A benchmark score answers one question: how a machine did on one test, on one day. Whether a laptop is fast is three different questions — how it compares with the machine you just replaced, where its CPU sits against the wider market, and whether it keeps that speed when the load stays on for hours. Three free tools cover those questions between them, and each was built for a different one, so the tool you pick depends on which question you are asking.

The next step is short. Run UserBenchmark's one-minute whole-PC sweep first; it is a free download and it flags hardware problems, not just speed [1]. If your question is where the CPU stands in the market, look it up on the public PassMark chart and read the sample count before you read the score [4] [5]. If the laptop will spend long stretches under sustained load, run UNIGINE's Superposition, which watches GPU clocks and temperatures [6].

Three tools, three different questions

Each tool measures a different slice of "fast", which is why running all three tells you more than running any one of them longer.

UserBenchmark's installed tool does the whole sweep in under a minute — "Speed test your PC in less than a minute", by its own page [1]. The CPU side runs integer, floating and string tests; the GPU side runs six 3D game simulations; the drive side checks read, write, sustained write and mixed IO; the RAM side checks single and multi core bandwidth plus latency [1]. The tool then compares your results against other users with the same components [1], so a lopsided build — a strong CPU sitting behind a slow drive, for example — shows up as a gap between component results instead of hiding inside one blended number. Run it on the old machine too, and the per-component results show what the replacement bought.

PassMark's CPU charts answer the market question. The CPU Mark on cpubenchmark.net averages eight tests, including integer math, compression, prime-number search and encryption, with a simultaneous test running on every logical CPU [4]. The baselines behind the chart come from results users submit after running the same software, plus PassMark's internal testing [4] [5]. A higher score is better on all eight [4].

UNIGINE's Superposition covers the third question, the one a quick run cannot reach: whether the machine holds its speed under sustained load. The page bills it as an "Extreme performance and stability test for PC hardware: video card, power supply, cooling system", run "in stock and overclocking modes with real-life load" [6]. During a run it monitors GPU temperature and clock speed, and it includes a loop stress-test for stability checks [6]. The page's own minimums: an NVIDIA GeForce GTX 6xx or higher, an AMD Radeon HD 7xxx or higher, or an Intel HD 5xxx or higher, with 2 GB of video memory and 5 GB of disk space [6]. The Basic edition is a free download [6], and the current version, 1.1, dates to 2019 [6].

PassMark chart numbers are averages that move

A chart score is a moving average of submitted runs, so the sample count behind the bar matters as much as the height of the bar.

The churn is constant. About 500 new benchmark results arrive at PassMark every day, and the graphs absorb them nightly [5]. CPU Mark values for a model therefore vary over time; older popular models hold stable scores, while newer ones can move a lot in their first weeks on the market [5]. Chart inclusion itself has a floor of one submitted result, and the separate overclocked chart requires five [5].

The sample count sits behind each bar. Five or fewer submissions is classed as a high margin for error, twenty or fewer as medium, twenty-one or more as low [5]. The count appears when you hover over a bar, and PassMark's own note is that a higher sample size makes the value likelier to be accurate [5].

Laptop buyers get one more caveat from the same page: some CPUs have a configurable TDP, a thermal limit the manufacturer can set above or below the typical value, and portable systems can run restricted profiles — the test software even checks whether a machine is running on battery power before it runs [5] — so the CPU Mark may include such systems. The same chip can therefore sit at different chart values in a thin laptop and in a desktop tower. PassMark's own advice, written for tender documents, is to take a print-out of the graphs on the day the tender is released, because the values change over time [5]; a screenshot taken the day you compare applies the same logic.

Two sites share the UserBenchmark name

The name on the tool now points to two different products, and each site's claims are its own.

The installed tool lives at userbenchmark.com and is the snapshot described above [1]. At userbenchmark.org, the page describes "a three-stage benchmark in your browser: sustained CPU throughput, adaptive GPU raster pressure, and path-tracing performance", with a CPU stage that saturates all logical cores with continuous math operations for 60 seconds [3]. That site's own limits are stated plainly: it "does not directly test RAM or storage", it is "not the same as full stress-test tools like Prime95 or FurMark", and "a browser benchmark does not establish performance in every game or application" [3]. The two pages describe two different products, and neither mentions the other [1] [3], so check the domain before you run anything.

On reputation, the vendors speak for themselves. UserBenchmark's own FAQ attributes the criticism around the site to "a coordinated misinformation campaign" and states that it is "the only major PC hardware publisher that is not paid by brands" [2]; the same FAQ credits fifteen years of operation [2]. PassMark's chart notes take the opposite approach, printing their own error margins and minimum sample sizes instead of contesting critics [5]. Both are the vendors' own accounts.

The routine and the total cost

Three steps cover the three questions, and the only listed price in the routine belongs to an optional paid edition.

  1. Run the whole-PC sweep [1]. The judgment: the weakest component result should match what the laptop feels like day to day. The failure signal: one component lands far below the rest, which points at a specific part — a drive, a memory configuration — rather than a verdict on the laptop.
  2. Look up your CPU on the PassMark chart. The judgment: weight bars with twenty-one or more submissions, the "Low margin for error" class, ahead of bars with five or fewer [5]. The failure signal: the value on your screenshot from last week no longer matches the chart, because the average moved; re-read the current number and take a fresh screenshot on the day you compare [5].
  3. If long exports, renders or other sustained loads fill your week, download Superposition and run its loop stress-test in stock mode — the factory-setting run, not the overclocked one [6]. The judgment: GPU clocks and temperatures hold steady across the loop [6]. The failure signal: a run that fails or turns unstable points you to the components the test is built around — the video card, the power supply, the cooling system [6].

The cost components: steps 1 and 2 cost nothing — the sweep is a free download by its own page [1] and the charts are public pages [5]. For step 3, Superposition's Basic edition is a free download [6]; the editions table lists features per edition, so match the feature list to the edition you download [6]. The one listed price in this piece belongs to Superposition's Advanced edition: $19.95 (USD, tax not included), an edition aimed at overclockers [6].

Run all three and the question stops depending on a single number: the sweep compares this machine's components against the same components in other users' machines, the chart shows where its CPU sits in the market, and the stability loop shows whether the speed survives a long afternoon of real work.