Simulation Speed Test
Four classic physics kernels, run on your GPU and your CPU, in this page. Each measures a different bottleneck, which is why there is no single score.
Your results never leave your browser. No timings, no scores and no device details are uploaded. The page itself counts visits, like every page on this site; see the note at the foot.
Run it
Results
| Kernel | Bottleneck | CPU, 1 thread | GPU | GPU advantage | Relative |
|---|---|---|---|---|---|
| No results yet. | |||||
Your run log
Every run you do is kept in this browser only, so you can compare machines or see whether closing other apps changed anything. It is never uploaded, and clearing your browser data deletes it. Nothing in this log is visible to us.
What each kernel measures
| Kernel | Physics | Why it is in the suite |
|---|---|---|
| N-body gravity | Every particle attracts every other. Work grows as N squared. | Arithmetic throughput. Each value loaded gets used many times, so memory is not the limit. This is where a GPU looks best. |
| Heat diffusion | Five-point stencil on a grid, the classic Jacobi sweep. | Memory bandwidth. Each cell does almost no arithmetic per value loaded, so the bus is the limit, not the maths. |
| Monte Carlo | Random sampling to estimate pi. | Raw parallel width. No communication at all between workers, so it measures how many things your hardware can do at once. |
| Serial chain | A sequence where each step needs the previous answer. | Single-thread speed, and the honest counterweight. No GPU can help here, and real simulations always contain some of it. |
Your hardware, as this page sees it
Why one number would be a lie
A benchmark that prints a single score hides the thing worth knowing: hardware is fast at some shapes of work and slow at others. The four kernels above disagree with each other on purpose.