Reviewer and press kit
Test it, cite it, link it.
For laptop reviewers, tech writers, video makers and forum regulars: what SystemCheck measures, how to run it the same way on every machine, and how to cite a result so readers can check it.
What it measures
SystemCheck is a PC stress test that runs in a browser tab. It holds a fixed workload for a fixed time and records how throughput holds up across the run. The Standard run is 3 minutes and four stages:
- Sustained CPU load, 60 seconds
- Throughput in million operations a second, counted every 250 ms on every logical processor, with the decay over the run and the shape of the curve.
- Memory bandwidth and latency, about 20 seconds
- Bandwidth in GB/s from one thread and from all threads, and the round trip to memory in nanoseconds.
- GPU raster at a fixed 1600 by 900, 50 seconds
- Frames per second timed to GPU completion, frame pacing and decay, at the same resolution whatever the window size.
- Path tracing at a fixed 960 by 540, to the end of the run
- Samples per second and how evenly they were delivered, with the GPU loaded for the whole stage.
The Deep run is 9 minutes and eight stages: it adds the screen, a longer CPU hold, the GPU pass looped three times, CPU and GPU together, soft shadows and a hot re-test. The Extended run is Deep with ten minutes of CPU load, 16.5 minutes in all. Every stage, its length and its sampling are specified on the methodology page.
Run it the same way every time
- Pick the run for the question
- Standard (3 minutes) for comparing many machines on one table. Deep (9 minutes) or Extended (16.5 minutes) for a cooling or thermals review, where a slow slide can take minutes to show. Use one mode for every machine you compare.
- Plug in
- On battery most laptops apply a lower power limit. A battery run and a mains run on one machine are two different measurements, and the report records which it was.
- Close everything else
- Other programs, tabs and extensions compete for the same cores and GPU and can look like a cooling problem. The report flags a run that lost 3% or more of its working time to something else.
- Keep the tab in front and let it finish
- A hidden tab pauses the stages, and the report says so. Do not switch away, and let the run reach its report.
- Same browser, same power mode
- Note the browser and the operating system power mode, and keep both the same across machines.
- Run it more than once
- Quote the spread, not one run. A difference of a few percent between two machines can sit inside one machine's run-to-run spread.
How to cite a result
Link the report itself. A saved report has its own address, /report/ followed by its id and a view-only key, from the report’s share control. Use that link, not the one in a receipt email, which also opens the paid half. Quote:
- The verdict line and the curve shape, in the report's own words.
- CPU sustained throughput and its decay.
- GPU sustained frame rate and path tracing samples per second.
- Memory bandwidth on all threads, if memory matters to the review.
- The run mode, the browser, mains or battery, and the measurement revision printed on the report.
The measurement revision today is 2026.09.18, scoring v3. Reports from different revisions, or from different GPU backends, are not directly comparable, and CPU figures are not compared across kernels; the report names all three. The overall score and grade are provisional while calibration on more machines continues, so lead with the measured figures.
What it cannot measure
In every case the reason is the same: no browser API exposes the value. Pair SystemCheck with a hardware monitor when the review needs these.
- Temperature
- No browser API exposes any thermal sensor.
- Power draw
- Package power, board power and voltage are unavailable to a web page.
- Clock speeds
- No API returns the current or boost frequency of any core. Throughput is measured instead.
- Fan speed
- Fan controllers sit behind the embedded controller and the operating system.
- Graphics memory
- Neither WebGL2 nor WebGPU reports installed or allocated video memory.
- Throttle reasons
- The processor records why it lowered its clock and a page cannot read it, so a curve shape is described as consistent with a cause, never named as one.
- Drives and memory faults
- The run does not touch storage, and memory is measured for speed only. A good result says nothing about a failing drive or a faulty module.
- Drivers and firmware
- The browser reports a renderer string and nothing about the driver version or firmware.
The badge
If you used SystemCheck in a review, you may show this badge with a link to the site. It is optional, and how you mark the link is up to you. The image files are plain SVG; you can also download one and host it yourself.
<a href="https://systemcheck.live/"><img src="https://systemcheck.live/badge/tested-on-systemcheck-light.svg" alt="Tested on SystemCheck" width="208" height="40"></a><a href="https://systemcheck.live/"><img src="https://systemcheck.live/badge/tested-on-systemcheck-dark.svg" alt="Tested on SystemCheck" width="208" height="40"></a><a href="https://systemcheck.live/"><img src="https://systemcheck.live/badge/tested-on-systemcheck-light-small.svg" alt="Tested on SystemCheck" width="158" height="28"></a>[](https://systemcheck.live/)Tested on <a href="https://systemcheck.live/">SystemCheck</a>Tested on [SystemCheck](https://systemcheck.live/)To point readers at one result instead of the home page, use the same snippet with the report’s share link in place of the site address.
Using results and the logo
Results are free to publish, quote and chart, with attribution to SystemCheck and, where you can, a link to the report or the site. The measurements belong to whoever ran them. The SystemCheck mark (SVG) may be used to refer to the site; please do not alter it or suggest an endorsement.
Questions
For a question about a figure, a correction, or anything a review needs, use the contact page. The changelog records every change to what the test measures.
Measured, not marketed. Run it on the machine you are reviewing.
Run the free benchmark