A GPU benchmark measures how much graphics work your card can do and turns it into a score. This one runs in your browser, with no download: four short scenes, each aimed at a different part of the graphics chip, rendered through WebGPU or WebGL2 at a fixed 1280 x 720. It reports an overall score, plus the average and 1% low frame rate of each scene, and shows where the score sits next to our graphics card tier data. The score is our own scale, with an ordinary laptop as the reference: 1,000 is the Intel Iris Xe graphics in a 12th generation Core i7 laptop running Microsoft Edge. Use it to check that a card performs as it should after a driver update, to compare your laptop plugged in and on battery, or to see roughly what class of graphics you have. It works as a video card benchmark on any desktop card, laptop chip or phone, so you can run the same GPU benchmark test on each of your devices. It is not a replacement for 3DMark or in-game benchmarks, and the guide below explains why.
GPU benchmark test: how the four scenes work
Each scene draws its work as many copies (layers or particles) in a single draw call, so the work per frame is set by one number. For every scene the benchmark:
- Raises that number by 1.6 times per step until the frame rate falls below your refresh rate, then narrows the step down. That is the most work your GPU finishes in one refresh.
- Sets the load to 1.5 times that, so the graphics card is always busy and the browser is never waiting for the screen.
- Lets it settle for about a second, then times every frame for 8 seconds.
- Multiplies the load by the frames per second. That throughput (work per second) does not depend on whether your monitor runs at 60 or 240 Hz.
Each scene's score is its throughput compared with our reference laptop, which scores 1,000 on every scene. The overall score is the geometric mean of the four, so one very strong or weak scene cannot dominate it.
The four scenes
- Fractal (40 steps of a 2D folding fractal per pixel per layer): pure shader arithmetic, the kind of work behind lighting and post-processing in games.
- Volume (96 raymarching steps per pixel, up to 7 iterations of a power-8 Mandelbulb per step): long loops with branching, like volumetric fog, clouds and the "volume shader" benchmarks people search for.
- Particles (small glowing quads, positioned in the vertex shader): vertex processing, rasterising and additive blending.
- Fill rate (see-through full-screen layers): blending and memory bandwidth, which limit high resolutions and heavy transparency.
Worked example: reading a result
Say a gaming laptop scores 6,000. That is six times the throughput of the reference Iris Xe laptop. The reference scores 38 in our GPU tier table (GeForce GTX 1060 6GB = 100), so the rough tier equivalent is 6 x 38 = 228, the class of a card about twice as fast as a GTX 1060. If your browser reports the card's name, the result also shows its own tier score, so you can see whether the two agree. A result well below the card's tier score usually means a power limit, the wrong graphics chip, or a driver problem.
How to benchmark your GPU: step by step
Step 1: Close games, video and other tabs
Anything else that draws on the graphics card lowers the score. On a laptop, plug in the charger and use the best performance power mode.
Step 2: Press Start and keep the tab in front
The four scenes run one after another, about 15 seconds each. Each scene first finds how much your GPU can draw per frame, then measures for 8 seconds at a load that keeps it fully busy.
Step 3: Read the score and the scene table
The score compares your GPU with our reference laptop (Intel Iris Xe graphics = 1,000). The table shows the average and 1% low frame rate for each scene.
Step 4: Run it twice
A second run should land within a few percent. A big difference means something else was using the GPU, or the card got hot and slowed down; the GPU stress test shows that over minutes.
What is a good GPU benchmark score?
Our scale starts from an everyday laptop, so the score reads as "times faster than a thin laptop". Built-in graphics in recent thin laptops land around 1,000. A gaming laptop or a mid-range desktop graphics card is several times that, and high-end desktop cards are higher still, though at the top end the browser itself and the test's limits start to hold the score back. Two scores within about 5% of each other are the same result. For a ranking of cards by name, see the GPU tier list, and to compare two cards directly use the GPU comparison.
GPU benchmark vs GPU stress test
A benchmark asks "how fast?": it runs for about a minute and gives a score. A stress test asks "does it stay fast?": it runs for minutes at full load and watches whether the frame rate falls as the card heats up. If your benchmark score is lower than expected, run the GPU stress test to see whether the card throttles. For a 20 second check that the card renders correctly at all, start with the GPU test.
Why is my GPU benchmark score low?
- Wrong graphics chip. Laptops with two GPUs may run the browser on the built-in one. In Windows, open Settings, System, Display, Graphics, choose your browser and set it to High performance.
- Battery or power mode. On battery, laptops cut the graphics power limit sharply. Plug in and choose the best performance mode.
- Hardware acceleration off. If the browser draws with software, the score measures the processor. The WebGL test shows whether acceleration is on.
- Old or broken driver. Update it from the maker's site. Our graphics driver finder points to the right download for your card.
- Heat. A clogged cooler makes the card slow down within seconds of full load.
- Something else is busy. Video playback, a game in the background or other browser tabs all share the GPU.
Can I compare this score with 3DMark or other GPU benchmarks?
No. Every benchmark has its own scenes and scale. 3DMark, Unigine Superposition and in-game benchmarks run native code with the full graphics API and measure at fixed settings. A browser adds its own layers: on Windows, WebGL and WebGPU are translated to Direct3D, and the browser's compositor shares the GPU. Some cards gain or lose more from that than others. Compare this score with itself: the same PC before and after a change, or friends running the same browser.
How the score maps to our GPU tier data
Our GPU tier table rates 241 graphics cards and chips on one scale, with the GeForce GTX 1060 6GB at 100. The reference laptop's Iris Xe scores 38 there. The benchmark multiplies that by your score / 1,000 to give a rough tier equivalent. This assumes browser throughput grows in step with game performance, which is only roughly true, so the result is labelled as a class, and it is left out when a scene hit the test's limits or the browser used software rendering. If the browser names your card, you also see its own tier score; to find your card's name, see what GPU do I have.
Browsers, WebGPU and WebGL2
WebGPU is the newer graphics interface for the web and is available in recent Chrome and Edge on Windows, macOS and ChromeOS, and in recent Safari. Where it is missing, the benchmark uses WebGL2 with the same scenes. The two can score differently on the same PC, so the result names the one it used, and you can choose under More options. To see what your browser supports, run the WebGL test.
Why we built this GPU benchmark
The question behind a GPU benchmark is usually simple: is my graphics card as fast as it should be? The best-known benchmarks answer it with a download, several gigabytes for 3DMark, and a score that only means something next to their own database. A simple browser test has the opposite problem. It draws a fixed scene, so every card fast enough to reach the monitor's refresh rate shows the same 60 or 144 fps, and the number tells you nothing.
So this benchmark keeps raising the load until your GPU falls behind and scores the work done per second, which the refresh rate cannot cap. The score is tied to a laptop you can picture, Intel Iris Xe graphics at 1,000, and the page says plainly what a browser cannot see. Nitin Raja, who builds this site, tried a lot of GPU benchmark sites and downloads before building this one, and kept only what people need: one minute, a score you can read, the frame rates behind it and the method.
What this GPU benchmark measures
The only setting is the graphics API, Auto by default. The page asks the browser for a WebGPU adapter, or a WebGL2 context where WebGPU is missing, and draws our own four scenes at a fixed 1280 x 720. It times every frame with requestAnimationFrame and performance.now() and keeps at most two frames queued (a fence in WebGL2, onSubmittedWorkDone in WebGPU), so the timing follows the GPU, not the browser. The load search and the scoring are explained in how the four scenes work.
The card name in the result comes from the browser's WEBGL_debug_renderer_info extension. The reference is one laptop we measured in October 2026: Intel Iris Xe graphics (96 execution units) with a Core i7-12700H, Windows 11 and Microsoft Edge on WebGPU. The tier equivalent uses our table of 241 graphics cards and chips, last checked in October 2026.
What it does not measure: temperature, clock speed, power draw, fan speed, video memory, ray tracing or frame rates in real games, because browsers do not expose any of that. 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, so on a laptop with two graphics chips the score is usually the built-in chip's until you set the browser to High performance in Windows graphics settings. Our reference laptop also has a GeForce RTX 3050 Ti, and Edge still ran on the Iris Xe.
How accurate is an online GPU benchmark?
- Repeatable, not absolute. Two runs on the same PC usually land within a few percent. Different browsers, driver versions and power modes change the score.
- Refresh rate does not cap the score. The load is raised until the GPU cannot keep up, and the score uses work per second.
- Very fast cards hit limits. If a scene reaches the most work the test can draw while still at the refresh rate, the score for that scene says "at least" and the real number is higher.
- No sensors. A browser cannot read clocks, temperature or power, so it cannot tell you why a score is low; the checklist above covers the usual reasons.
- One reference machine. The scale is anchored to one laptop we measured. That makes it concrete, but it is not a lab average.
Why this GPU benchmark is different
Each of these points can be checked on this page:
- Nothing runs until you press Start. No score is filled in beforehand, and the run stops by itself when you switch tabs.
- The refresh rate does not cap the score. The load rises until the GPU falls behind, so a 60 Hz and a 240 Hz monitor give the same result on the same card.
- A reference you can picture. 1,000 is the Iris Xe in a thin laptop, so 3,000 means three times that laptop.
- It says when the number is not the whole story. A scene that reaches the test's limit shows "at least", and software rendering is flagged as a score for the processor.
- The method is on the page. Load search, timing, reference machine and scoring are all described above.
- No account, nothing uploaded. The graphics engine makes no network requests; the result stays on your device unless you copy it.
This browser benchmark vs native benchmarks
| This browser benchmark | Native benchmarks (3DMark, Unigine) | |
|---|---|---|
| Install | None | Download, several GB for 3DMark |
| Time | About a minute | A few minutes per test |
| Scenes | Four focused scenes, 1280 x 720 | Full game-like scenes at set resolutions |
| Score | Our own scale, Iris Xe laptop = 1,000 | Their own scales, with large public databases |
| Per scene frame rates | Average and 1% low | Yes, plus detailed graphs |
| Ray tracing | No | Yes (3DMark Port Royal, Speed Way) |
| Works on | Any browser with WebGL2 or WebGPU, including Macs, Chromebooks and phones | Mostly Windows; 3DMark has mobile apps |
| Cost | Free | 3DMark basic edition is free; full tests are paid. Unigine has free editions |
Sources
- W3C, WebGPU specification; Khronos, WebGL 2.0 specification (instanced drawing, fences).
- Intel Xe on Wikipedia, the graphics architecture of the reference chip (Iris Xe, 96 execution units).
- MDN, requestAnimationFrame and the Page Visibility API.
- Our GPU tier scores: compiled from spec sheets and published reviews, GeForce GTX 1060 6GB = 100.
GPU benchmark FAQ
How do I benchmark my GPU online?
Press Start on this page. It runs four short graphics scenes through WebGPU or WebGL2 in your browser, about a minute in total, and gives a score with average and 1% low frame rates. No download or account is needed.
Is a browser GPU benchmark accurate?
It is repeatable on the same PC and browser, usually within a few percent. It is not a precise ranking of graphics cards: browsers add overhead, very fast cards can hit the test's limits, and drivers and browser versions change the result. Use it to compare your own runs, or as a rough class.
Can I compare this score with 3DMark?
No. 3DMark, Unigine and other native benchmarks use their own scenes and scales. This score is ours: 1,000 is an Intel Iris Xe laptop in Edge. Two cards can rank differently here than in native tools, especially very fast ones.
What is a good GPU benchmark score here?
Around 1,000 is entry-level built-in graphics, like the Iris Xe in thin laptops. Gaming laptops and desktop graphics cards score several times higher. Look at your score next to the tier equivalent and the GPU tier list rather than chasing a number.
Why is my score lower than expected?
Common reasons: the laptop is on battery, the browser runs on the built-in graphics instead of the dedicated card, hardware acceleration is off, another app is using the GPU, or the card is hot. Check each, then run again.
What is 1% low fps?
The frame rate that 99% of frames beat. It shows how even the frame pacing is: a 1% low far below the average means stutter, even when the average looks good.
Is benchmarking bad for my graphics card?
No. A minute at full load is less than a gaming session. Graphics cards lower their clocks by themselves if they get too hot.
Related tools
GPU testA quick health check: renders correctly, runs smoothly, no artifacts
GPU stress testLoad your graphics card for up to 10 minutes and watch for throttling
Volume shader testA raymarched 3D fractal cloud with a live frame rate
WebGL testWebGL 1, WebGL 2 and WebGPU support, renderer and limits
Bottleneck calculatorWhich part limits first, CPU or GPU, at 1080p, 1440p and 4K
Can my PC run itCheck popular games against your graphics card, CPU and RAM, no download