Performance testing statistics show a persistent gap between desktop and mobile experiences. In 2024, 43% of mobile websites had good Core Web Vitals (CWV), compared with 54% of desktop websites. The figures below come from The Web Almanac by HTTP Archive — Performance | 2024, whose measurements combine CrUX real-user data with HTTP Archive lab data; the chapter identifies the dataset type where relevant. Percentages are website or page shares, and millisecond and kilobyte values retain the source units.
Contents
- Core Web Vitals by device and site group
- Loading speed and thresholds
- What drives Largest Contentful Paint
- Interactivity and responsiveness
- Long tasks and Total Blocking Time
- Visual stability and cache-related performance
Core Web Vitals by device and site group
The overall 2024 CWV pass rate was 43% for mobile websites and 54% for desktop websites. The metric choice matters: using FID rather than INP, 48% of mobile websites had good CWV, while using INP, 43% did. These are not interchangeable readings of the same result; they reflect different interactivity measures.
Site scale also changes the distribution. Among the top 1,000 websites, 40% of mobile pages and 54% of desktop pages had good CWV. Among the top 10,000, the shares were 33% and 46%. Among the top 100,000, they were 31% and 43%, while among the top 1,000,000 they were 36% and 48%, respectively. All of these figures are 2024 measurements and refer to mobile first, then desktop.
Page type provides another useful testing cut. Home pages reached good CWV on 45% of desktop pages and 38% of mobile pages in 2024. A performance test that reports only an aggregate site result can therefore hide meaningful differences between devices, site groups, and page types.
Loading speed and thresholds
The 2024 report classifies a Time to First Byte (TTFB) of 800 milliseconds or less as good. Even with that threshold, only 42% of mobile websites had good TTFB in 2024; 40% needed improvement and 19% were poor.
First Contentful Paint (FCP) was stronger on desktop. In June 2024, 68% of desktop websites had good FCP, compared with 51% of mobile websites. The report defines FCP below 1.8 seconds as good.
Largest Contentful Paint (LCP) shows a similar device gap. In 2024, 72% of desktop websites had good LCP, 20% needed improvement, and 8% were poor. Mobile websites recorded 59% good, 27% needing improvement, and 14% poor. The good LCP threshold was below 2.5 seconds.
Page type matters here as well. For home pages, 63% of desktop pages and 53% of mobile pages had good LCP in 2024. Secondary pages performed better: 82% of desktop pages and 72% of mobile pages had good LCP.
| 2024 LCP result | Desktop websites | Mobile websites |
|---|---|---|
| Good | 72% | 59% |
| Needs improvement | 20% | 27% |
| Poor | 8% | 14% |
These distributions are useful for setting test baselines, but they are not forecasts for a particular product. They describe website shares in the 2024 measurement chapter.
What drives Largest Contentful Paint
Images were the LCP element on 83.3% of desktop pages and 73.3% of mobile pages in 2024. Text was the LCP element on 16.3% of desktop pages and 26.3% of mobile pages. Inline images accounted for 0.3% of desktop pages and 0.4% of mobile pages.
The LCP sub-parts show how the bottleneck changes with performance quality. For origins with good p75 LCP in July 2024, median times were 600 ms TTFB, 350 ms image-load delay, 160 ms image-load duration, and 230 ms render delay. Origins needing LCP improvement had medians of 1,360 ms, 720 ms, 270 ms, and 310 ms for those same sub-parts. Poor p75 LCP origins reached 2,270 ms TTFB, 1,290 ms image-load delay, 350 ms image-load duration, and 360 ms render delay.
| Origin LCP group, July 2024 | TTFB | Image-load delay | Image-load duration | Render delay |
|---|---|---|---|---|
| Good p75 LCP | 600 ms | 350 ms | 160 ms | 230 ms |
| Needs improvement | 1,360 ms | 720 ms | 270 ms | 310 ms |
| Poor p75 LCP | 2,270 ms | 1,290 ms | 350 ms | 360 ms |
Discovery and prioritization remain common test targets. Thirty-five percent of mobile websites did not have an LCP element that was statically discoverable in the document in 2024. Sixteen percent of mobile websites lazy-loaded their LCP image: 9.5% used the native loading=lazy attribute and 6.7% used a custom approach. Conversely, 15% used fetchpriority=high on the LCP image, up from 0.03% in 2022.
Image weight also varies. Forty-eight percent of mobile pages had LCP images between 100 and 200 KB in 2024, while 8% of both desktop and mobile pages had LCP images larger than 1,000 KB.
Interactivity and responsiveness
A good Interaction to Next Paint (INP) result requires at least 75% of sessions to have INP of 200 milliseconds or less. In 2024, 97% of desktop websites had good INP, 2% needed improvement, and less than 1% were poor. Mobile results were 74% good, 24% needing improvement, and 2% poor.
Mobile INP improved across the measured years: good results rose from 55% in 2022 to 64% in 2023 and 74% in 2024. This is a historical comparison of reported shares, not a forecast.
The largest site groups did not all match the overall mobile result. Among the top 1,000 websites, 53% of mobile pages had good INP, 41% needed improvement, and 6% were poor. For the top 10,000, the shares were 49%, 44%, and 7%. For the top 1,000,000, they were 61%, 35%, and 4%. Among the top 100,000,000, they were 74%, 24%, and 2%.
Home and secondary pages were close but not identical: good INP was recorded for 96% of desktop home pages, 73% of mobile home pages, 96% of desktop secondary pages, and 72% of mobile secondary pages in 2024. In RUMvision data, presentation delay contributed 36 milliseconds to median INP, while input delay reached 155 milliseconds at the 90th percentile.
Long tasks and Total Blocking Time
The Long Tasks API treats a task longer than 50 milliseconds as a long task. In 2024, task duration was 61 ms on desktop and 71 ms on mobile at the 25th percentile. Median duration was 90 ms on desktop and 108 ms on mobile. At the 75th percentile it was 161 ms and 187 ms, and at the 90th percentile it was 331 ms and 377 ms, respectively.
Total Blocking Time (TBT) below 200 milliseconds is classified as good. The device difference was pronounced: at the 25th percentile, TBT was 84 ms on desktop and 417 ms on mobile. Median TBT was 67 ms on desktop and 1,209 ms on mobile. At the 90th percentile, it reached 718 ms on desktop and 5,955 ms on mobile.
| 2024 percentile | Task duration, desktop | Task duration, mobile | TBT, desktop | TBT, mobile |
|---|---|---|---|---|
| 25th | 61 ms | 71 ms | 84 ms | 417 ms |
| Median | 90 ms | 108 ms | 67 ms | 1,209 ms |
| 75th task duration | 161 ms | 187 ms | — | — |
| 90th | 331 ms | 377 ms | 718 ms | 5,955 ms |
The TBT values reinforce why a mobile performance test should inspect the tail of the distribution, not only a typical run. The supplied report gives the percentile measurements above; it does not establish a cause for every device-level difference.
Visual stability and cache-related performance
A Cumulative Layout Shift (CLS) score of 0.1 or less is good, 0.1 to 0.25 needs improvement, and above 0.25 is poor. In 2024, 72% of desktop sites had good CLS, 18% needed improvement, and 11% were poor. Mobile sites recorded 79% good, 12% needing improvement, and 9% poor.
The mobile share with good CLS rose from 60% in 2020 to 62% in 2021, 74% in 2022, 76% in 2023, and 79% in 2024. Desktop good CLS rose from 54% in 2020 to 62% in 2021, 65% in 2022, 68% in 2023, and 72% in 2024.
Caching and lifecycle behavior can affect repeat navigation. Among the top 1,000 sites, unload handlers made 35% of desktop pages and 27% of mobile pages ineligible for the back-forward cache (bfcache) in 2024. Twenty-one percent of sites used Cache-Control: no-store.
Layout dimensions remain a concrete stability check. Sixty-six percent of mobile pages had at least one image without explicit dimensions in 2024. At the 90th percentile, pages had 26 unsized images on desktop and 23 on mobile.
Together, these results suggest a practical test matrix: separate desktop and mobile runs; report CWV by page type; check TTFB, FCP, and LCP thresholds; identify the LCP element and its sub-parts; inspect INP and long-task percentiles; and test CLS, image dimensions, bfcache eligibility, and cache headers as distinct performance conditions. The measurements remain tied to the 2024 HTTP Archive chapter and should not be presented as 2026 estimates or forecasts.