LAV Filters - DirectShow Media Splitter and Decoders

[clsid]

Turning off HQ processing does not make a difference.

I have a VP4 card. DXVA checker shows this:
ModeH264_VLD_NoFGT: DXVA2, 720x480 / 1280x720 / 1920x1080

Edit:
And for deinterlacing (NV12) it shows:
DXVA1/2, 720x480 / 1280x720 / 1920x1080 / 3840x2160

[nevcairiel]

DXVA Checker only tests common resolutions, like the ones you listed there, it doesn't test for 1920x1200, or any other in-between steps.

[6233638]

You can try turning off the "High Quality Processing" checkbox for CUVID, that will then access the hardware more directly without any DXVA/Windows influence, maybe it helps. :DStill using avcodec when I do that.

[wanezhiling]

My vp5 card gt520 works fine, I will test my vp4 card gts450 tomorrow.

[nevcairiel]

http://files.1f0.de/lavf/LAVFilters-0.53.2-27-g8bf80c7.zip

Please try this version. It should hopefully be smart enough to figure out if the movies resolution is supported by your GPU, and thus avoid black-screen situations in native DXVA2.
This means, there is no hardcoded limit anymore for the DXVA2-Native resolution, assuming it doesn't break in some ugly way for someone.

If you can, please test running videos in various resolutions with this build, including 4K content. If your GPU is not capable, it should fallback to software properly, and if is capable, it should use DXVA2 Native.
Its important to test if this fallback works on all major GPUs. I tested on my Intel HD3000, and it properly falls back to Software, and on my NVIDIA GTX 680, it goes ahead and uses DXVA2. I'm specifically interested if this auto-detection works properly on AMD (and possibly also if AMD finally bothered to fix 4K decoding or if it still results in a half green frame :D)

So, run videos in supported resolutions, and ensure it uses DXVA, and try videos in unsupported resolutions (if any, 4K for most older GPUs), and see if it properly falls back to software!

PS:
I tested DXVA2-Native with 4K content and madVR, and the performance was just terrible. DXVA2 CB on the same file was like 20x faster. I've already asked madshi about performance issues with native DXVA, we'll see what happens.

PPS:
You can get 4K/2160p samples from here:
http://xhmikosr.1f0.de/samples/2160p/

Note that these samples are also high-framerate (50p usually), which is actually too much for NVIDIAs decoder (not sure about AMD) - but it can still show you if it works at all, even if it doesn't play fluidly.

[aufkrawall]

Cool, thanks nev. :)
4k files play with DXVA2n without any dropped frames for me in madVR, VPU is used and thus CPU usage very low.
Great.

Having a GTX 670.

[VipZ]

DXVA2n detection working fine for me. No 4K here sadly, instead of AMD trying to fix they simply disabled completely :(

[nevcairiel]

As long as 4k is properly falling back to software, thats at least something.

Otherwise, sounds just like AMD. :)

[VipZ]

Yep 4K is falling back correctly for me, AMD can still even crash with just NV12 output on some 4K files at present for me.

[aufkrawall]

AMD can still even crash with just NV12 output on some 4K files at present for me.
I remember blue screens some years ago with a 5870 at 4k files.
Ridiculous that it's still not fixed (or does it crash in some other way instead now? :D ).

[VipZ]

I remember blue screens some years a go with a 5870 at 4k files.
Ridiculous that it's still not fixed (or does it crash in some other way instead now? :D ).

Yep, still BSOD. I can only play that particular file when I force RGB32 output on software decoding (DVB_2012-09-09_16-38-03_Service 2_UHDTV_2160p_cut.ts)

[Prinz]

CrowdRun_2160p50.x264.CRF24 with DX native and evr crashes the graphic driver on my ATI 2600 XT.

[nevcairiel]

Guess some older GPUs don't do too well with 4k DXVA, even the attempt to open a decoder.
I had planned to add a checkbox to control 4K DXVA anyway (also because of the performance issues), guess i better add it right away.

Just to give some performance perspective, using Jinc3+AR for Chroma and Catmull-Rom+Linear Light for Image Downscaling pushes my GTX 680 over its limit playing a 3840x2160 file, downscaling to 2560x1440. Guess we can be lucky 4K content isn't popular yet, our hardware is still too slow (or we need better screens too, so downscaling isn't required!) :D

[6233638]

http://files.1f0.de/lavf/LAVFilters-0.53.2-27-g8bf80c7.zip

Please try this version. It should hopefully be smart enough to figure out if the movies resolution is supported by your GPU, and thus avoid black-screen situations in native DXVA2.
This means, there is no hardcoded limit anymore for the DXVA2-Native resolution, assuming it doesn't break in some ugly way for someone.That sample is still using avcodec here, whether CUVID/DXVA2 is selected, and even my CPU doesn't appear capable of handling the 4K samples.

I wonder if the issue is something to do with the videos being above 1080p resolution. I have other MKV files where DXVA decoding works just fine.

PS:
I tested DXVA2-Native with 4K content and madVR, and the performance was just terrible. DXVA2 CB on the same file was like 20x faster. I've already asked madshi about performance issues with native DXVA, we'll see what happens.This was my experience when testing last night. VPU load with some samples was as much as 20% higher with DXVA2 native compared to DXVA2 copy-back.

Just to give some performance perspective, using Jinc3+AR for Chroma and Catmull-Rom+Linear Light for Image Downscaling pushes my GTX 680 over its limit playing a 3840x2160 file, downscaling to 2560x1440. Guess we can be lucky 4K content isn't popular yet, our hardware is still too slow (or we need better screens too, so downscaling isn't required!) :DI wonder if it's the decoder or the GPU load failing.

Before the queues drop to 0, I'm only seeing about 60–70% GPU load. (Nvidia Inspector tells you GPU, MCU, and VPU load)

[nevcairiel]

That sample is still using avcodec here, whether CUVID/DXVA2 is selected, and even my CPU doesn't appear capable of handling the 4K samples.

It wasn't meant to change that, but if using DXVA2 Native, it should now properly fallback to software (instead of giving you a black screen), because it trys to probe the hardware for support, instead of using a fixed resolution limit.

[6233638]

It wasn't meant to change that, but if using DXVA2 Native, it should now properly fallback to software (instead of giving you a black screen), because it trys to probe the hardware for support, instead of using a fixed resolution limit.Sorry, I misunderstood that. It does fall back to avcodec correctly.

So does that mean you need a 600-series card to decode anything above 1080p with hardware acceleration?

[nevcairiel]

I wonder if it's the decoder or the GPU load failing.


I tried software decoding for this, and my CPU managed to keep up, lowering the scaling settings made it work just fine, GPU load was in the 90s with my above mentioned settings.

Sorry, I misunderstood that. It does fall back to avcodec correctly.

So does that mean you need a 600-series card to decode anything above 1080p with hardware acceleration?

Well i thought at least 1920x1200 was still supported by the older hardware, but i can't say i really had any proof.

[6233638]

I tried software decoding for this, and my CPU managed to keep up, lowering the scaling settings made it work just fine, GPU load was in the 90s with my above mentioned settings.What kind of CPU are you using? I was really surprised that my 2500K couldn't keep up, even though it's overclocked to 4.5GHz. 6-core? (hyperthreading shouldn't make a difference, as all cores were at 100%)

[romulous]

http://files.1f0.de/lavf/LAVFilters-0.53.2-27-g8bf80c7.zip

Please try this version. It should hopefully be smart enough to figure out if the movies resolution is supported by your GPU, and thus avoid black-screen situations in native DXVA2.
This means, there is no hardcoded limit anymore for the DXVA2-Native resolution, assuming it doesn't break in some ugly way for someone.



Thanks nev - the file I posted now plays with DXVA2 Native (it falls back to avcodec, like Copy-Back and CUDA), rather than showing the black screen. NVIDIA GTX460 here, 310.61 Beta drivers. It's interesting that my card doesn't seem to support 1920x1200 for hardware accleration - 1920x1080 files seem ok.

The clip was originally an uncompressed FRAPS encode, captured at my displays native resolution (1920x1200). Because raw FRAPS encodes tend to be on the large side, and because you need a far more powerful system than my Quad Core to be able to play those uncompressed clips without massive frame loss, I encoded it to H264 MKV. I can't recall what tool I used to do this now - it was probably XMedia Recode though. I would have selected one of the preset profiles - I don't know too much about the fine details of video encoding.

[DragonQ]

(hyperthreading shouldn't make a difference, as all cores were at 100%)
That statement makes no sense. All cores would be at 100% when video encoding too, doesn't stop the addition of HyperThreading greatly increasing speed.

[6233638]

That statement makes no sense. All cores would be at 100% when video encoding too, doesn't stop the addition of HyperThreading greatly increasing speed.I guess I don't properly understand how HyperThreading works then. I was under the impression that it helped more with multi-tasking operations by allowing the virtual cores to make use of resources that may be stalled and waiting for a task to finish on another core.

With fully multithreaded tasks that max out all cores, such as video encoding, the tests I had seen only showed around a 5% improvement at the same clockspeed for a 50% increase in price, and were actually detrimental to performance with tasks that didn't know how to handle HT properly.

Maybe things have changed with the latest Ivy Bridge processors though.


I was just very surprised to find that my CPU couldn't handle those videos when it's still faster than the current Ivy Bridge equivalents at their stock clockspeeds. (at least in the tests I've run, maybe not video decoding?)

[Tacio]

I have Intel DN2800MT motherboard with embedded Atom 2800 and DXVA checker shows next capabilities of atom's graphics:
http://s8.postimage.org/j90dz8o9x/image.png
Using latest dev build of LAV filters I tried to play 2160p videos from this link http://xhmikosr.1f0.de/samples/2160p/. DXVA checker says pretty the same info about them:
http://s14.postimage.org/to3o0wsup/image.png
However, when I tried to benchmark of play in DXVA checker I got this for any movie:
http://s8.postimage.org/mc5pgcsmd/image.png
When I tried to play Time Scapes 4K (actually 2560x1440 resolution) I had little bit better results :))
http://s8.postimage.org/41ozsmlt1/image.png
MPC-HC shows the same results. But with less resolution movies no problem at all though. Is it possible to fix it or it's a hardware/drivers limitation? I use the latest drivers from intel website and windows 7 sp1 x32 without aero.

[nevcairiel]

What kind of CPU are you using? I was really surprised that my 2500K couldn't keep up, even though it's overclocked to 4.5GHz. 6-core? (hyperthreading shouldn't make a difference, as all cores were at 100%)

I'm on a 2600k at 4.5Ghz as well.
And like someone else explained, HT does give you benefits in highly threaded tasks, even if all cores are at 100%.

What it basically does is improve the pipelining of the CPU, because even at 100%, internally not all stages of the CPU are always 100% busy, and by running two threads on the same core, you can improve this percentage.

In my test some time ago, HT added about 30% performance in video decoding (with an appropriate increase in thread-count, "Auto" setting in LAV does that for you)

Is it possible to fix it or it's a hardware/drivers limitation? I use the latest drivers from intel website and windows 7 sp1 x32 without aero.
Its most likely a hardware limitation, i would've actually been surprised if the Atom had 4k decoding capabilities.

I already added options to control whether it uses DXVA for SD, HD and UHD(4k) content, with 4K off by default, so even if your hardware claims support for 4K, you can turn it off again to avoid this. ;)

[6233638]

I'm on a 2600k at 4.5Ghz as well.
And like someone else explained, HT does give you benefits in highly threaded tasks, even if all cores are at 100%.

What it basically does is improve the pipelining of the CPU, because even at 100%, internally not all stages of the CPU are always 100% busy, and by running two threads on the same core, you can improve this percentage.

In my test some time ago, HT added about 30% performance in video decoding.So much for reviews then. I had a look around to check, and at least with the first few reviews I came across, they did only show roughly the 5% difference with video encoding/decoding when comparing the 2500K/2600K at the same clockspeeds that I had remembered.

I suppose it must have been a problem with their testing methodology, and it looks like I'll be buying a HT-enabled CPU next time I upgrade.

I also wasn't expecting to be needing to do any kind of encoding/decoding on the CPU when I bought this though. Sorry for going off-topic.

[nevcairiel]

Its not like 4K content is an every-day thing, besides some tech-demo clips, what else is there? :D

[Tacio]

I already added options to control whether it uses DXVA for SD, HD and UHD(4k) content, with 4K off by default, so even if your hardware claims support for 4K, you can turn it off again to avoid this. ;)
But as you understand Atom is not capable to play 2160p in software mode :)) Ok, doesn't matter.
BTW, DXVA checker also claims that this Atom capable of playing MPEG4 ASP, is it possible to add such an opting in LAV video filter? For such low power processor it could be great!

[6233638]

Its not like 4K content is an every-day thing, besides some tech-demo clips, what else is there? :DWell not yet. Hopefully with all this retina stuff Apple is pushing, we'll see 4K monitors show up some time next year, and with 4K TVs and Blu-ray players available, surely content has to be coming soon.

[wanezhiling]

Turn uhd on,play 4k in native mode, greenscreen and BSOD on 6850 card... cant fallback to sw..

[nevcairiel]

AMD seems to simply be stupid, their driver claims 4K is supported, but when you try to use it, it just doesn't work (or even BSODs). Thats what the checkbox is for (and why its off by default). :D

If people prefer, i could simply blacklist 4K DXVA on AMD instead of hoping that users don't turn on the checkbox.

[wanezhiling]

Good idea,amd destroy my system!

[RealSnoopyDog]

Its not like 4K content is an every-day thing, besides some tech-demo clips, what else is there? :D

Not yet, but (e.g.) ASTRA will start 4k test broadcasting next year:

http://www.digitalfernsehen.de/Astra-Erste-4K-Ausstrahlung-2013.95013.0.html (german language). But they even don't know, which video compression they will use. Maybe, HEVC or H.265...

But hey, i'm pretty satisfied with 1920x1080 on a 46'' flat screen. :D Higher resolutions make sense only on bigger displays.

The other thing is 3D. I think except the latest Cyberlink software, there is no codec so far that can decode 3D content properly. On Blu-Rays, additional information is stored in the SSIF subdirectory along the normal m2ts streams.

[VipZ]

Nev, I have noticed WMA/WMV are not configurable on LAV Source/Splitter, would this be possible to add now that LAV fully supports WMV playback?

[nevcairiel]

Nev, I have noticed WMA/WMV are not configurable on LAV Source/Splitter, would this be possible to add now that LAV fully supports WMV playback?

Its "asf", which is the same format. :p

[VipZ]

Its "asf", which is the same format. :p

Ah ok, wasn't sure if it was just ASF or covered WMA/WMV as well :)

Thanks, maybe adjust the description to indicate WMA/WMV as well then?

[wanezhiling]

nev, UHD only supports Nvidia VP5? I tested Intel hd4000, failed.

[nevcairiel]

LAV Filters 0.54

LAV Splitter
- Subtitle selection in "Default" mode now properly takes the "Default Track" flag into account
- Fixed an issue that could cause graph flush events to be send out of sync

LAV Video
- The DXVA2 Native decoder can now properly detect hardware support for video resolutions
- Allows 4K DXVA2 decoding on NVIDIA/Intel
- New options to control which resolutions are being handled by the hardware decoder (SD, HD and 4K/UHD)
- Added support for DVD menu overlays and DVD subtitles with DXVA2 Native
- Improved DXVA2 Native support with madVR
- Fixed a freeze when playing corrupted H.264 streams in DXVA2 Native
- Fixed a few rare DVD menu issues when using the QuickSync decoder


Download: Installer (both x86/x64) (http://files.1f0.de/lavf/LAVFilters-0.54.exe) -- Zips: 32-bit (http://files.1f0.de/lavf/LAVFilters-0.54.zip) & 64-bit (http://files.1f0.de/lavf/LAVFilters-0.54-x64.zip)

Even though this versions change list is somewhat shorter, i still decided to go with a new major version, instead of just a point release, because the changes are somewhat important.

DXVA2 Native auto-detection of hardware support
Before this version, DXVA2 Native mode had a hard-limit of 1920x1200 for the video resolution, and any video over this was refused and software decoding was used.
In this version, LAV will now try to ask the hardware about its supported resolutions, and based on this decide if DXVA2 Native should be used, or not.

This means that recent NVIDIA cards will now allow DXVA2-Native to decode 4K material (and possibly Ivy Bridge GPUs will as well).
AMD on the other hand is blacklisted in LAV, because some AMD driver versions accept 4K material, but then crash or even BSOD (and newer driver versions simply refuse it completely, so nothing lost). As before, only material up to 1920x1200 is allowed with AMD hardware.

Additionally, there is also a new option to control which resolutions are handled by the HW Accel, which allows you to use DXVA, CUVID or QS for just SD/HD, but always use Software for 4K, even if your hardware supports it. You could also simply disable SD HW decoding, if you have one of the AMD cards which have problems with SD DXVA. Note that Ultra-high-definition (UHD) / 4K is disabled by default because of performance and/or possibly hardware compat issues.

Full DVD Menu and DVD Subtitle support with DXVA2 Native
One could say this build focuses mainly on Native DXVA, and it wouldn't even be all wrong.

DXVA2-Native can now also be used for DVD decoding without any disadvantages over the other decoders. DVD Menus and DVD subtitles should now work equally well as with software decoding or any of the other HW decoders.
Note that i have not performed wide-spread performance tests on the DXVA2 Subtitle support, but i hope that even AMD can manage to deal with this.


Have a good weekend!

[nevcairiel]

nev, UHD only supports Nvidia VP5? I tested Intel hd4000, failed.

Its possible that Intel only allows it through QuickSync for some reason. I know that QS can do 4K decoding on my HD4000, i didn't test DXVA (because the HD4000 is not in my Dev PC)

What does DXVAChecker say on it?

[wanezhiling]

Its possible that Intel only allows it through QuickSync for some reason. I know that QS can do 4K decoding on my HD4000, i didn't test DXVA (because the HD4000 is not in my Dev PC)

What does DXVAChecker say on it?

My fault, now it works.

Maybe you should change the log:
- Allows 4K DXVA2 decoding on NVIDIA/Intel;)

[DragonQ]

I guess I don't properly understand how HyperThreading works then. I was under the impression that it helped more with multi-tasking operations by allowing the virtual cores to make use of resources that may be stalled and waiting for a task to finish on another core.
That's about right. However, your CPU being at 100% in Task Manager doesn't mean every single component is being used. HyperThreading uses components that a CPU core isn't using right now to form another virtual core that can be used for other things, essentially.

With fully multithreaded tasks that max out all cores, such as video encoding, the tests I had seen only showed around a 5% improvement at the same clockspeed for a 50% increase in price, and were actually detrimental to performance with tasks that didn't know how to handle HT properly.

Maybe things have changed with the latest Ivy Bridge processors though.
Dunno what benchmarks you were looking at but HyperThreading typically increases speed by 30-50% for video encoding (x264 specifically). This has always been the case, I tested it myself many years ago on my Nehalem CPU.

HyperThreading being detrimental to performance is usually overblown. It's possible it makes a slight difference (<5%) but the main complaints I see about this is in regard to certain games, which are obviously coded poorly if they can't handle HyperThreading.

For 99% of people, HyperThreading is beneficial. Whether it's worth it for the price is another matter though.

[kasper93]

FWIW, maybe someone could contact AMD folks and tell about problems with 4k dxva decoding. I guess they are aware of that, but maybe it will encourage them to fix this faster. Or if they can't do DXVA 4k decoding at least their driver shouldn't report they can. I would do this, but I can't test DXVA decoding right now, and I shouldn't say about things that I can't test myself.

[VipZ]

FWIW, maybe someone could contact AMD folks and tell about problems with 4k dxva decoding. I guess they are aware of that, but maybe it will encourage them to fix this faster. Or if they can't do DXVA 4k decoding at least their driver shouldn't report they can. I would do this, but I can't test DXVA decoding right now, and I shouldn't say about things that I can't test myself.

AMD has updated their drivers to simply remove 4K decoding, its only old drivers which fail with a crash or BSOD. I have reported HDMI audio issues few times and pretty sure others have done the same and AMD has yet to fix it so don't really feel like wasting time to try get AMD to fix something like 4K which is hardly used.

[wanezhiling]

AMD has updated their drivers to simply removed 4K decoding
Could you post any info about this you knew?

Of course, I will try Catalyst 12.11 Beta8 later.

[VipZ]

Could you post any info about this you knew?

Of course, I will try Catalyst 12.11 Beta8 later.

http://dl.dropbox.com/u/19338638/dxva.JPG

And LAV fallback worked as expected to use software on 4K content pre blacklist.

[wanezhiling]

http://dl.dropbox.com/u/19338638/dxva.JPG

And LAV fallback worked as expected to use software on 4K content pre blacklist.

Thanks.

AMD is aware of their "lie":p

[aufkrawall]

New versions fixes the menu issue, thanks.

[kasper93]

So blacklist added in cea9c97 will not be needed with newer driver, fallback should work fine :)

[wanezhiling]

So blacklist added in cea9c97 will not be needed with newer driver, fallback should work fine :)
Still lots of people use the older drivers.

[kasper93]

Thats is why I said "will not be" :) Anyway hardware acceleration UHD/4K is disabled by default, I guess people are not so stupid to enable it, there is even a note in the tooltip ...On AMD GPUs, 4K support is very fragile, and may even cause crashes or BSODs, use at your own risk. but @nevcairiel will do as he thinks :) I don't like when software decide for me, but maybe it's better this way. (・へ・)

BTW. No need to quote post above ;p

[nevcairiel]

I'll still keep the blacklist for AMD right now, because there is no downside. People have tested on the latest drivers, and it doesn't work, people have tested on older drivers, and it crashes. Its MPEG-2 support all over again, they claim it works, but it takes like 2 hardware generations more until it actually works. :p

[Prinz]

And don't forget: All ATI 2000,3000,4000 are on legacy support now and can't use the current drivers.