본문으로 건너뛰기

이 페이지는 아직 사용하시는 언어로 제공되지 않습니다. 영어 원문을 읽고 계십니다.

Blog

Measurements, written up

Every entry here comes out of a measurement we took and can show you: an acceptance run on the built engine, a before and after on the same machine, a count taken from the code rather than from memory. Where there is no measurement, there is no entry.

6개 글

  1. ARMANOS 팀7 min

    A guard that watches half the places

    Every number on this site is meant to come from a run you can repeat, and what holds that together is not care but the checks standing over it. Over two days four exclusions turned up inside those checks, and each exclusion was a permission for the place it skipped.

  2. ARMANOS 팀7 min

    What a green count does not prove

    Before a build of the browser ships, it runs a self-test, and the run ends in one line: so many readings out of so many. A full 29 of 29 still does not prove the noise inside is doing anything, so the verdict rests on a second thing as well.

  3. ARMANOS 팀6 min

    The hundred steps of a heap ceiling

    In a Chromium browser a page can ask how much memory its scripts are allowed to hold. The browser has two ways to answer that, and the rounded one puts the number on one of a hundred steps we computed from the engine's own source.

  4. ARMANOS 팀7 min

    Numbers that do not add up

    We read about a hundred pages across the showcases of this category. The part worth your time is not how large the numbers on them are, it is that the same quantity comes out at a different value on the next page.

  5. ARMANOS 팀6 min

    What a real machine reports about its GPU

    A profile can hand a site a graphics card name you pick from the list its platform offers. What it cannot hand over is the state of that card, and a rented Windows machine answered that second question in 14 rows without a single accelerated one.

  6. ARMANOS 팀6 min

    A window wider than the screen it claims

    A profile tells a site how big its screen is. The window it runs in is a separate number, and nothing forces the two to agree. We drew 900 profiles to see how often they disagree, and the answer was 64 percent.