NVIDIA vs. Radeon VDPAU Mesa 17.2 Video Decode Performance

[jakubo]

has anyone experience with this?

https://bugzilla.redhat.com/show_bug.cgi?id=1305699

i get rather bad performance with a lot of flash movies / live streams.
Not sure if set up properly though... a lot of confusion for me... documentation mostly outdated for fedora..

[dheitmueller]

There is nothing misleading about them, qvdpautest is publicly available and has been for years, all the questions can be answered by looking at it, as well as the Phoronix Test Suite's test profile wrapped around it.

Certainly to be fair to Michael, he's relying on an existing app (qvdpautest) and I have no reason to believe the benchmark numbers he is quoting are not consistent with whatever the tool reports.

That said, IIRC the mixer tests (i.e. for the deinterlacing modes) are run without anything else happening (i.e. no MPEG2/H.264/etc video decoding is going on). Hence while the raw numbers may be accurate, they are not representative of what you would see in a real-world use case where you are trying to watch a video -- in which case you would be using the decoder and mixer at the same time.

For some designs the above might not matter - if the mixer and decoder are not sharing the same resources then they could be tested in isolation. I don't know enough about AMD's hardware design to offer an opinion there. However if the deinterlacing is implemented, for example as an OpenGL shader, then you may end up with very different numbers than exercising each of the units in isolation.

And with regards to qvdpautest and SPU blending: IIRC it doesn't exercise that functionality at all.

In summary, I'm not arguing that Michael's numbers are factually incorrect, and in fact they are interesting from an academic standpoint and/or to compare across cards. However I wouldn't rely on them to get a realistic assessment of whether a card can handle video playback for a given codec/resolution/etc without stuttering or frame dropping.

It certainly feels like qvdpautest could use some improvement though to test some more realistic use cases (something I had considered adding in the past but there are so many hours in the day). :-)

Regards,

Devin

[Hi-Angel]

I wonder if everybody see something like this

[leipero]

I just don't use VDPAU at all, one year back it introduced problems with VLC on my hardware, even tho i do not use VLC anymore (MPV FTW! ), it also had a problem with cracking sound on youtube etc., removing that package solved it, I didn't test it recently, I might as well install and try it again.

[pal666]

"MPEG-4" itself is rather inspecific. Are you talking MPEG-4 Part 2? Simple profile? Advanced simple profile?

he is talking about dvd codec. h264 is part 10 for blueray

[Holograph]

he is talking about dvd codec. h264 is part 10 for blueray

DVDs only support MPEG-2. And no MPEG standard was created specifically for DVD or Bluray or HD-DVD or anything similar. MPEG-4 including part 10 H.264 is significantly older than Blu-ray.

I too can assume what he means by MPEG-4 but it's still worth fixing the labeling on those tests by being specific.

It almost definitely means MPEG-4 part 2 simple profile or advanced simple profile, essentially stuff like Microsoft MPEG4v3 codec, DivX, XviD, etc.

I'm not saying it's incorrect, just that it would be good to be more specific.


Edit: oh and legume, on the last page, posted about "1080p MPEG" decoding... Unless he's using a hacked encoder and decoder, MPEG-1 doesn't support HD at all. Offhand I want to say it maxes at 352x240.. maybe it can do SD/VGA.. not 1080p.. Now, again, he's probably also just not being specific.. but he also mentions MPEG-4 decoding and H.264 decoding... just really confusing all-around. Not sure why people are so eager to benchmark these things but have no concern about labeling their benchmarks in a useful way. Again, I'm not saying it's wrong, just confusing. Edit: Yes I've realized this is directly from that test program and not legume here, but yeesh, I'd think the developers of that program would know better than to give us such amazingly confusing information.

[DragonDaddyBear]

Do any of the cards do H.265 or HDR?

[Holograph]

Do any of the cards do H.265 or HDR?

Nvidia/VDPAU: http://www.phoronix.com/scan.php?pag...U-1.0-Released

AMD: http://www.phoronix.com/scan.php?pag...VC-Main-10-UVD

[aufkrawall]

it is easier to install custom kernel package than nv driver. You can work with desktop working and going back is easy with grub.

On Arch and derivates you simply install the NV driver by ticking a box in your graphical package manager, if you want to use something like that. If you know the exact package names, way easier than on Windows.

Does amdgpu clock up with VDPAU/VAAPI when there is additional load on the GPU, e.g. because of advanced mpv features, or do you have to rely on hacks for that?
With Nvidia you have the opposite problem, GPU doesn't clock down when using cuda video decoding (but supports more formats than vdpau, nvidia vdpau does not support HEVC 10 bit nor VP9 8/10 bit).

[Holograph]

nvidia vdpau does not support HEVC 10 bit nor VP9 8/10 bit

My reply just before yours links to a Phoronix article from 2015 that claims VDPAU does in fact support H.265 (AKA HEVC). Has that changed?