Back to ProjectsNexaBench · Device Diagnostics
NexaBench · in-browser diagnostics

Know what you're
actually running on.

CPU, GPU, memory, storage and real-DOM web workload stress tests that run entirely in this tab - single and multi-core throughput, rendering / export / AI compute on the GPU, RAM bandwidth, disk read/write, and a Speedometer-style browsing suite - plus a battery efficiency score measured from the drain those tests cause. Run everything for a full report, or drill into one test at a time.

Platform
-
Processor
-
Logical cores
-
Core topology
not exposed by browser
CPU frequency
not exposed by browser
Memory (approx)
-
GPU vendor
-
GPU
-
GPU adapter
-
GPU generation
not exposed by browser
GPU cores
not exposed by browser
Pixel ratio
-

Exact P/E-core counts, live CPU frequency, GPU core counts and vendor driver IDs are hidden by browser privacy sandboxes on many Intel, AMD, NVIDIA and Apple devices; NexaBench shows native WebGL/WebGPU/UA Client Hint data wherever the browser exposes it.

Run everything
CPU, GPU, RAM, storage, web, battery & display refresh - estimated ~25s
0% ·

CPU throughput

Runs blended integer + floating-point workloads for a fixed time on a single worker thread, then again split across every logical core - the same single-core / multi-core split industry benchmarks use.

StatusIdle - press run
Single-core
Integer ops/sec-
Floating point ops/sec-
Multi-core (4 threads)
Integer ops/sec-
Floating point ops/sec-
Scaling efficiency-
CPU score
--
not run

Runs entirely off the main thread via Web Workers, so the UI stays smooth while it runs. Score is relative (0–100), not an absolute index.

GPU compute & rendering

Three passes: an animated WebGL particle field (rendering), a 4K offscreen raster with pixel readback (export/encode workloads), and a WebGPU matrix-multiply compute shader (the core op behind AI/ML inference).

StatusIdle - press run
Renderer-
Rendering
Particles / FPS- · - fps
Export
4K raster FPS / readback- fps · - MB/s
AI / ML compute
Matmul throughput-
GPU score
--
not run

Memory bandwidth

Sequential write, sequential read and random-access passes over a large typed array, run in a worker so timing isn't skewed by UI work. On unified-memory systems this also reflects the memory pool the GPU shares.

StatusIdle - press run
Test buffer size-
Sequential write-
Sequential read-
Random access-
Memory architectureUnknown
Memory score
--
not run

Buffer size scales down on lower-memory devices to avoid tab crashes. VRAM figures are an approximation bound by the WebGL upload path, not raw silicon bandwidth.

Storage throughput

Writes and reads a real multi-megabyte file through the Origin Private File System - genuine disk I/O through your browser's sandboxed storage, not a synthetic estimate. Falls back to IndexedDB where OPFS isn't available.

StatusIdle - press run
Engine-
Test file size-
Sequential write-
Sequential read-
Storage score
--
not run

The test file is deleted immediately after the run. Numbers reflect your browser's storage layer, which sits below native disk speed on every platform.

Web browsing

Speedometer-style suite: DOM node churn, forced layout/reflow, simulated text input, list sort & filter re-rendering, and JSON serialize/parse - the same categories of work real web apps spend most of their time on. Combined via geometric mean, so one weak workload can't be masked by strong ones.

StatusIdle - press run
DOM churn (create/remove)-
Forced layout / reflow-
Text input simulation-
List sort & filter-
JSON serialize/parse-
Web score
--
not run

Runs directly against the real DOM, so it briefly uses the main thread like any actual web page would - this is the one test that isn't perfectly smooth by design.

Battery status

Reads live battery telemetry where the browser exposes it. Records the charge level the moment you start your first test, then scores how much it drained by the time you check back - a real efficiency reading from the load you just put on the device.

Support-
Charge level-
Charging-
Time to full / empty-
Session drain (since first test)-
30s drain sample-
Battery efficiency score
--
not run

Chrome and Edge on desktop and Android expose the Battery Status API. Safari, Firefox and iOS block it for privacy - check system settings instead there.

Extrapolated from a short sample, so it reads more extreme than steady-state use - accurate direction, noisy magnitude. Not counted in the overall device score.

Display

Resolution, pixel density and refresh rate read directly from the screen, plus full-screen colour panels for dead-pixel and uniformity checks.

Resolution-
Device pixel ratio-
Colour depth-
Colour gamut-
HDR-
Measured refresh rate-

Dead-pixel test cycles through full-screen red, green, blue, white, black and a grey gradient. Look closely across the whole panel on each colour for a spot that doesn't change, or banding on the gradient.

Click any swatch for a quick fullscreen check, or use "Dead-pixel test" for the full guided sequence.

Speakers & audio

Isolates left/right channels and sweeps frequency range so you can catch a blown driver, a dead channel, or a rattling enclosure.

440 Hz

Left / right only isolates each channel so you can confirm both speakers actually work and aren't swapped.

The sweep plays every audible frequency in order - most phone and laptop speakers roll off before 20kHz and below ~150Hz, that's normal. Listen for crackling, distortion, or a channel that cuts out partway through, which isn't.

Audio contextnot started
Output channels-

Device report

A summary you can screenshot or copy before making an offer, or before wiping a machine you're about to sell.

TestResultScoreStatus
CPU throughput--not run
GPU compute & rendering--not run
Memory bandwidth--not run
Storage throughput--not run
Web browsing--not run
Battery--not run
Display--not run
Speakers--not run
Overall device score
--
run tests to generate

Overall blends CPU (25%), GPU (25%), memory (15%), storage (15%) and web browsing (20%). Battery gets its own efficiency score but isn't part of this blend; display and speakers are diagnostic reads only.