How come an older computer can't play 2160p video at 1 fps?

[LG7]

I am new here. If this is not the right place to post this, apologies.

We have a new computer that can play 2160p@60fps video smoothly (nVidia GTX980M GPU).

We also have a ~7-year-old computer (GPU: ATI HD 6970M) that plays 1440p@60fps smoothly, it also plays 1080p@120fps on an 1080p 120Hz screen, but not 2160p@1fps.

With deactivated hardware acceleration, the old computer plays 2160p@30fps footage normally for half a second and then the video freezes while the audio keeps playing.

I am sure the old computer has more than 1/60 of the performance of the newer computer. How come it fails to play back 2160p at ONE frame per second (from a test video file)?

Both 2160p@30fps and 1080p@120fps have exactly 248832000 pixels per second. How come only the latter can be played by the older computer, despite the same number of pixels per second?

[johnmeyer]

There are many variables involved in playing video smoothly on any computer, new or old. I have a 12-year old computer that plays pretty much everything, with no problem. In my case, they key was finding the right software program.

Software video players are NOT all created equal.

So, what video player are you using?

[jagabo]

Are you saying you have a 1 fps 2160p video that isn't playing on the old computer? Or are you saying you expect a 30 fps 2160p video to play at some lower frame rate but still show you all the frames?

[Selur]

As a side note: What format is you video in?
Often 2160p is HEVC (might even be HDR) while 1080p is H.264 and your graphics card will not be able to decode HEVC @ 2160p.

[LG7]

Sorry for late response. I have just seen it.

There are many variables involved in playing video smoothly on any computer, new or old. I have a 12-year old computer that plays pretty much everything, with no problem. In my case, they key was finding the right software program.

Software video players are NOT all created equal.

So, what video player are you using?

I tested it using VLC Media Player, Media Player Classic – Home Cinema, Media Player Classic – Black Edition and xplayer (on Windows and Linux respectively).

[LG7]

Sorry for late response.

As a side note: What format is you video in?
Often 2160p is HEVC (might even be HDR) while 1080p is H.264 and your graphics card will not be able to decode HEVC @ 2160p.

No, it was not HEVC, but plain H.264 (self-recorded video footage and video files from GSMArena and DKamera.de shot on various devices).

And the 1080p@50fps was AVCHD-progressive.

[LG7]

Are you saying you have a 1 fps 2160p video that isn't playing on the old computer? Or are you saying you expect a 30 fps 2160p video to play at some lower frame rate but still show you all the frames?

I am wondering how 1080p@120fps plays smoothly but 2160p@30fps not, despite both have the exact same pixel rate of [strike]299792458[/strike] edit: 248832000 pixels per second (3840×2160×30 and 1920×1080×120).

[Selur]

MPC and MKC HC should both allow to see wether the VPU is used or not.
My guess is that you use a combination the VPU on your graphic card does not support.

[jagabo]

I am wondering how 1080p@120fps plays smoothly but 2160p@30fps not, despite both have the exact same pixel rate of [strike]299792458[/strike] edit: 248832000 pixels per second (3840×2160×30 and 1920×1080×120).

With CPU decoding this could be an issue of cache size. Once the working set size (1 frame of memory*) exceeds the L3 cache size memory access becomes much more expensive. A seven year old computer probably has a cache size of 6 MB -- that's large enough to hold one frame of 1920x1080 YUY2 video (~4 MB). But not large enough to hold one frame of 3840x2160 YUY2 video (~16 MB). So even though the total throughput of the two is the same the latter exceeds the size of the L3 cache, resulting in cache thrashing.

* With modern codecs the working set while decompressing video is typically several frames. But the issues still holds -- the 4x larger working size is going to result in less effective caching.

[LG7]

I am wondering how 1080p@120fps plays smoothly but 2160p@30fps not, despite both have the exact same pixel rate of [strike]299792458[/strike] edit: 248832000 pixels per second (3840×2160×30 and 1920×1080×120).

With CPU decoding this could be an issue of cache size. Once the working set size (1 frame of memory*) exceeds the L3 cache size memory access becomes much more expensive. A seven year old computer probably has a cache size of 6 MB -- that's large enough to hold one frame of 1920x1080 YUY2 video (~4 MB). But not large enough to hold one frame of 3840x2160 YUY2 video (~16 MB). So even though the total throughput of the two is the same the latter exceeds the size of the L3 cache, resulting in cache thrashing.

* With modern codecs the working set while decompressing video is typically several frames. But the issues still holds -- the 4x larger working size is going to result in less effective caching.

Very interesting indeed, thank you.

About the GPU: The GPU that the 7 year old computer has has a maximum display resolution of 1440p, while the new GPU certainly supports 2160p.

Also, the newer computer supports toggling between an CPU-integrated Intel GPU and a dedicated Nvidia GPU, while the old computer does not offer the option to use the integrated GPU, although the hardware (Intel Core i7-2630QM) is equipped with one.

[LG7]

The other thread thread is not the exact same subject as this one.

They overlap somewhat, but one thread asks about why two devices with the exact same GPU play videos differently, while the other is about why an old computer fails to play back a video at a higher resolution with proportionally decreased framerate.

It is, although somewhat similar, not exactly the same thing.

[Selur]

same gpu: same source, same drivers, player and settings?
MPC-HC: allows to see which GPU is used (assuming one is used)

[babygdav]

https://trac.videolan.org/vlc/ticket/214
Vlc has a know limitation - can't play 1fps.

Most media players attempt to play back files in real time, dropping video frames and audio as needed to stay on time.

Try turning off any "drop frames" settings that are checked.

Don't have it, but you may need a program like VirtualDub to play back files frame-by-frame without dropping at non-real time playback.

[gdgsdg123]

I am wondering how 1080p@120fps plays smoothly but 2160p@30fps not, despite both have the exact same pixel rate of 248832000 pixels per second (3840×2160×30 and 1920×1080×120).

For the decoding complexity change of such videos is not linear in the first place...

I don't have the interest yet to write an explanation on why and how exactly, but this post should give you some ideas on the concept of modern codecs.

With CPU decoding this could be an issue of cache size. Once the working set size (1 frame of memory*) exceeds the L3 cache size memory access becomes much more expensive. A seven year old computer probably has a cache size of 6 MB -- that's large enough to hold one frame of 1920x1080 YUY2 video (~4 MB). But not large enough to hold one frame of 3840x2160 YUY2 video (~16 MB). So even though the total throughput of the two is the same the latter exceeds the size of the L3 cache, resulting in cache thrashing.

* With modern codecs the working set while decompressing video is typically several frames. But the issues still holds -- the 4x larger working size is going to result in less effective caching.

True though... it's not the major limit of the decoding performance. (unless you're talking about uncompressed content)

...isn't the GPU mainly responsible for video playback (and recording)?

It isn't unless you're playing (decoding) / recording (encoding) videos of certain constraints (supported by the GPU's VE (Video Engine)) and chose to use the GPU for the job. (i.e. hardware acceleration)

https://trac.videolan.org/vlc/ticket/214
Vlc has a know limitation - can't play 1fps.

More info on the subject:
https://forum.videohelp.com/threads/393903 ("Further Off-Topic", "VLC")

[jagabo]

https://trac.videolan.org/vlc/ticket/214
Vlc has a know limitation - can't play 1fps.

Good find. I verified it has problems with low frame rates (less than 6 to 8 fps, or so). The exact failure mode varies. Sometimes it will show a blank window. Sometimes just the first frame. Sometimes it would show several frames quick then pause for a while, repeating.

[LG7]

Ok, thanks for clearing it up. I now understand DPB is the reason.
—————

But is still something I don't understand:

How come the UleFone Armor 2 fails to play back 4K video smoothly, despite having appropriate hardware: Mediatek Helio P25 CPU and a Mali T880 GPU, both of which support 2160p video according to data sheet?

(Original thread)

[gdgsdg123]

I now understand DPB is the reason.

"How come an older computer can't play 2160p video at 1 fps?" ...I fear it's not.

The DPB in this case is basically defined by the media player (or its underlying hardware acceleration method).

How come the UleFone Armor 2 fails to play back 4K video smoothly, despite...

What if the software failed to properly utilize the available hardware capabilities? (which is so common)

Mediatek Helio P25 CPU and a Mali T880 GPU, both of which support 2160p video according to data sheet?

Highly dubious on this line.

Also note: the VE is usually considered as an independent part on SoCs.

[LG7]

[QUOTE=gdgsdg123;2571696]

What if the software failed to properly utilize the available hardware capabilities? (which is so common)

Mediatek Helio P25 CPU and a Mali T880 GPU, both of which support 2160p video according to data sheet?

Highly dubious on this line.

Also note: the [video engine] is usually considered as an independent part on SoCs.

Interesting.

But the software used to play back the video (both VLC Mobile and Android Media Framework) play it back on a Galaxy S7 Edge (Snapdragon 820 CPU and Mali T880 GPU) smoothly.

So the only way this could happen is that the software is somehow incompatible with the Helio P25, although it shouldn't be, because both P25 and Snapdragon 820 the same processor architecture, ARMv8.

Highly dubious on this line.

Why?

The Helio P25 data sheet mentions 1080p display resolution, but 2160p video resolution.

Also, it plays 1080p@60fps and 1440p@30fps smoothly.

[gdgsdg123]

Why?

After 30 seconds' research... I believe MediaTek Helio P25 is a SoC instead of a CPU.

And it's extremely rare for a smartphone SoC to have multiple VE units integrated.


Note: the same for Qualcomm Snapdragon 820.

So the only way this could happen is that the software is somehow incompatible with the Helio P25...

Think of how the video could be decoded with the available hardware. (pure CPU; GPU assisted; VE (ASIC) handover)

[LG7]

@gdgsdg123 Thanks for the information.

But I just noticed: That P25 data sheet ( https://www.mediatek.com/products/smartphones/mediatek-helio-p25) says “capture FPS: 24fps”.

I will now try playing a 2160p@24fps video on the UleFone Armor 2.

Note that video encoding needs much more ressources than decoding.

But in the meantime, here are some more observations and tests I made:

Using the app “Open Camera” (third party camera software for Android), I have access to frame rate modes (15, 24, 25, 30, 60, 96, 100, 120) that were not accessible through the precluded camera software that always uses VFR (variable frame rate) 10-30fps (depending on brightness conditions) at these resolutions: QCIF (176×144), CIF (352×288), D1 (720×480), HD 1280×720, FullHD 1920×1080.

According to my recording tests using the OpenCamera on the rear camera:

2160p@15fps: Fails to start recording.
1440p: OpenCamera has no 1440p mode, at least not the somewhat old version I use.
1080p@120fps mode: Skips many frames. Apparently, the effective framerate is ~30fps
1080p@60fps: Similar as 120fps, but less frequent stuttering. Sometimes records a few frames without skipping in succession. But average frame rate still appears like 30fps.
720p@120fps: Looks 90% smooth, ~10% skipped frames
720p@60fps: Looks 95% smooth, ~5% skipped frames (surprisingly).
864×480@120fps: Always smooth, no skipped frames.

Note that the Armor 2 uses a crappy image sensor (Samsung S5K3P3), which could also be the cause of the poor performance at 720p or higher.

I have not done much testing for the front camera (also max. 1080p@30fps), but I have made another interesting observation:

Selecting 480p@60fps for the front camera actually works in the viewfinder (the preview looks 60fps smooth), but when starting the video recording, the camera software (OpenCamera) freezes up and the front camera becomes inaccessible to any app until next reboot (even after forcibly taskkilling OpenCamera). Strange phenomenon.

[gdgsdg123]

But I just noticed: That P25 data sheet ( https://www.mediatek.com/products/smartphones/mediatek-helio-p25) says “capture FPS: 24fps”.

I will now try playing a 2160p@24fps video on the UleFone Armor 2.

I doubt it's related...



"Looks smooth"?..

Caveat: what you recorded may not be of the expected frame rate, and you may not be able to tell merely by look.

Also note: the existence of the "consecutive duplicates".

[LG7]

But I just noticed: That P25 data sheet ( https://www.mediatek.com/products/smartphones/mediatek-helio-p25) says “capture FPS: 24fps”.

I will now try playing a 2160p@24fps video on the UleFone Armor 2.

I doubt it's related...



"Looks smooth"?..

Caveat: what you recorded may not be of the expected frame rate, and you may not be able to tell merely by look.

Also note: the existence of the "consecutive duplicates".

The test files used for playback were not the same as the ones recorded on the Armor 2.

I tested a DCi 4K video (4096×2160@24fps) now.

Playback using VLC: Smooth for first second, then only the audio keeps playing.

Playback using stock media player: smooth.

[LG7]

But I just noticed: That P25 data sheet ( https://www.mediatek.com/products/smartphones/mediatek-helio-p25) says “capture FPS: 24fps”.

I will now try playing a 2160p@24fps video on the UleFone Armor 2.

I doubt it's related...



"Looks smooth"?..

Caveat: what you recorded may not be of the expected frame rate, and you may not be able to tell merely by look.

Also note: the existence of the "consecutive duplicates".

I played the recorded videos back at a quarter speed to see whether it skips frames.
The other test videos (e.g. 4K videos) at original speed.