Get informed about your system compute pressure.
The Compute Pressure API offers high-level states that represent the pressure on the system. It allows the implementation to use the right underlying hardware metrics to ensure that users can take advantage of all the processing power available to them as long as the system is not under unmanageable stress.
Current status
| Step | Status |
|---|---|
| 1. Create explainer | Complete |
| 2. Create initial draft of specification | Complete |
| 3. Gather feedback & iterate on design | In progress |
| 4. Origin trial | Complete |
| 5. Launch | Complete (Chrome 125) |
Try out the Compute Pressure API
To experiment with the Compute Pressure API locally, read this page.
Register for the origin trial
From Chrome 115, the Compute Pressure API is available as an origin trial. It is expected to end in Chrome 123 (May 29, 2024). Register for the origin trial.
Use-cases
The primary use cases enhanced by the current Compute Pressure API are video conferencing and video games.
These popular real-time applications are classified as soft. That is, the quality of service degrades if the system is exercised beyond certain states, but does not lead to a total system failure. These soft real-time applications greatly benefit from being able to adapt their workloads based on CPU consumption or pressure.
Specifically, the first version of this API aims to enable the following adaptation decisions.
Video conferencing
- Adjust the number of video feeds shown simultaneously during calls with many participants.
- Reduce the quality of video processing (video resolution, frames per second).
- Skip non-essential video processing, such as some camera filters.
- Disable non-essential audio processing, such as WebRTC noise suppression.
- Turn quality-versus-speed and size-versus-speed knobs towards "speed" in video and audio encoding (in WebRTC, WebCodecs, or software encoding).
Video games
- Use lower-quality assets to compose the game's video (3D models, textures, shaders) and audio (voices, sound effects).
- Disable effects that result in less realistic non-essential details (water, cloth, fire animations, skin luminance, glare effects or physical simulations that don't impact gameplay).
- Tweak quality-versus-speed knobs in the game's rendering engine (shadows quality, texture filtering, view distance).
Technically these can be accomplished by knowing thermal (for example, is the system being passively cooled) and CPU pressure states for the main thread and workers the site is using. System thermal state is a global state and can be affected by apps and sites other than the observing site.
Interfaces
The Compute Pressure API can be run in the following contexts:
- Window or main thread
- Dedicated Worker
- Shared Worker
The Compute Pressure API defines two new interfaces.
PressureObserver: An object to observe the compute pressure of any number of sources at a predefined sample interval. First iteration in Chromium exposes "cpu" as source. See section about parameters for more details. Each observer can asynchronously observe pressure changes trends in a system.
PressureRecord: Describes the pressure trend at a specific moment of transition. Objects of this type can only be obtained in two ways: as an input to your PressureObserver callback, or by calling the takeRecords() method on the PressureObserver instance.
PressureObserver
When a PressureObserver object is created, it's configured to watch the pressure of supported sources, at a given sample interval. The supported sources can be individually observed or unobserved at any time during the lifetime of the PressureObserver object. The sample interval cannot be changed after the creation of the object.
Constructor
PressureObserver(callback): Creates a new PressureObserver object which will invoke a specified callback function when it detects that a change in the values of the source being observed has happened.
The constructor takes a mandatory callback function.
Callback
callback(): The callback is called with an array of unread PressureRecord objects.
Methods
PressureObserver.observe(source, options): Tells the 'PressureObserver' which source to observe and optional options, as parameters.
PressureObserver.unobserve(source): Tells the 'PressureObserver' to stop observing a source.
PressureObserver.disconnect(): Tells the 'PressureObserver' to stop observing all sources.
PressureObserver.takeRecords(): Returns a sequence of records, since the last callback invocation.
static PressureObserver.knownSources() (read only): Return the user agent's known source types in alphabetical order.
Options
PressureObserverOptions: Contains the sample interval, sampleInterval in milliseconds, at which the user requests updates.
Parameters
source: The source to be observed, for example "cpu". This must be one of the supported source types.
In the current version of Compute Pressure, only "cpu" is supported.
PressureRecord
The PressureRecord interface of the Compute Pressure API describes the pressure trend of a source at a specific moment of transition.
Instance Properties
PressureRecord.source (Read-only): Returns a string representing the origin source from which the record is coming.
PressureRecord.state (Read-only): Returns a string representing the pressure state recorded.
PressureRecord.time (Read-only): Returns a number representing a high resolution timestamp.
Examples
The following sections list exemplary usage examples.
Determine API support
if('PressureObserver'inglobalThis){// The Compute Pressure API is supported.}Create a pressure observer
Create the pressure observer by calling its constructor with a callback function to be run whenever there is a pressure update:
constobserver=newPressureObserver((records)=>{/* ... */});Usage of the pressure observer
There is only one way to start a pressure observer. For each source call observer.observe(source).
observer.observe("cpu"{sampleInterval:2_000});In this example the "cpu" is the pressure source we are interested in. For now, it is the only source available. In the future, there may be other sources such as "gpu", "power" or "thermals".
A sample interval, sampleInterval, of 2000 ms, means that there will be updates at most every two seconds.
If the sample interval requested cannot be served by the system, the system will provide samples at the best suitable interval that exists. For example, if an interval of 2000 ms is requested, but the system can only provide samples at maximum 1000 ms, 1000 ms will be selected.
To stop observing a source, use the unobserve() method, as in the following example:
observer.unobserve('cpu');In order to unobserve all sources at once, use the disconnect() method, as in the following example:
observer.disconnect();Retrieve pressure records
Pressure records can be retrieved with a callback function, which will be invoked every time a change is happening in the pressure state.
functioncallback(records){constlastRecord=records[records.length-1];console.log(`Current pressure ${lastRecord.state}`);if(lastRecord.state==='critical'){// Reduce workers load by 4.}elseif(lastRecord.state==='serious'){// Reduce workers load by 2.}else{// Do not reduce.}}constobserver=newPressureObserver(callback);awaitobserver.observe('cpu',{sampleInterval:1_000});The user can also force the reading of PressureRecord by calling the takeRecords() method.
The takeRecords() method of the PressureObserver interface returns an array of PressureRecords objects stored in the pressure observer, emptying it out.
The most common use case for this is to immediately fetch all pending pressure records, not yet processed by the observer's callback function, prior to disconnecting the observer, so that any pending records can be processed when shutting down the observer.
Calling this method clears the pending records list, so the callback won't be run.
constobserver=newPressureObserver((records)=>{/* Do something with records. */});awaitobserver.observe('cpu',{sampleInterval:1_000});setTimeout(()=>{// Forced records reading.constrecords=observer.takeRecords();observer.disconnect();// Do something with last records if any.},2000);Share your feedback
Is there anything about the API that does not work as you expected? Do you see any missing method or property for your usage of the API? File a spec issue or comment on an existing one in the corresponding GitHub repo.
Report a problem with the implementation
Did you find a bug with Chromium's implementation? Or is the implementation different from the spec? File a bug at new.crbug.com. Be sure to include as much detail as you can, instructions for reproducing, and enter Blink>PerformanceAPIs>ComputePressure in the Components box.
