[m121]

Sorry for this newbee comparison. I know kodi has a lot more functionalities than vlc, and this is why I prefer and I use kodi.
But once you play a video, it seems to me that vlc and kodi are doing the same job (playing a video with optional additional filters), this is why I am comparing the 2 software.

If, as you said, kodi is doing higher post processing, is there a way to disable it? I am very satisfied with the quality and gpu usage of vlc when playing a video. So I just wanted to know if there is a way to have lower gpu usage on kodi.

I have tested libreelec, but as far as I remember gpu usage was high too.

[Dangelus]

I'm not I understand why you are so concerned about the GPU usage being higher when using Kodi as opposed to VLC. It really is comparing apples to oranges.

Is the video playback performance suffering from the high GPU usage?

Are you trying to multi-task on your device and the high GPU usage is causing problems?

If everything is working fine then what is the problem? I'm not trying to be funny here just that you haven't really claimed to have a performance issue so I am confused
edit: Sorry - just re-read the thread and the higher load is causing your fan to be loud which I agree is distracting.

[m121]

Hi,
Firstly, as I said on my previous post "But on a specific platform (acepc gk3v, intel j4125), the extra gpu load of kodi will quickly starts the fan, which sounds very loud and is disturbing while watching a movie. Using VLC on the same platform let me watch a movie without hearing the fan."

Secondly, maybe have you heard about global warming? It's a shame to use more power for an identical result (or an identical result for my eyes). I can understand that some people have better eyes (or different screens...) but in my case I am just searching a way to "downgrade" kodi quality or post processing to lower the power usage.

As I said a lot of times, yes I know kodi and vlc are different (apple and orange). But when playing a movie, they are very comparable (2 apples or 2 oranges as you prefer).

[Dangelus]

As you can see I did go back and re-read your statement so I understand what the detrimental issue is for you. 

As for the global warming comment, you are not helping by using an Intel device, you would solve both your concerns by switching to an ARM device such as Android TV

But to return to some semblance of seriousness, I am not sure how much "tweaking' can be done to downgrade the playback experience on Kodi at a user level (the settings are not terribly extensive even at expert level) but I am interested to see if any of the media gurus have any ideas for you.

You may have better luck using VLC as an external player with Kodi.

[sarbes]

And as I said in my last post, Kodi on Linux does NOT utilize the GPU for video rendering if the system allows for it. This is far better than VLC in most cases.

The video rendering workflow with a compositor:
Video decoder -> RAM -> GPU -> RAM -> Scanout engine
And without (using GBM/DRM Prime):
Video decoder -> RAM -> Scanout engine

This makes Kodi far more (power) efficient in most cases, as it can exploit multiple benefits by going the shorter path. Otherwise, systems like the Raspberry Pi series wouldn't be able to offer up to 4k video support. Heck, I'm convinced this introduces jank as I suspect that the GPU is powering down completely on such systems when no GUI update is rendered for a while.

Running Kodi in a DE (and thus rendering via the GPU) is not a primary use case, especially on low-power systems. The argument could be made to to optimize the GPU path by making some features optional, but the main focus is on the efficient path. Use LE 11, as setting up a proper software stack for GBM is not trivial.

So no, you can not compare a media player for a desktop environment (VLC) with one intended running standalone (Kodi).

[m121]

Thanks for re-reading, editing your post and understanding my problems!

I remember using ARM devices a long time ago, but I was quite disappointed. Support was very poor, and devices where stucked to a kernel and a GPU driver version. Upgrading system was very difficult, and the only solution I found was changing the hardware every 2 years... I could try again to see if it is better nowadays.
Google devices are not a solution for me, and non google "android" systems (like lineageos) lacks support for applications like netflix.
For now, I was trying to use a more supported and with upgradable software platform, like x86 processors. By the way, I use low end CPU, which are also low power (5W TDP).

I agree with you that even in expert mode, I cannot play a lot with settings to reduce GPU usage. Maybe some kodi developer will answer here?

As my first post in this thread is quite old, I do not remember all the things I tried. But I remember having tried VLC as external player and integration was not so good. But maybe I have not done it correctly.

[m121]

Thanks sarbes for all this technical information, which unfortunately I do not understand for now . I will try do document myself...

Maybe I misunderstood, but I get confused:

• You say "This is far better than VLC in most cases" and "This makes Kodi far more (power) efficient in most cases". In my tests, kodi is not more power efficient than VLC. What am I doing wrong in my tests?
  
• you say "Kodi on Linux does NOT utilize the GPU for video rendering" and "Running Kodi in a DE (and thus rendering via the GPU)" : so does is it render or not via the GPU?
  
• you say "you can not compare a media player for a desktop environment (VLC) with one intended running standalone (Kodi)": I am comparing both desktop environment versions of VLC and kodi. The last test I need to make is comparing (on same hardware, and with same video) VLC on desktop environment with standalone kodi (librelec)
  

Please correct me where I am wrong

[sarbes]

You say "This is far better than VLC in most cases" and "This makes Kodi far more (power) efficient in most cases". In my tests, kodi is not more power efficient than VLC. What am I doing wrong in my tests?

Hard to tell from my site. But most likely, you are running an outdated software stack besides a compositor.

you say "Kodi on Linux does NOT utilize the GPU for video rendering" and "Running Kodi in a DE (and thus rendering via the GPU)" : so does is it render or not via the GPU?

Kodi can render videos on the GPU, but doesn't need to. If the SW/HW supports it, the decoded frames can be read directly by the scanout engine. This is often not possible with VLC.

you say "you can not compare a media player for a desktop environment (VLC) with one intended running standalone (Kodi)": I am comparing both desktop environment versions of VLC and kodi. The last test I need to make is comparing (on same hardware, and with same video) VLC on desktop environment with standalone kodi (librelec)

Using a desktop environment for a fullscreen media application is a artificial limitation by your side. I don't deny that there is room for improvement when it comes to GPU rendering (like exposing more user settings). But "complaining" that a playback path with a (possibly) more advanced feature set is not performing as well as a basic one, seems a bit contrived to me. Especially since a far more optimized path is available.

I don't complain that HDR output is not available for VLC on X11 (and Linux in general, I think), while Kodi using GBM is capable of it (with some external limitations). Kodi can leverage its unique position as an all-in-one mediacenter to achieve this. VLC can't, making it a different class of software and only partially comparable.

Just go for a recent LE 11 nightly and make sure you enable DRM Prime rendering.

[m121]

Ok, thanks for all this technical information!
I understand that kodi (desktop environment version) and vlc are technically different. But once they are playing a movie they are functionally equal. And one (vlc) is more power efficient than the other one in my tests.

So as you recommended I tried LibreELEC-Generic.x86_64-11.0-nightly-20221222-afbe3d7 on a n5000 intel processor.
The results are good:
render 33.95% and video 21.79% (gpu @103Mhz)

on linux mint 19.3 with VLC (on the same video), I have:
render 57.76% and video 13.52% (gpu @153Mhz)

LE is better on render and VLC is better on video. But gpu frequency is lower on LE.
LE seems more power efficient than mint+vlc which seems also normal

[Klojum]

LE is expected to have the least amount of overhead, because of its design. Mint is a full-blown OS, where LE is plain-vanilla Linux with only the necessary bells & whistles. I would only start looking at CPU and GPU frequency stuff when videos start to stutter and such.

I have a J5040 "office machine" with a UHD605 gpu, it basically runs all under Linux, but Windows 11 does run smoother on that box. Ubuntu is my daily driver however.

[sarbes]

Ok, thanks for all this technical information!
I understand that kodi (desktop environment version) and vlc are technically different. But once they are playing a movie they are functionally equal. And one (vlc) is more power efficient than the other one in my tests.

No, they are not equal. Kodi does frame interpolation, VLC doesn't (as far as I know). Equal would be Kodi running on GBM, which can be done with Mint as well (if Kernel and Mesa are not too old). And apart from developing, I don't see a reason not to do so.

So as you recommended I tried LibreELEC-Generic.x86_64-11.0-nightly-20221222-afbe3d7 on a n5000 intel processor.
The results are good:
render 33.95% and video 21.79% (gpu @103Mhz)

Was the GUI active? Without it, the renderer should sit idle (as there is nothing to render).

LE is better on render and VLC is better on video. But gpu frequency is lower on LE.
LE seems more power efficient than mint+vlc which seems also normal

No, VLC and Kodi are using roughly the same video resources as you would expect. You have to normalize the percentage values to the clock of the IP core (assuming that the video decoder is living in the same clock domain as the stated one). If you want to have a more objective comparison, you would need to peg all system clocks to fixed values, as performance usually doesn't scale linearly with clock rate.

Also, you save ~60% memory resources (depending on multiple factors), which should further reduce power consumption.