Gustavo Garcia tests the results of the CPU usage of different encoders in the same test environment.
In WebRTC, recognized as excellent and most popular codecs are VP8 and H.264, but these two codec are not our only choice. VP9 has been available for a while, and some large services are using it, such as recent Chrome adds to AV1 support.
When comparing the codec, some interesting factors, such as interoperability and licenses, but the most important factors may be how the codec is the performance of compression, and the codec is cheap in CPU and memory usage. degree.
The compression ratio is usually our first thing to consider, and there are many comparisons available for this, but if we want to be able to use the codec for real-time use case, resource consumption is equally important.
Given that AV1 is available in the Chrome Canary version, I decided to run some tests to evaluate the four CPU usage of 4 available codecs in the WebRTC ecosystem. The purpose of this test is to compare the entire video pipe with these four codecs, not just individual comparison codecs.
test environment
These tests are completed by a simple web page, which has been connected between 2 Peerconnections (a send and another reception). If you want yourself to run test, see the test page:
https://jsfiddle.net/tvo7czxs/
Tests that use this page have changed 3 variables:
Codec: VP8, VP9, H264, AV1
Resolution: HD, VGA, QVGA
Bit rate: 200kbps, 800kbps, 2Mbps
If you look at the test page, you can easily change these 3 parameters to run tests in other configurations or other devices.
The use of the Chrome version is the latest version (1/2/21) from Git (1/21) this week, tested in MacBook Pro (2.4 GHz 8 Nuclear Intel Core I9).
In order to check the use of CPUs, after waiting for 30 seconds, view the CPU of the Chrome process consumed in the System Activity Monitor to provide time to provide a WebRTC bandwidth estimate and resolution / frame rate adaptive stability. When the result is 100%, it is indicated that the machine has 1 intact core.
No fancy, but I hope this can be easily understood.
In that environment, I have run several 36 parameters combinations, take the average value, and summarized in the following sections:
QVGA test results
For QVGA resolutions, the result is expected, where VP9 requires more CPUs than VP8, and AV1 requires 2 times that of VP8 is almost VP8. H.264 is a small CPU usage because it uses hardware acceleration for this purpose.
% of CPU Usage with Different Codecs
QVGA200KBPS800KBPS2MBPS
VP8182228
VP9202833
H.264101415
AV1364650
VGA test results
For VGA, the result is not very different, but at low bit rates, only VP9 can hold resolution, while the bit rate limit is increased to 2 Mbps, AV1 uses more than 1 core. The quality of H.264 at 200kbps is really poor, and the frame rate is very low, the blocking is also obvious. In this case, the adaptability of the Chrome browser is clearly unable to work.
HD (1280x720) test results
The results of HD are similar to the results of VGA, but AV1 cannot encode the original resolution, narrowing the resolution in all bit rate tests. The performance of H.264 at low bit rate is also very unbelievable, and the difference between VP8 and VP9 cost is much higher than VGA.
(In addition, the high-definition resolution AV1 often crashes because of the memory problem of MAC-related code, but maybe this bug has been fixed when you read this article)
Code VS decoding cost
I have conducted a test to divide costs between encoding (sender) and decoding (recipient). This test is for VGA at 800 kbps, and the test result is the result of the next four codecs that are considering.
The result is not large, but the decoding of VP9 and AV1x is relatively inexpensive than the coding.
Only the decoding cost is compared to different codecs, and it seems that the price of AV1 is 2 times more expensive than other decoders. The price of VP9 is slightly higher than the price of VP8, and the price of VP8 is slightly higher than H.264, but there is not much difference between the three.
Summarize
With a new codec, it is surprising, there is no doubt that av1 is the future of real-time video communication, but it looks patiently waiting for some time, so that it can be used in general video conferencing applications in the future. . At the same time, we may also use it for specific usage, such as broadcast, dedicated powerful devices, or encoding the low resolution versions of the video stream when using the network.
For other use cases, VP8 and VP9 look still the best choice unless you are not worried about the low bit rate, or you are using high resolution, and battery / CPU consumption is a big problem, you can consider H .264.
In addition, it is clear that the new libaom patch is about to be available, and performance can be increased by 15%, so repeating these tests on Chrome's future versions and different devices (AV1 may optimize ARM CPUS).
Editor: Lyn, read full article, original title: WebRTC video codec performance assessment
Article Source: [Micro Signal: Livevideostack, WeChat public number: microwave radio forum] Welcome to add attention! Please indicate the source of the article.
Our other product:
