LAV CUVID Decoder - High Quality Hardware decoding for NVIDIA

[CruNcher]

Mixer73 all this frame dropping stuff depends on many more then just 1 part of the system also Windows is no realtime OS so you have to be careful what is going on the system (everywhere) any load somewhere can cause frame drops (using a exclusive mode is more secure but also not a 100% avoiding solution) it's not easy to solve by anyone if you don't know exactly your system and it's configuration and i mean really in depth analyzing it directly on the machine.

If you have a cluttered system and even EVR CP in Exclusive mode drop frames you have to kill everything you know that could introduce this the best is really to test on a absolute clean system installation with no more 3rd party code on it, and if you put more 3rd party code on it you should always check if that interferes with your Playback

[Mixer73]

Originally Posted by CruNcher

Mixer73 all this frame dropping stuff depends on many more then just 1 part of the system also Windows is no realtime OS so you have to be careful what is going on the system (everywhere) any load somewhere can cause frame drops (using a exclusive mode is more secure but also not a 100% avoiding solution) it's not easy to solve by anyone if you don't know exactly your system and it's configuration and i mean really in depth analyzing it directly on the machine.

Thanks, I'm gonna do the clean format anyway, this machine is a little long in the tooth, hell maybe it will just do the trick.

I am adding an SSD to the system as well with the format so it can't get worse :P

[CruNcher]

Originally Posted by Mixer73

Thanks, I'm gonna do the clean format anyway, this machine is a little long in the tooth, hell maybe it will just do the trick.

I am adding an SSD to the system as well with the format so it can't get worse :P

Yeah sometimes the best solution though if you think you can handle it you could first look in your Performance information since NT 6 Windows has some nice self Diagnostic Layer that can be really helpful to find not sooo deep Performance problems @ least and some of them could also be the cause for overall system issues

Boot up/down time analysis code 100,102,103 ect are the first thing to look @

[jmonier]

I've noticed two anomalies with LAV CUVID which, while they don't seem to be a problem for me, may be of interest. I haven't tested really extensively, but in my setup (which includes Zoomplayer and madVR), they happen reliably with LAV CUVID and not with other decoders.

1) GPU memory usage (as displayed by GPU-Z). Whenever I stop playback and then start it again, the memory usage is higher than it was before. This continues until the upper limit of memory is reached, then it stays at that value. Playback is fine, even after the upper memory limit is reached. The memory only goes back to zero when I exit Zoomplayer.

2) madVR Windowed Mode with interlaced material. This causes constant dropped and delayed frames. It's fine in Full Screen Exclusive mode and also fine with non-interlaced material in both modes.

[CruNcher]

@nev

http://forum.doom9.org/showthread.ph...10#post1525110 <- here you can see the current efficiency Intel reached themselves with the copy overhead on SB, most probably the driver team)

will be super interesting to compare that vs Lav Cuvid

[TheShadowRunner]

Nevcairiel, thank you very much, between your filters and Madshi's renderer, you've pretty much given a new life to HTPCs.
I'm wondering, do you ever plan to integrate LAV CUVID into Lav Video Dec eventually?
So that 8bit h264 could be CUDA accelerated and 10bit could be software decoded, with a single codec? (since these video streams share the same subtype, setting 2 decoders is pretty much impossible (?))

Also, i take it you've no hints regarding nvidia's ivtc feature available in their driver?

[CruNcher]

Nope nev was one of the first who thought about this Lav Cuvid supports to fallback (it wont allow connections for 10 bit 4:2:2 neither 4:4:4 and also not Mpeg-2 4:2:0 content) you just have to configure your player right

You put Lav Cuvid on top of the chain and enable H.264 and Lav Video bellow it (or any other you prefer for this type of content, makes it much more flexible in your decision what you want to use), now if you open 10 bit content it will use Lav Video same for Mpeg-2 Studio if you select also Mpeg-2.

Though the Mpeg-2 fallback is a special case it works only combined with Lav Splitter.

[TheShadowRunner]

Brilliant, thanks for the tip, CruNcher. Now I need time for some testing
Strange this isn't documented anywhere though ^^;

[clsid]

It is not documented because it is 'normal' behavior. For example the MPC and ffdshow DXVA decoders also deny connection for stuff they do not support to allow fallback to the next available decoder.

[mindbomb]

how is the mpeg 4 asp decoding in cuvid?

i don't have a compatible graphics card to test it, but dxva for it on the radeon 6000 series using the arcsoft video decoder results in a generally blurrier picture and there are frequently bugs. I'm wondering if similar problems are found with cuvid.

[betaking]

Originally Posted by mindbomb

how is the mpeg 4 asp decoding in cuvid?

i don't have a compatible graphics card to test it, but dxva for it on the radeon 6000 series using the arcsoft video decoder results in a generally blurrier picture and there are frequently bugs. I'm wondering if similar problems are found with cuvid.

Last cyberlink video decoder dixv decoder or daum potlayer built deocoder can support radeon 6000 series mpeg 4 asp dxva!
cuvid only for nvidia cuda graphics card!

[CruNcher]

Originally Posted by mindbomb

how is the mpeg 4 asp decoding in cuvid?

i don't have a compatible graphics card to test it, but dxva for it on the radeon 6000 series using the arcsoft video decoder results in a generally blurrier picture and there are frequently bugs. I'm wondering if similar problems are found with cuvid.

The difference is that they can chose the PP method they want to apply themselves Mpeg-4 Asp has no standard defined PP or any kind of inloop it's up to the implementer if they want to use it or not and how (bitstream information based, default strenght,manual) everything's possible and most probably will be different from implementer to implementer (resulting in different visual output)
Arcsoft seems by default then to use a heavy PP to make sure blocks are decimated as much as possible taking into account that it kills fine details (though they could also use a more adaptive approach that works only in their own framework,needing bitstream information).
This and that the IDCT,Qpel could differ (between encoder/decoder) where the 2 major reasons for it to fail and to replace it with H.264
Mpeg-4 ASP had no Commercial Stability that was needed for HD-DVD,Blu-Ray and more Advanced Interoperable Internet (Cloud) solutions we see today in combination with a Hardware Ecosystem that works very good (unthinkable with Mpeg-4 ASP, that's why DivX Networks(Inc) initiated their Certification they saw that others would lose in this catastrophe without any guidance and that a own Ecosystem with their own Standards was highly needed for those sheep's @ that time)

[Andy o]

Originally Posted by nevcairiel

Yes it is a pure software decoder. The increased GPU load is probably because the GPU actually goes into a lower power state. CUDA forces it into the high-power mode for some reason..

Is there only a low and a high power mode, or is this a mid power state like AMD does with many mid-high-end cards like the 5770 (and the 4670)? When UVD is being used, I get 400/900 (locked so there's no DPC latencies). Idle/low state is 157/300 and high is 850/1200. I'm considering swapping to my GTX460 once again to check out your latest developments especially re: deinterlacing.

[nevcairiel]

There is typically 3 states with NVIDIA cards that are used. P0 is the full 3D power mode, P8 is the "Video" mode at around half speed, and P12 is the 2D low-power mode (which seems to actually be even lower then your low state, just going by clocks).
DXVA uses the P8 mode, unless the 3D load from post-processing is too high and forces P0. But CUDA always forces the P0 mode, unless you use a tool like NVIDIA Inspector to change the modes per-application or by some other rules.

Recently, NVIDIA changed the way the P-state switching is done. However, in the upcoming 285 driver release, they went back to a more "conservative" P-state logic, because it apparently caused quite some issues.
Using my DXVA-HD deinterlacer, it usually starts out in P0 mode, but if i let it run for a while, after 30s or so it goes into P8 (which can be seen in my debug timings, as the values double. )

[pankov]

Andy,
yes there are three power states in the recent NVIDIA cards (including the GTX460).
P0 - full 3D power
P8 - video
P12 - idle
There is a nice application called NVIDIA Inspector which has a feature called "Multi Display Power Saver" which allows you to select which apps trigger which state and also at what level (Threshold) of GPU/VPU usage should each level be activated.
I use GTX 460 SE and I'm very glad I changed my 5750 for it because now I can watch my 1080i recordings with perfect deinterlacing and perfect display synchronization thanks to madVR which works perfectly with LAV CUVID Decoder.

Edit:
nevcairiel was quicker

[CruNcher]

Also nev could put this multi display power saver part right into Lav Cuvid utilizing NVAPI additionally and implementing a switch (always keep P8 state on/off)
For most this would be useless though as you need they dynamic of their system else you always gonna endup with to slow results and Nvidia has to find this best dynamic on the driver level which is hard as they have to take a lot of different scenarios into account it never will be perfect in (performance/power/latency) for everything that's for sure (Web,Games,Video,Folding)
The last time Nvidia Engineers tried to lower the latency most cards couldn't hold that stress and errored out, especially Gamers where not happy with that

[thuan]

On my GTX560 card, normally my card has 3 power states like nevcairiel mentioned, but after using CUVID 0.12 my card won't go to P12 and stucks at P8 when idle, I need to restart to have the correct behavior back. @Dev: do you have the same problem?

EDIT: DXVA is fine BTW.

[nevcairiel]

Originally Posted by thuan

On my GTX560 card, normally my card has 3 power states like nevcairiel mentioned, but after using CUVID 0.12 my card won't go to P12 and stucks at P8 when idle, I need to restart to have the correct behavior back. @Dev: do you have the same problem?

EDIT: DXVA is fine BTW.

I don't have that problem, it always goes back into P12 for me eventually.

[CruNcher]

Also Windows Vista/7 with Aero is more complex it should standby any application in the background though but who knows what a faulty application could do with your GPU
Best is to check with Process Explorer and the OS GPU state and see if something in that context could trigger a P state switch also or keep P state constantly high for example i guess some changing wallpapers could trigger it.
A javascript driven Picture Change in the Browser can already if additional you scroll around for example (if the spike occurs @ the same time)
Nvidia also is going to introduce a FPS capper in the driver which could be used to avoid such things if Applications dont come up with such sollutions themselves

[Andy o]

Thanks guys, Nvidia Inspector seems like a pretty useful tool. I'll probably be trying it soon.

Originally Posted by nevcairiel

Using my DXVA-HD deinterlacer, it usually starts out in P0 mode, but if i let it run for a while, after 30s or so it goes into P8 (which can be seen in my debug timings, as the values double. )

Does this mean I'd have to use LAV Video instead of CUVID?