If you are a mac-user, you might have used both browsers, and it doesn’t take long. Google Chrome and Safari are two popular browsers. MacBook battery life has been a huge feature for Apple in.
![]() 59% 2388 MHz: 1.0% 2592 MHz: 0% 2772 MHz. 79% 1956 MHz: 1.2% 2184 MHz. 42% 1056 MHz: 1.4% 1284 MHz: 3.0% 1500 MHz. It is most useful for getting CPU, GPU and wakeup measurements in a precise and easily scriptable fashion (unlike Activity Monitor and top)Below is a sample log from powermetrics output after running the following command:Sudo powermetrics -i 1000 -samplers cpu_power,gpu_power -a -hide-cpu-duty-cycle -show-usage-summary -show-extra-power-info -u ~/powerlogs.txt*** Sampled system activity (Thu Feb 11 00:10:50 2021 -0800) (1004.97ms elapsed) ***E-Cluster HW active residency: 11.15% (600 MHz: 2.8% 972 MHz: 88% 1332 MHz: 2.5% 1704 MHz: 3.0% 2064 MHz: 3.5%)P-Cluster HW active residency: 2.34% (600 MHz: 90% 828 MHz. Powermetrics is a Mac-only command-line utility that provides many high-quality power-related measurements. Peripherels attached: Apple Magic Keyboard and Magic Trackpad (Bluetooth)Power Consumption was measured using Apple’s powermetrics utility. Photo editor online for macI also tested Chrome in software-decode mode by forcing the hardware VP9 decoder off using the Chrome flag at chrome://flags/#videotoolbox-vp9-decoding.This gave us the following test matrix: PlayerTo compare against idle state, there was an idle period of 60 seconds before and after playing the video in both scenarios. Therefore the video is played using the software-based decoder. VLC currently cannot use the hardware VP9 decoder. For our purpose, I extracted the following fields: Extracted FieldTotal power consumption of the 4 high-efficiency coresTotal power consumption of the 4 high-performance coresTotal power consumption of the M1 chip including the CPU, GPU, DRAM and other components like Thunderbolt and SSD controller, media decoding engines, ISP, Secure Enclave etc.For an overview of the M1 architecture, see Apple’s official product page and the press release.If you’re interested in exploring the code that was used to parse the logs, visit the following Github repoI tested the following video - “Japan in 8K” - while streaming from YouTube in Safari, Chrome browsers and local file in VLC player.I included the VLC player in this test to compare against the local playback. 05% 528 MHz: 0% 720 MHz: 1.5% 924 MHz: 0% 1128 MHz: 0% 1278 MHz: 0%)GPU requested frequency: (396 MHz: 0% 528 MHz: 0% 720 MHz: 1.5% 924 MHz: 0% 1128 MHz: 0% 1278 MHz: 0%)I wrote a custom Python script to parse out these logs and plot the graphs below. 36%)GPU active residency: 1.51% (396 MHz. Chrome Or Safari On For Better Battery Life Mac OS EME PipelinePower Consumption for Chrome software decoding is more efficient than VLC software decoding □Looking at the CPU and GPU usage leads to some interesting observations: Power Consumption for Chrome hardware decoding = 2x Safari hardware decoding □ Surprisingly, Chrome’s power consumption is higher than Safari’s while both are decoding hardware decoding of VP9 codec.EDIT 5/21/21: This tweet explains why Chrome’s hardware decoding might have a higher power consumpton.I am not surprised chrome consumes more power than safari because chrome likely isn’t using the OS low power video pipeline for security or lack of access.For example, Chrome doesn’t have access to Safari / Mac OS EME pipeline because different DRM systems.— gurupanguji May 20, 2021Looking at the average and total power consumption, I made the following observations: ![]() The footage in this test was played for 60 seconds.The are two interesting parts to Netflix streaming between these browsers: Just like the last test to compare against the idle state, there was an idle period of 60 seconds before and after playing the video in both scenarios. Video not playing).For this test, I streamed Queen’s Gambit from Netflix in both Safari and Chrome browsers. With VLC, the GPU is consistently around 715 MHz, even in the idle phase ( i.e. The average frequencies are almost neck and neck for hardware decoding modes ( except GPU) and higher in software decoding modes. It’s most likely because Safari is streaming 1080p vs. On the M1, the DRAM usage in Safari is really high! It’s more than both the CPU clusters. Safari streamed this title in H.265/HEVC codec while Chrome streams in VP9.Netflix Queen’s Gambit streaming on SafariNetflix Queen’s Gambit streaming on ChromeOverall, both browsers end up being similar in the average and total power consumption numbers.However, once you dig down into individual components, we notice some interesting details. Netflix states that to stream in Ultra HD, one needs a device with an Apple T2 Security chip.Netflix requirements for streaming in Ultra HD on Mac The performance cluster runs around 600 Mhz most of the time but has periodic jumps to almost 3.2 GHz ( max frequency for the cluster as confirmed by Anandtech).Hopefully, that was an insightful look into the M1 power characteristics! Stay tuned for Part 2, where I’ll dig into similar details but this time with local playback.Also, I hear Big Sur 11. In Safari, the GPU is clocking up and down (zoom in on the details), while in Chrome, the GPU is constantly running around 710 MHz. GPU usage in Chrome is much higher.Some interesting things are going on with the GPU on the frequency charts. It’s the component that consumes the most power.No surprises looking at the usage charts than what we already gleaned from the power consumption charts i.e. Just like the YouTube test, Chrome seems to be leveraging quite a bit of GPU. This is most likely due to the T2 chip being leveraged because of hardware DRM implementation.
0 Comments
Leave a Reply. |
AuthorGeorge ArchivesCategories |