test works and the thresholds in how to read the result. The example chart above is our own run from October 2026, exported with the test's CSV button.

What it does not measure: temperature, clock speed, power draw, fan speed, video memory errors or ray tracing hardware. It also cannot choose the graphics chip. The page asks for the high-performance GPU, but Chrome and Edge on Windows do not act on that request, which is why the example run landed on the built-in Iris Xe of a laptop that also has a GeForce RTX 3050 Ti. Set the browser to High performance in Windows graphics settings before you test.

How accurate is a browser GPU stress test?

The frame rate and frame times are measured, not estimated: the page times every frame it hands to the GPU. What a browser test cannot do:

  • No sensors. No temperature, clock speed, fan speed or power draw. Browsers do not expose them.
  • Lighter than native tools. The browser adds its own layers (the compositor, security checks, and on Windows a translation of WebGL to Direct3D), and on a very fast card the refresh rate can cap the load. The verdict says so when that happens.
  • One kind of load. The test is shader arithmetic. It does not check video memory for errors or test ray tracing hardware.
  • Other apps count. Video playback, games or other tabs using the GPU change the result. Close them first.
  • Software rendering. If hardware acceleration is off, the browser draws with the processor, and the page warns you.

So treat a stable result as "fine for normal use", and a throttling or reset result as a real finding worth checking with the tools above.

Why this GPU stress test is different

Each of these points can be checked on this page:

  • Nothing runs until you press Start. No result is filled in beforehand, and the test stops by itself when the time is up or you switch tabs.
  • The load is set for your GPU, then fixed. A falling line can only mean the card got slower, not that the test changed.
  • It says when the card was not fully loaded, instead of calling a capped run stable.
  • Published thresholds. Stable, mostly stable, throttling and stutter follow the rules in the table above.
  • A real example run. The chart above is our own 3 minute run, not a mock-up.
  • No account, nothing uploaded. The result stays on your device. When you copy it, it ends with a link that opens the test with the same settings, and that link never starts the test by itself.

This GPU stress test vs FurMark, 3DMark and OCCT

This browser testNative tools (FurMark, 3DMark, OCCT)
InstallNone, runs in the browserDownload and install
LoadFull GPU load with a fractal shader; can be capped by the refresh rate on very fast cardsHeavier and more varied loads, including power-virus style tests
Temperature, clocks, powerNot available in a browserShown live by most of them
Frame rate over timeYes, chart and CSVYes
Throttling verdictYes, from the frame rate at a fixed loadVaries; often you read the graphs yourself
Video memory error checkNoOCCT has a VRAM test
Phones, Macs, ChromebooksYes, any browser with WebGL2 or WebGPUMostly Windows; 3DMark has mobile apps
CostFree, no accountFurMark and the OCCT free tier are free; 3DMark has a free basic edition

A good order: run this test first because it takes three minutes and no install. If it shows throttling or a reset, or you are tuning an overclock, move to FurMark, OCCT or 3DMark with a sensor tool open.

Sources

Checking the whole PC? The CPU stress test loads every processor thread the same way, and the refresh rate test confirms the Hz your monitor really runs at.

GPU stress test FAQ

Is a GPU stress test safe?

Yes, for a healthy card. Graphics cards protect themselves: when they reach their temperature or power limit they lower the clock speed rather than overheat. A stress test only does for minutes what a demanding game does for hours. Stop if you see artifacts, flicker or a crash, because those point to a problem worth fixing, not to damage from the test.

How long should I stress test my GPU?

Three to five minutes shows whether the card is stable and whether it slows down as it warms up, because most coolers reach a steady temperature within a few minutes. For an overclock, run a native stress tool for 20 to 30 minutes and then play your heaviest game, since games find errors that a single test can miss.

Can a website stress test my GPU?

Yes. WebGL2 and WebGPU let a page run shader programs on your graphics card, and this test keeps it busy with a heavy fractal shader for the time you choose. It is lighter than native tools such as FurMark or OCCT, and a browser cannot read temperature, clocks or power, so pair it with a monitoring tool.

What does a GPU stress test check?

It shows whether the card keeps the same speed under a long full load (no throttling), whether frames arrive evenly (no stutter) and whether the driver stays up (no crash or reset). It does not test video memory for errors and cannot show temperature in a browser.

Why does my FPS drop during a GPU stress test?

At a fixed load, a frame rate that falls over minutes means the card lowered its clock speed. The usual reasons are heat (dust, weak fans, poor case airflow), a power limit, or a laptop on battery. A single dip can also come from another app using the GPU at the same moment.

Is GPU throttling bad?

Throttling protects the card, so it is not harmful in itself. It does cost performance: a card that drops 15% after a few minutes loses that much frame rate in long gaming sessions. Cleaning the cooler, better airflow or a slightly lower power limit usually fixes it.

Does this test work on a laptop or phone?

Yes. It runs in any browser with WebGL2 or WebGPU, including laptops, Chromebooks, Macs, Android phones and iPhones. Laptops and phones throttle sooner because they have small coolers, so start with Light or 1 minute and keep the device on a hard surface.

Why does the test say my GPU was not fully loaded?

On a very fast graphics card the frame rate can stay at your refresh rate even at the test's maximum load, so the card is never fully busy and throttling cannot show. The verdict says so. For a full load on a high-end card, use a native tool such as 3DMark, OCCT or FurMark.

Related tools