GPU Test

What a GPU test measures

The GPU turns vertices, textures, and shader math into a finished frame. Vertex work builds geometry. Rasterization fills pixels. Fragment shaders color those pixels. Memory bandwidth feeds textures and buffers. A useful test keeps that pipeline busy and reports how many frames completed.

FPS is frames per second. Frame time is milliseconds per frame. Min FPS, or a 1% low when a HUD shows it, tells you about hitch better than the average. Stability is how often frame time stays consistent. The timed 3D rendering test shows FPS, frame time, and a verdict on the HUD. Volume Shader prints live FPS on the canvas while it ray marches a Mandelbulb.

Those numbers are graphics readings for this scene, this browser, and this complexity. They are not CPU scores, disk scores, or battery health. They do not prove a phone is fake. They do not report clock speed, wattage, or true VRAM size the way GPU-Z does on Windows. Treat them as a load test, not a hardware inventory.

• FPS: frames finished per second
• Frame time: milliseconds per frame on the 3D, shader, and VRAM HUDs
• Stability: how often samples stay near the mean FPS on the 3D test
• Not measured: CPU, disk, battery wear, fake phone detection, true VRAM capacity, sensor temperature

Which test matches the question

Pick the load that matches the bottleneck you care about. Mixing four FPS numbers into one score hides the pipeline. Volume Shader is ray marching. The 3D test is raster meshes and lights. The shader test is fragment arithmetic. The VRAM test is texture sampling. The four canvases, side by side, are on live GPU tests.

• General graphics check or timed score: 3D rendering test
• Classic volume clip or ray marched FPS: Run Volume Shader
• Fragment shading without a Mandelbulb: shader test
• Texture bandwidth and cache: VRAM test
• Hold the load until heat shows: GPU stress test
• Compare two cooled devices: GPU benchmark
• Prep the device first: how to test a GPU

WebGL, vsync, and why FPS caps

These tests use WebGL 1.0 so mid range phones still start. WebGPU can go further on some desktops. It is not required here. If a canvas fails, try another browser or check that WebGL is enabled before you assume the GPU is gone.

A 60 Hz or 120 Hz display will not show more FPS than its refresh. On the 3D, shader, and VRAM tests the HUD names that headroom. Raise the level until FPS falls. That drop is the GPU, not the panel. Chrome, Safari, Firefox, and Edge can disagree on the same device because compositors and GPU process limits differ. Write the browser down with the score. More on the counter: what FPS means.

Browser GPU test vs Furmark, 3DMark, and GPU-Z

An online GPU test lives inside the browser, WebGL, and the device power limits. Phones hit those limits fast. Furmark and Unigine still win for a long desktop burn in. 3DMark still wins for published game-like scores and driver comparison. GPU-Z, HWiNFO, and similar tools still win for clocks, VRAM size, and sensor readouts. MSI Afterburner still wins if you want overlay graphs on a Windows game.

Use this site when you want a result in half a minute on a phone or a PC you do not want to install on. DirectX, Vulkan, and Metal benches are different APIs. They are not interchangeable with a WebGL canvas. The longer split is in browser vs Furmark and 3DMark.

GPU test on a phone

Chrome or Safari is enough. Start Volume Shader on Standard, or the 3D test on Medium. Extreme or Ultra on a mid range Android can sit under 30 FPS. Heat will pull the number down after a minute. That is the SoC protecting itself, not a failed test. Turn off battery saver. Close other apps. Do not start on a phone that just finished a game.

Adreno, Mali, and Apple GPU parts all run these canvases through WebGL. The brand name does not change the method. A phone focused walkthrough is GPU stress test on a phone.