Methodology

How BottleneckRadar measures browser bottlenecks

BottleneckRadar is built around practical browser diagnostics rather than laboratory hardware certification. The goal is to answer a narrower question: what is most likely limiting this browser session right now? That distinction matters because a browser can be affected by power mode, thermal throttling, extensions, graphics-driver settings, hardware acceleration, privacy restrictions, Wi-Fi quality, VPN routing, and other tabs running in the same profile.

CPU responsiveness

The CPU test uses Web Workers to run repeatable mixed workloads outside the main page thread. The workload combines integer operations, floating-point math, branching, and small memory accesses. The result includes a single-thread rate, a multi-thread rate, a scaling estimate, a sustained-performance ratio, and a main-thread responsiveness check. This is useful for spotting slow browser execution, weak parallel scaling, and performance drops during short bursts.

GPU rendering

The GPU test renders a deliberately heavy WebGL scene and measures frame rate, pixel throughput, and WebGPU availability. A low result can indicate weak integrated graphics, disabled hardware acceleration, remote-desktop rendering, driver issues, or a browser that is not using the expected GPU path. The score is not a game benchmark: games use different engines, APIs, shaders, and resolutions.

Memory and browser capability

The memory test allocates and reads browser-managed data to estimate memory throughput inside the browser sandbox. The browser capability test checks whether features required by modern web apps are available, including WebGL, WebGPU, worker support, secure context status, and online state. Browser vendors intentionally limit exposed hardware details, so BottleneckRadar reports unknown or hidden values as not exposed instead of guessing.

Jitter and bandwidth

The video call jitter test sends small same-origin timing requests and looks for latency variation and lag spikes. That is useful for diagnosing choppy voice or video calls. The full bandwidth test goes further by measuring download, upload, idle latency, jitter, and latency under load. It transfers generated data only; it does not upload user files.

Browser AI benchmark

The AI benchmark runs a real MobileNet v2 ONNX model locally in the browser. It compares CPU/WebAssembly inference with WebGPU acceleration when available and reports initialization time, median inference latency, p90 latency, and inferences per second. Model input is generated locally and inference output is not submitted to BottleneckRadar.

How to use the scores

Scores are designed to rank the current browser session, not to provide a universal hardware grade. A high CPU score and lower GPU score means graphics rendering is more likely to limit browser graphics workloads. It does not prove that a specific game is GPU-bound. For better comparisons, run tests with the same browser, power mode, display refresh rate, VPN status, and background workload.