TestBenchmark logoTestBenchmark

GPU Benchmark Test

This GPU benchmark test measures how fast your graphics card really is — shader, geometry and particle speed in about 15 seconds, right in your browser. No download, no sign-up, no admin rights, nothing uploaded. One honest limit up front: a browser score only compares within the same browser at the same load, so this page prints its scoring formula and reports the raw numbers, not a made-up tier.

Your graphics cardChecking this browser…

The line below shows what your browser says about this device: whether WebGL is available, which graphics card it reports, and the largest texture it supports. The benchmark needs WebGL — if it is off, the page says so and tells you how to turn it on instead of showing a fake number.

Reading this browser…

1 · Run the benchmark — about 15 seconds

Press Start and the test runs three fixed workloads in turn — a shader stage, a geometry stage and a particle stage, about 5 seconds each — on the canvas below. Pick a load first: Low is light enough for a phone, Medium is the usual comparison, High is the stress load for a desktop card. The score for each stage is its average FPS times the load factor times ten, and the total is the three stage scores added together. Keep this tab in front for the whole run.

Not run yet — shader 5 s, then geometry 5 s, then particles 5 s.

GPU score

—

Average FPS

—

1% low FPS

—

Load

—

2 · What the three stages measure

Each stage stresses a different part of the card, so a card that is strong on shaders but weak on geometry reads differently from one that is strong everywhere. The shader stage runs a Mandelbrot fractal — pure fragment math, the same kind of work a game spends on lighting and effects. The geometry stage draws thousands of triangles in one draw call — vertex work, the kind a 3D scene spends on models. The particle stage moves thousands of points — the kind of work a game spends on smoke, sparks and rain. Each stage reports its own average FPS, its 1% low (the slow frames an average can hide) and its own score.

Shader

—

Geometry

—

Particles

—

3 · Stress test — 20 seconds at High load

Press Start and the shader stage runs at High load for 20 seconds. The test compares the frame rate in the first 5 seconds with the frame rate in the last 5 seconds: a card that holds its pace is cooling well, one that fades is throttling. Stop any time — closing the tab stops the load immediately, and the page never reads your temperature, because no web page can.

Not run yet — 20 seconds at High load.

First 5 s FPS

—

Last 5 s FPS

—

Drop

—

4 · Runs this session

Every run you do on this page is kept here, so you can see the pattern instead of trusting one number. Run the benchmark three times at the same load: the first run is often the lowest, and a score that falls on every run points to cooling, while one that is low every time points to a setting. Scores from different loads cannot be compared — the load factor is part of the number.

RunGPU scoreAvg FPS1% lowLoad
No runs yet — press “Start benchmark” above.

5 · Your GPU report

The report reads the latest run, the weakest stage and the run history, and says in plain words what the numbers mean — whether the card is shader-bound or geometry-bound, whether it looks like it is throttling, and what to check first. Copy the text version to paste into a support ticket, a sale listing or a message to a friend.

Honest limits

A web page only sees the work the browser lets it do. It cannot read your GPU temperature, clock speed, fan speed, VRAM size or the exact graphics card model — no browser API exposes those on any platform — so this page never claims to know them. It counts the frames your screen shows, which means the reading is capped at your monitor's refresh rate: a 60 Hz screen cannot show more than 60 FPS, however fast the card is. The score measures how fast this browser's WebGL runs this workload on this machine, so Chrome, Edge, Firefox and Safari give different numbers on the same computer. Compare only scores taken in the same browser at the same load, and read a run against your own other runs, never against another machine's number.

How to benchmark your GPU in 3 steps

1 · Set the machine up. Plug the laptop in — many laptops run the graphics card slower unplugged, even without battery saver — and set the power mode to Best performance (Windows) or turn off Low Power Mode (Mac). Close heavy apps and other browser tabs, especially anything else using the GPU such as a game or a video editor, and wait a minute after start-up for background tasks to settle. A benchmark measures whatever the machine is doing right now, so a video call or a cloud sync running beside it becomes part of your score.

2 · Run it three times at the same load. Pick Medium load and press “Start benchmark”, then leave the page alone for about 15 seconds: five seconds on the shader stage, five on geometry, five on particles. Then press it again twice more. Each run is added to the runs table below the tool, so you can compare them. Use the best or the typical result, not the first one — the first run is often lower while the browser warms up. Keep the load the same for all three runs: a High-load score and a Medium-load score are different numbers, because the load factor is part of the score.

3 · Read the three stages, not one number. The total GPU score is the three stage scores added together, and each stage score is its average FPS times the load factor times ten. A card whose shader stage is strong but whose geometry stage lags is shader-strong and geometry-weak — it will handle effects-heavy games better than model-heavy ones. Here is a worked example. A laptop averages 58 FPS on the shader stage at Medium load: the stage score is 58 × 2 × 10, which is 1,160. If it also averages 62 FPS on geometry (1,240) and 55 FPS on particles (1,100), the total GPU score is 1,160 + 1,240 + 1,100, which is 3,500. The 1% low matters as much as the average: a stage with a high average and a low 1% low stutters, even if the headline number looks good.

What the three stages measure

Shader: a Mandelbrot fractal, calculated fresh for every pixel on every frame. This is fragment work — the same kind of math a game spends on lighting, shadows and effects. A card that is slow here will struggle in effects-heavy scenes. Geometry: thousands of triangles rotating in a single draw call. This is vertex work — the kind a 3D scene spends on models and terrain. A card that is slow here will struggle in model-heavy scenes. Particles: thousands of points moving on screen, positioned in the vertex shader. This is the kind of work a game spends on smoke, sparks, rain and explosions. A card whose particle stage lags the other two is usually bandwidth-bound rather than compute-bound. No single stage tells the whole story, which is why the page reports all three instead of hiding them in one number.

Why your FPS caps at your refresh rate

The frames this page counts are the frames your screen actually shows, and a screen cannot show more frames per second than its refresh rate: a 60 Hz monitor caps the reading at 60 FPS, a 144 Hz monitor at 144 FPS, however fast the card is. That is why a strong card can score exactly 60 on every stage at Low load — the card is faster than the test can show at that load, and the reading is telling you about the screen, not the card. Re-run at High load: each frame is heavier, the frame rate usually drops below the ceiling, and the differences between cards become visible again. The 1% low is the more useful number when the average is capped: it shows the slow frames the average hides.

What affects the score

Thermal throttling: scores drop on every run. Let it cool — on an older laptop, dust in the fans or dried-out thermal paste is the usual cause. Power mode: every run is low. Windows: Balanced or Best performance. Mac: turn off Low Power Mode. Running on battery: lower than when plugged in. Plug in — many laptops run the graphics card slower unplugged, even without battery saver. Hardware acceleration off: the score is far lower than it should be, and the device panel shows a software renderer such as SwiftShader. Turn hardware acceleration on in your browser settings. Background apps: scores jump around. Close games, video editors, video calls and other tabs. The browser: different numbers on the same PC. Compare only in the same browser, ideally the same version. When a graphics card gets hot, it slows down to protect itself, so a score that falls on every run points to cooling, while one that is low every time points to a setting.

How to compare two computers fairly

Use the same browser on both, ideally the same version. Plug both in and set the same power mode. Pick the same load level on both — a High-load score cannot be compared with a Medium-load score. Close everything else, and wait a minute after start-up for background tasks to settle. Run the benchmark three times on each and compare typical results, not a single run. Note the graphics card name the browser reports, and check the real card in your system settings so you know what you are comparing. Expect some spread between runs — differences of a few percent are noise. A big gap between two machines with the same graphics card usually means one is throttling, set to a power-saving mode, or busy in the background.

Benchmarking a used PC before you buy

On a second-hand laptop or desktop, benchmarks show whether the machine performs like its spec says and whether it is overheating. Open your system settings and check the graphics card, memory and operating system match the listing. Run the benchmark three times in a row at the same load — scores that fall noticeably each run point to throttling, which often means a cooling problem. Then run the 20-second stress test and watch whether the frame rate holds steady: a card whose rate falls by more than 15% from the first 5 seconds to the last 5 is throttling under sustained load. Watch the canvas while it runs, too: flickering, strange colours or shapes that should not be there can mean a failing card. Finish with the display and input tests on this site — a fast graphics card in a machine with a dead screen corner or a failing keyboard is still a bad buy.

What this test cannot see

A web page only sees what the browser passes on. It cannot see your GPU temperature, clock speed, fan speed, VRAM size or the exact graphics card model, and it cannot tell a failing card from a driver fault — that is not its job, and it does not pretend otherwise. It cannot measure disk speed either: browsers have no access to raw disk performance. It cannot predict real gaming performance, which depends on the game engine, the drivers and the graphics card as much as on the raw rendering speed. And a 15-second benchmark will not reveal thermal throttling that only appears after 5 to 10 minutes of load — that is what the 20-second stress test is for, and even that is a quick check, not a replacement for a 30-minute native run. What it can do, honestly, is tell you how fast this browser renders this workload on this machine right now, and whether that speed holds up when you run it again.

Frequently asked questions

How does a browser GPU benchmark work?

It hands your graphics card three fixed jobs through WebGL — a shader job (a Mandelbrot fractal, calculated fresh for every pixel), a geometry job (thousands of triangles rotating in one draw call) and a particle job (thousands of points moving on screen) — and counts how many frames the card finishes per second on each. The score is the average frame rate times the load level times ten, added across the three stages, so the formula is printed on the page and the number can be checked. Nothing is downloaded and nothing leaves your browser: the work runs inside the page, and closing the tab stops it.

Is the stress test safe — can it damage my graphics card?

No. A browser stress test cannot bypass your hardware's own safeguards: a graphics card that gets too hot clocks itself down, and if that is not enough the driver or the machine protects itself. The test runs inside the browser's sandbox, lasts 20 seconds, and is lighter than most games — you are asking the card to do the same kind of rendering work a game or a video editor asks it to do. Expect fan noise and a warm laptop. If you would rather not have those, press Stop — the load ends the moment you do.

Why is my FPS capped at my monitor's refresh rate?

Because the frames this page counts are the frames your screen actually shows, and a screen cannot show more frames per second than its refresh rate: a 60 Hz monitor caps the reading at 60 FPS, a 144 Hz monitor at 144 FPS, no matter how fast the card is. That is a property of on-screen measurement, not of your graphics card. If the average sits exactly at your refresh rate on every stage, the card is faster than the test can show at that load — re-run at High load, which makes each frame heavier and usually drops the reading below the ceiling, where the differences between cards become visible again.

Why does the test show a different graphics card name?

The name comes from the browser, and browsers increasingly hide the exact model for privacy: Firefox, for example, reports the same generic name to every owner of a similar card instead of the real one. Anti-detect browsers and extensions substitute names too, and dual-GPU laptops or remote sessions can surface a different adapter — the integrated chip instead of the dedicated card, or a software renderer. When the name looks like a software renderer (SwiftShader, llvmpipe or Microsoft Basic Render), hardware acceleration is probably off in your browser settings. The result is unaffected: the score measures the real performance of whichever GPU renders the page.

Can I compare my score with another computer?

Only if you run the test the same way on both. Use the same browser, ideally the same version, on both machines. Plug both laptops in and set the same power mode. Pick the same load level — a High-load score cannot be compared with a Medium-load score, because the load factor is part of the number. Close everything else, and wait a minute after start-up for background tasks to settle. Run the benchmark three times on each and compare the typical result, not a single run. Differences of a few percent are noise. A big gap between two machines with the same graphics card usually means one is throttling, set to a power-saving mode, or busy in the background.

Why is my score different each time I run it?

Usually by a few percent: the test counts frames for five seconds per stage, and a busy machine gives a slightly different number every time. Bigger swings come from other programs using the graphics card in the background, a hot device, or a battery-saving mode — close what you can and run it again. The run-history table on this page keeps every run this session so you can see the pattern: the first run is often the lowest while the browser warms up, and a score that falls on every run points to cooling, while one that is low every time points to a setting. A score that jumps wildly between runs usually means something else was using the GPU at the same time.

Can a browser test show my GPU temperature or VRAM?

No — and any page that claims to is guessing. No browser API exposes the temperature sensor, the clock speed, the fan speed, the VRAM size or the exact model of a graphics card, on any platform. What a web page can see is the work the card does (frames per second), the renderer name the browser chooses to report, and the maximum texture size. That is enough to answer “how fast is this card, and does it hold up under load?” — but if you need temperatures and clocks, pair this test with a local monitor such as GPU-Z or MSI Afterburner, which read the hardware directly.

Does the test work on a phone or tablet?

Yes — the benchmark runs on any device with a modern browser: Windows, macOS, Linux, ChromeOS, Android and iOS, with nothing to install. On a phone it shows lower frame rates, because phone graphics chips are built for low power draw rather than raw rendering speed, and they throttle sooner. Start on Low load on a phone, and plug it in if you are testing on battery. Keep the tab in front for the whole run — browsers throttle background tabs, and the readings stop meaning anything. One honest limit: this page cannot show your phone's GPU temperature or VRAM — no web page can — so pair it with a local monitor if you need those numbers.

How this test works: every number is measured in your browser from WebGL rendering on a <canvas> — a Mandelbrot fragment shader, instanced triangle geometry and a point-sprite particle field, each run for about 5 seconds at a fixed internal resolution per load level. The score for each stage is its average FPS times the load factor (Low ×1, Medium ×2, High ×3) times ten, and the total is the three stage scores added together. Frames are counted with requestAnimationFrame, so on-screen FPS is capped at your monitor's refresh rate. No downloads, no sign-up and nothing uploaded.