Extreme GPU usage while Element app is visible

[Mershl]

Steps to reproduce

1. Install Element from .rpm or flathub.
2. During Login no excess CPU usage is visible
3. Login
4. While the Element app is now visible an excessive CPU usage (full load on one core) as long as it's visible
5. Minimize the Element app
6. CPU usage of Element app drops back to 0%
7. Open Element app back up.
8. CPU usage of Element app raises to 100%

What happened?

Element app has a constant, very high CPU load which is noticable by running fans and the machine heating up.

What did you expect?

Element app should not show a noticeable CPU usage while it's idling on the desktop.

While Element app is visible:

While Element app is minimized the CPU load drops to 0%.

Operating system

Fedora 34 (GNOME 40.4 wayland, rpm and flathub tested) and ArchLinux (GNOME 40.5 Xorg, flathub tested)

Application version

reproduciable on 1.8.5 and 1.9.0

How did you install the app?

tested rpm and flathub, seperatly on a clean installation of Fedora and ArchLinux

Homeserver

matrix.org

Have you submitted a rageshake?

Yes

[germain-gg]

Would you be able to share a performance profile of the renderer process so that we can investigate this issue further.
Thank you for sharing a /rageshake, but unfortunately those logs do not contain the rigth information for this kind of problem.

You can do so by opening the developer console in the electron app and recording a performance profile

[Mershl]

Would you be able to share a performance profile of the renderer process so that we can investigate this issue further. Thank you for sharing a /rageshake, but unfortunately those logs do not contain the rigth information for this kind of problem.

You can do so by opening the developer console in the electron app and recording a performance profile

Yep, will do. Recorded 20sec while Element app idles. htop confirmed the issue during the recording.

Performance profile:
Profile-20211004T173606.zip

[germain-gg]

It appears that your CPU is fine, but something obliterates your GPU.
I'm going to close this ticket as it appears to be a duplicate of #19217 which experiences the same problem as you.

Could you comment on there answering the questions that were asked

[Mershl]

@gsouquet The issue seems related, but different as it happens in Firefox, not Chromium (which Element desktop is relied on). The other issue seems fixed when enabling WebRender.

[SimonBrandner]

@gsouquet, I think this manually needs to be moved out of Closed?

[Mershl]

Tested again with latest version (1.9.1). The issue is still reproduciable.

If it helps the following lines are printed to console on startup. Is the libva line interesting?

[2 preload-host-spawn-strategy] Running: /app/bin/zypak-helper child - /app/Element/element-desktop --type=zygote
/var/home/mershl/.var/app/im.riot.Riot/config/Element exists: yes
/var/home/mershl/.var/app/im.riot.Riot/config/Riot exists: no
Gtk-Message: 18:01:41.671: Failed to load module "canberra-gtk-module"
Gtk-Message: 18:01:41.671: Failed to load module "canberra-gtk-module"
Starting auto update with base URL: https://packages.element.io/desktop/update/
Auto update not supported on this platform
Fetching translation json for locale: en_EN
Changing application language to en-us
Fetching translation json for locale: en-us
Could not fetch translation json for locale: 'en-us' Error: Cannot find module './i18n/strings/en-us.json'
Require stack:
- /app/Element/resources/app.asar/lib/language-helper.js
- /app/Element/resources/app.asar/lib/tray.js
- /app/Element/resources/app.asar/lib/electron-main.js
- 
    at Module._resolveFilename (internal/modules/cjs/loader.js:887:15)
    at Function.n._resolveFilename (electron/js2c/browser_init.js:257:1128)
    at Module._load (internal/modules/cjs/loader.js:732:27)
    at Function.f._load (electron/js2c/asar_bundle.js:5:12913)
    at Module.require (internal/modules/cjs/loader.js:959:19)
    at require (internal/modules/cjs/helpers.js:88:18)
    at AppLocalization.fetchTranslationJson (/app/Element/resources/app.asar/lib/language-helper.js:76:20)
    at /app/Element/resources/app.asar/lib/language-helper.js:89:39
    at Array.forEach (<anonymous>)
    at AppLocalization.setAppLocale (/app/Element/resources/app.asar/lib/language-helper.js:88:17) {
  code: 'MODULE_NOT_FOUND',
  requireStack: [
    '/app/Element/resources/app.asar/lib/language-helper.js',
    '/app/Element/resources/app.asar/lib/tray.js',
    '/app/Element/resources/app.asar/lib/electron-main.js',
    undefined
  ]
}
Resetting the UI components after locale change
Resetting the UI components after locale change
libva error: vaGetDriverNameByIndex() failed with unknown libva error, driver_name = (null)
Changing application language to en-us
Fetching translation json for locale: en-us
Could not fetch translation json for locale: 'en-us' Error: Cannot find module './i18n/strings/en-us.json'
Require stack:
- /app/Element/resources/app.asar/lib/language-helper.js
- /app/Element/resources/app.asar/lib/tray.js
- /app/Element/resources/app.asar/lib/electron-main.js
- 
    at Module._resolveFilename (internal/modules/cjs/loader.js:887:15)
    at Function.n._resolveFilename (electron/js2c/browser_init.js:257:1128)
    at Module._load (internal/modules/cjs/loader.js:732:27)
    at Function.f._load (electron/js2c/asar_bundle.js:5:12913)
    at Module.require (internal/modules/cjs/loader.js:959:19)
    at require (internal/modules/cjs/helpers.js:88:18)
    at AppLocalization.fetchTranslationJson (/app/Element/resources/app.asar/lib/language-helper.js:76:20)
    at /app/Element/resources/app.asar/lib/language-helper.js:89:39
    at Array.forEach (<anonymous>)
    at AppLocalization.setAppLocale (/app/Element/resources/app.asar/lib/language-helper.js:88:17) {
  code: 'MODULE_NOT_FOUND',
  requireStack: [
    '/app/Element/resources/app.asar/lib/language-helper.js',
    '/app/Element/resources/app.asar/lib/tray.js',
    '/app/Element/resources/app.asar/lib/electron-main.js',
    undefined
  ]
}
Resetting the UI components after locale change

[tycho-kirchner]

Hi. Maybe I can add something to this (also running Element: 1.9.8):
On my systems the high load disappears after approx. 15 seconds but the other symptoms are the same.

When I launch with element-desktop --hidden there is no noticable load.
As soon as I make the window visible there is a considerable amount of cpu-load for more than 10 seconds with in the following processes (see [1]). The load occurs in the desktop app, in firefox and google-chrome regardless of hardware acceleration enabled or disabled. The load is not dependent on the selected chat and even a later switch to a chat with a lot of media does not trigger the symptoms again. Running on Debian Buster and tested on the following hardware:

Device-1: Intel HD Graphics 520 driver: i915 v: kernel 
Display: x11 server: X.Org 1.20.4 driver: intel resolution: 1600x900~60Hz 
OpenGL: renderer: Mesa DRI Intel HD Graphics 520 (Skylake GT2) v: 4.5 Mesa 18.3.6

Device-1: Intel Xeon E3-1200 v2/3rd Gen Core processor Graphics driver: i915 v: kernel 
 x11 server: X.Org 1.20.4 driver: modesetting resolution: 1920x1080~60Hz 
renderer: Mesa DRI Intel Ivybridge Desktop v: 4.2 Mesa 18.3.6 

Device-1: Intel driver: i915 v: kernel 
Display: x11 server: X.Org 1.20.4 driver: modesetting unloaded: fbdev,vesa 
resolution: 1368x768~60Hz 
OpenGL: renderer: Mesa DRI Intel HD Graphics 500 (Broxton 2x6) 
v: 4.5 Mesa 18.3.6

The symptoms occur only in element, not on the many other websites I visit with firefox or google chrome.

It is always reproducible so please, what info can I provide in order to solve this?
A shoot in the dark: I think something is loaded/polled for in the background with an unreasonably high loop-frequency.

[1]

tycho     4576 42.7  2.3 57053424 181532 pts/1 Sl+  12:07   0:07 /opt/Element/element-desktop --type=renderer --field-trial-handle=6416784470168092594,10345997723189300962,131072 --enable-features=WebRTCPipeWireCapturer --disable-features=CookiesWithoutSameSiteMustBeSecure,HardwareMediaKeyHandling,MediaSessionService,SameSiteByDefaultCookies,SpareRendererForSitePerProcess --lang=de --standard-schemes=vector --enable-sandbox --secure-schemes=vector --bypasscsp-schemes --cors-schemes --fetch-schemes=vector --service-worker-schemes --streaming-schemes --app-path=/opt/Element/resources/app.asar --num-raster-threads=2 --enable-main-frame-before-activation --renderer-client-id=4 --no-v8-untrusted-code-mitigations --shared-files=v8_context_snapshot_data:100

tycho     4553  8.1  1.0 358084 83624 pts/1    Sl+  12:07   0:01 /opt/Element/element-desktop --type=gpu-process --field-trial-handle=6416784470168092594,10345997723189300962,131072 --enable-features=WebRTCPipeWireCapturer --disable-features=CookiesWithoutSameSiteMustBeSecure,HardwareMediaKeyHandling,MediaSessionService,SameSiteByDefaultCookies,SpareRendererForSitePerProcess --gpu-preferences=UAAAAAAAAAAgAAAQAAAAAAAAAAAAAAAAAABgAAAAAAAwAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAgAAAAAAAAACAAAAAAAAAAAAAAAAAAAAAIAAAAAAAAAAgAAAAAAAAACAAAAAAAAAA= --shared-files

[abb128]

I was also experiencing the issue of constant high CPU use while the window is not minimized. In my case, it turned out to be caused by having an animated gif as my profile picture. Most UI elements do not show the animation, but the sidebar would play the animation and presumably do some CPU intensive blurring every frame.

[barathrm]

My element causes between 50-80% cpu usage according to top, even though there are no visible changes in the GUI. I don't have an animated profile picture.

I'd love to help debugging this via performance profiles, but it seems those contain screenshots of the app. I don't know what else they might contain, so I'll not start with posting one now. I'd also recommend warning users about that when devs request them as users might not notice/know that they contain screenshots, thereby accidentally leaking private information.

Distro: opensuse tumbleweed
Element version: 1.9.8
Olm version: 3.2.8

[m-o-e]

Can confirm high idle CPU usage on OSX (Monterey, 12.3.1).
Element version 1.10.9 (1.10.9)

For me it appears to correlate to the number
of messages in the backlog.

I can reproduce it like this:

1. Create a new, empty room

2. While the room is empty: <1% CPU

3. Type one message. Idle CPU raises to ~15%.

4. Every additional message adds around 2% to the idle CPU load
   until it plateaus at around 50%.

It sticks to this high idle load while Element window is visible,
even when it's in the background. When hiding the window (with ⌘H)
it drops to <1% and stays there.

I have no animated avatar. None of the toggles under Preferences
seems to affect it (CPU load remains unchanged when toggling them all to off).