{"id":101565,"date":"2021-10-08T10:22:52","date_gmt":"2021-10-08T08:22:52","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/rezultaty-optimizaczii-chromium-realizovannye-proektom-renderingng"},"modified":"2021-10-08T10:22:52","modified_gmt":"2021-10-08T08:22:52","slug":"rezultaty-optimizaczii-chromium-realizovannye-proektom-renderingng","status":"publish","type":"post","link":"https:\/\/prohoster.info\/en\/blog\/news\/rezultaty-optimizaczii-chromium-realizovannye-proektom-renderingng","title":{"rendered":"Results of Chromium optimization implemented by the RenderingNG project","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"
Chromium developers have presented the first results of the RenderingNG project, launched 8 years ago, aimed at continuously improving the performance, reliability, and scalability of Chrome. <\/p>\n
For instance, the optimizations introduced in Chrome 94 compared to Chrome 93 led to an 8% reduction in page processing delays and a 0.5% increase in battery life. Given Chrome's user base size, these figures translate to more than 1,400 years of CPU time saved globally each day. Compared to versions from previous years, modern Chrome renders graphics over 150% faster and is six times less likely to crash in GPU drivers on problematic hardware. <\/p>\n
Among the methods implemented to achieve performance gains is the parallelization of rasterization operations of different pixels on the GPU and a more active distribution of handlers across different CPU cores (such as JavaScript execution, page scrolling processing, video and image decoding, and pre-rendering of content). A limiting factor in active parallelization is the increased CPU load, which results in higher temperatures and increased power consumption, making it crucial to achieve an optimal balance between performance and energy consumption. For example, when running on battery power, it is acceptable to sacrifice rendering speed, but scrolling processing in a separate thread cannot be compromised, as this would noticeably reduce user interface responsiveness. <\/p>\n
The technologies implemented in the RenderingNG project completely change the approach to compositing and allow for adaptive utilization of various optimization technologies for computations on the GPU and CPU related to specific parts of the pages, taking into account such factors as resolution and refresh rate, as well as the presence of support for advanced graphics APIs in the system, such as Vulkan, D3D12, and Metal. Examples of optimizations include the active use of GPU texture caching and rendering results of parts of web pages, as well as considering only the areas of the page visible to the user when rendering (there is no point in rendering parts of the page covered by other content). <\/p>\n
An important element of RenderingNG is also performance isolation when processing different parts of the pages, such as isolating computations related to displaying ads in iframe blocks, rendering animations, playing audio and video, scrolling content, and executing JavaScript.
<\/center> <\/p>\n
Implemented optimization techniques: <\/p>\n
\n
In Chrome 94, a mechanism called CompositeAfterPaint was introduced, allowing for the compositing of separately rendered parts of web pages and dynamically scaling the load on the GPU. According to data collected from users through telemetry, the application of the new compositing system has reduced scrolling latency by 8%, increased user interaction responsiveness by 3%, increased rendering speed by 3%, reduced GPU memory consumption by 3%, and extended battery life by 0.5%.\n
GPU Raster \u2014 a rasterization mechanism on the GPU side, was implemented for all platforms in 2020 and allowed for an average speedup of 37% in MotionMark tests, and 150% in HTML-related categories. This year, GPU Raster was enhanced with the ability to use GPU acceleration for rendering Canvas elements, resulting in a 1000% speedup in contour rendering and a 130% improvement in passing MotionMark 1.2 tests.\n
LayoutNG \u2014 a complete overhaul of the page element layout algorithms, aimed at improving reliability and predictability. The project is expected to be delivered to users this year.\n
BlinkNG \u2014 a refactoring and cleanup of the Blink engine, separating rendering operations into individually executed phases to enhance caching efficiency and simplify deferred rendering considering the visibility of objects in the window. The work is planned to be completed this year.\n
Separation into distinct threads for scroll, animation, and image decoding handlers. The project has been evolving since 2011 and this year has achieved the capability to offload animated CSS transformations and SVG animations to separate threads.\n
VideoNG \u2014 an efficient and reliable engine for video playback on web pages. This year, support for displaying protected content in 4K resolution has been implemented. HDR support was added earlier.\n
Viz \u2014 separate processes for rasterization (OOP-R \u2014 Out-of-process Raster) and rendering (OOP-D \u2014 Out of process display compositor), separating browser interface rendering from page content rendering. The SkiaRenderer process is also being developed, using platform-specific graphic APIs (Vulkan, D3D12, Metal). This change has reduced crashes due to graphics driver issues by six times. <\/ul>\n
![\"Results](\"\/wp-content\/uploads\/2021\/10\/494e5178556944a3a47f3f717488127c.jpg\")
<\/center> <\/p>\n