Make Chrome Go Brrrrrr....
Table of Contents
Everything started with a simple question: “Why does my laptop get hot watching YouTube?”, and led to a rabbit hole of hardware acceleration, Chrome flags, and Intel iGPU monitors.
Alright 44°C isn’t that hot, but it’s the principle of the thing!
What is Hardware Acceleration?
Even though (Intel) CPUs are general purpose processors, they include specialized hardware and instructions to accelerate common tasks. This ranges from multi-purpose instructions like SSE for vector operations, to specialized instructions like AES-NI for efficient encryption.
My system features an 11th Gen Intel i7-11800H, which includes Intel QuickSync: “a dedicated video encoding and decoding hardware core”. On Linux, this functionality is exposed via LibVA an implementation of the Video-Acceleration-API (VAAPI).
With this hardware prowess available, my laptop should be more than fine to handle a YouTube video, right?!
Is Chrome using Hardware Acceleration?
The first step was to check if Chrome was using hardware acceleration at all.
Chrome uses Chrome URLs to provide (stupidly detailed) information about the browser.
For anyone looking for their next rabbit hole, you should check out chrome://chrome-urls/.
First stop, Chrome settings! This should let us know if hardware acceleration is enabled. Checking chrome://settings/?search=accel reveals 🥁🥁🥁
Hardware acceleration is enabled!
Okay, down to the next level, as there are several features hiding behind hardware acceleration, e.g., 2D drawing, composition and video de-/encoding, maybe Chrome is just not able to accelerate video decoding. Let’s continue our journey to chrome://gpu/.
Just to discover that video decoding is hardware accelerated. 🤨🤨🤨
So - any sane person might call it a day here, accept that the system runs a little warm, and move on with their life. But not me! We can go a level deeper!
Let’s check out chrome://media-internals/. This page is a treasure trove of information about media playback in Chrome. It shows media sessions, the codecs used, and their hardware acceleration status. Let’s play a video and see what we can gleam from this page.
We can see that the video uses the VP9 codec, for which our Tiger Lake CPU has hardware support, but what does kVideoDecoderName: VpxVideoDecoder mean?
Turns out LibVPX is “the reference software implementation for the VP8 and VP9 video coding formats” and there is “no hardware encoding or acceleration support of any kind”.
Switch to the Video Acceleration API (VAAPI)
Armed with this knowledge we can finally start to ask the right question: “How can we make Chrome use the VAAPI hardware acceleration?”
The answer hides deep in the Chromium Docs: “VA-API on Linux is not supported, but it can be enabled using the flags below”. Not what I wanted to hear, but what do we have to loose at this point? Let’s try it!
Even though the documentation states the flags --use-gl=angle --use-angle=gl are required,
they had no effect for my system. The GPU configuration in chrome://gpu/ stayed the same.
Adding the flags --enable-features=VaapiVideoEncoder,VaapiVideoDecodeLinuxGL on the other hand did the trick!
Power Consumption
What did we actually gain from all this? Battery life!
To actually quantify the difference, I measured the power consumption of my laptop using powerstat, while watching a YouTube video (1440p 60fps) in Chrome and Firefox (as a reference).
| Setup | Description |
|---|---|
| Chrome (VAAPI) | 12.5W ± 0.49W |
| Chrome (VPX) | 14.6W ± 0.86W |
| Firefox | 15.7W ± 1.94W |
This reduced my power consumption by 15%! 🎉🎉🎉
Conclusion
Use the following flags to enable VAAPI hardware acceleration in Chrome:
--enable-features=VaapiVideoEncoder,VaapiVideoDecodeLinuxGL
These if you are on Wayland:
--enable-features=VaapiVideoEncoder,VaapiVideoDecodeLinuxGL,UseOzonePlatform --ozone-platform=wayland
See Chrome flags in chrome://version/ to double check that flags are actually recognized.
Use chrome://media-internals/ to verify that hardware acceleration is used.
References
ArchLinux Forum and Wiki are amazing resources on the topic.
Use intel_gpu_top from igt-gpu-tools to
actually see the usage of your iGPU.
