CPU Benchmark
Measure how fast your processor runs real work: prime sieves, floating point maths, sorting, hashing, JSON and physics. The test runs once on a single thread and then on every core at the same time, so you see both single-core speed and multi-core scaling.
Close heavy apps and plug in a laptop for a fair score.
| Workload | Single-thread | Multi-thread | Scaling |
|---|
How to use the CPU Benchmark
- Close heavy apps and plug in a laptop so the CPU can run at full speed.
- Choose the test length and how many threads to use (Auto uses every logical core).
- Press Start benchmark and keep this tab in front until the progress bar finishes.
- Read your scores: single-thread shows one core, multi-thread shows all cores together.
- Compare runs in the history list, or Copy result to share it.
What the six workloads measure
| Workload | What it does | CPU strength it shows |
|---|---|---|
| Integer: prime sieve | Sieve of Eratosthenes up to 100,000 (checks that it finds 9,592 primes) | Integer maths, memory writes and branches |
| Floating point: Mandelbrot | Up to 100 iterations of z = z² + c for 6,912 pixels | Floating point multiply and add speed |
| Sorting | Sorts 10,000 random numbers with a comparison function | Branch prediction, function calls and cache |
| Crypto: SHA-256 | Hashes 16 KB with a full SHA-256 implementation (verified against the standard "abc" test vector) | 32-bit bit operations and rotations |
| Text & JSON | Builds 150 records, converts them to JSON and back, then counts words with a regular expression | Strings, memory allocation and garbage collection |
| Physics: N-body | Simulates the Sun and four outer planets for 1,500 time steps | Floating point division and square roots |
How the scores are calculated
Each workload first runs for 150 ms to let the browser's JIT compiler optimise the code, then runs again for the chosen time (0.5, 1.1 or 2.5 seconds) while the number of completed runs is counted. That gives operations per second for each workload.
- Workload ratio = operations per second ÷ the fixed reference speed for that workload.
- Score = geometric mean of the six ratios × 1,000. The geometric mean stops one very fast workload from dominating the total.
- Multi-thread score uses the same formula with the operations of all threads added together.
- Multi-core scaling = multi-thread score ÷ single-thread score. Efficiency = scaling ÷ number of threads.
The single-thread test also runs in a background Web Worker, so the page stays responsive and the result is not affected by page drawing.
Why scores differ between devices and browsers
- Turbo and power limits: one busy core can boost higher than eight busy cores, which is why efficiency is usually below 100%.
- Hybrid CPUs: chips with performance and efficiency cores scale less than their core count suggests.
- Battery and thermal state: laptops and phones on battery or running hot can score 30 to 50% lower.
- JavaScript engine: Chrome and Edge (V8), Firefox (SpiderMonkey) and Safari (JavaScriptCore) optimise code differently, so always compare in the same browser.
- Background tabs: browsers slow down hidden tabs, so keep the page visible until the test ends.
Frequently asked questions
How is the CPU benchmark score calculated?
Each of the six workloads runs as many times as it can in a fixed time and the result is converted into operations per second. Every workload is divided by a fixed reference speed and the six ratios are combined with a geometric mean, then multiplied by 1,000. Because the scale is fixed, a score twice as high means the CPU did the work twice as fast.
What is the difference between single-thread and multi-thread scores?
The single-thread score runs the workloads on one Web Worker, so it measures the speed of one core, which matters most for web browsing, office apps and many games. The multi-thread score runs the same workloads on one worker per logical core at the same time, which matters for video export, compiling code and 3D rendering.
Why is my multi-core scaling lower than my core count?
Logical cores from Hyper-Threading or SMT share the resources of a physical core, so they add roughly 20 to 30 percent rather than doubling speed. Many laptop and phone CPUs also mix fast performance cores with slower efficiency cores, and all cores share power and heat limits, so clock speeds drop when every core is busy.
Can I compare this score with Geekbench or Cinebench?
No. This test runs JavaScript in your browser, so the result depends on both the CPU and the browser's JavaScript engine. It is good for comparing devices, browsers, power modes and cooling with the same test, but the numbers are on a different scale from native benchmarks.
Why do my scores change between runs?
Modern CPUs change their clock speed constantly based on temperature, power mode and background tasks. Battery saver modes, a hot laptop, browser extensions and other open tabs all lower the score. Run the test two or three times with the laptop plugged in and use the best or average result.
Does the benchmark use real work or synthetic loops?
It uses real algorithms: a Sieve of Eratosthenes for integers, a Mandelbrot set for floating point maths, sorting 10,000 numbers, a full SHA-256 hash of 16 KB, building and parsing JSON, and an N-body planetary simulation. Each result is checked so the browser cannot skip the work.
Is this CPU test safe and private?
Yes. It runs for a few seconds to a minute, like any busy web page, and stops by itself. Nothing is installed or uploaded; your scores are stored only in your browser's local storage.