LAV Filters - DirectShow Media Splitter and Decoders

[NikosD]

All three modes use the same amount of CPU for decoding.

P.S
The intel on chip gpu is permanently disable in the bios.
I talk strictly about the GTX600, In P8 State! which is downclocked to around 320 Mhz for both GPU and MCU.

Don't know much of Nvidia HW but it seems that you are not using hybrid mode since GPU usage is zero and it looks like more SW decoding since the CPU usage is the same for all modes.

You probably "cheat" LAV Video thinking that your GPU can hybrid decode HEVC and you (?) have managed to put your card in a state being incapable of hybrid decoding.

Can you put Nvidia in "normal" state or it goes in that state automatically ?

[huhn]

can you try it at stock clocks.

the hybrid decoder usually use a ton of GPU power...

[nevcairiel]

Don't know much of Nvidia HW but it seems that you are not using hybrid mode since GPU usage is zero and it looks like more SW decoding since the CPU usage is the same for all modes.

He said VPU usage is 0%, which is normal for hybrid. He never mentioned his GPU usage - which wouldn't be zero if he uses madVR anyway.

[NikosD]

He said VPU usage is 0%, which is normal for hybrid. He never mentioned his GPU usage - which wouldn't be zero if he uses madVR anyway.
Another non careful reader.

He wrote and you replied that it's not using GPU at all - which means GPU usage 0%

Hybrid?
It's not using the GPU at all, VPU is at 0% and still it says "Active Decoder: dxva2cb direct", resulting in a lot of glitches.

[nevcairiel]

Thats your interpretation. Mine is that he only refers to the VPU. He uses madVR, there is no way GPU usage is absolutely 0%.

[James Freeman]

Nev is right, I meant "not using the GPU at all FOR DECODING thus VPU is at 0%", but the main GPU usage is definitely heavily loaded in p8.
I edited the post with P0/P2 (full power) info.

[nevcairiel]

Are you forcing it to stay in P8? It should automatically switch modes as it sees the need for performance.

[NikosD]

Nev is right, I meant "not using the GPU at all FOR DECODING thus VPU is at 0%", but the main GPU usage is definitely heavily loaded in p8.
I edited the post with P0/P2 (full power) info.
I asked you in my previous post :

" Why don't you use DXVA native and report back your results by monitoring GPU usage ?"

You didn't reply about GPU usage and thought it was 0% like the first mention of not using GPU at all.

Well, can you report the GPU usage using GPU - Z for the two modes :

P0/P2 and P8 ?

[James Freeman]

Are you forcing it to stay in P8?
Yes! Have done for years.
When all works properly 320Mhz is more than enough with moderate (no NNEDI3 stuff) madVR settings.

Besides, the P state is always switching when I'm browsing the internet, so I lock it to P8.
In other words, Adaptive mode in with Nvidia is not acceptable for my needs and I don't play games at all.

My i7 is locked in 1600Mhz (0% = maximum & minimum power in windows Power Management), and the GPU is always in P8.
The setup draws less than 50W even at load.... waste not.

[nevcairiel]

That would explain it then, Hybrid decoding can need quite a chunk of performance from both the CPU and GPU, so its probably pushing something over the edge.

[James Freeman]

Well, can you report the GPU usage using GPU - Z for the two modes :
P0/P2 and P8 ?

GPU usage (not MCU, not VPU), HEVC On.

DXVA2 copy-back,
P8 (320 Mhz): 70%
P2 (1020 Mhz) : 12%

DXVA2 native,
P8 (320 Mhz): 70%
P2 (1020 Mhz) : 12%

Software,
P8 (320 Mhz): 46%
P2 (1020 Mhz) : 8%

Apparently you are right NikosD, there is more GPU load in DXVA2 than in software, probably because the Hybrid topology.

Okay thanks everybody, I'm sticking with HEVC Off.
I don't see a reason to load the GPU when the CPU is doing absolutely nothing.

For x264, VPU is doing its job very well.
I would love to load the VPU with x265 with my next video card...

[NikosD]

GPU usage (not MCU, not VPU), HEVC On.

DXVA2 copy-back,
P8 (320 Mhz): 70%
P2 (1020 Mhz) : 12%

DXVA2 native,
P8 (320 Mhz): 70%
P2 (1020 Mhz) : 12%


I can't interpret - as I would like to - your results.

A GPU usage of 12% in DXVA Hybrid decoding and P2 state looks like something is not working good.

It seems that it's actually the CPU that does the decoding and GPU is helping a little or maybe your HEVC file is not so demanding in terms of resolution, bitrate, FPS or the GPU driver is not optimized for hybrid decoding.

Can you try a 4K HEVC file with high frame rate or the same HEVC file that you see 12% GPU usage in "Decode Benchmark" mode using DXVA Checker ?

In the last case you should see 100% GPU usage for any clip.

[Fadeout]

Not sure if it's a problem of LAV, but recording videos at 1080i with Avermedia box creates files at 50 fps. If then I load these in MPC-HC with LAV configured to use Nvidia CUVID I end up with playback going at twice the speed.

If I remove CUVID and do the decoding normally, everything plays normal.

I tried various things but I couldn't make CUVID play these files properly...

[NikosD]

@Ligh


He is reporting the right monitoring parameter - which is GPU usage in his case - since he doesn't have a pure HW HEVC decoder in his GPU in order to use it and he can only decode HEVC in hybrid mode using shaders.

You weren't so careful reading his post.

I was wrong and you were right (like nevcairiel)

[LigH]

... since he doesn't have a pure HW HEVC decoder in his GPU in order to use it and he can only decode HEVC in hybrid mode using shaders.

Maybe I just got confused between "hybrid decoding" and "Hybrid by Selur"; or I wasn't aware that shader units are indeed active in this hybrid decoding mode. :o

But if you found an explanation, fine. Such hardware is far beyond my budget, anyway.

[Budtz]

I’m wondering if Lav filters could have any effect on Reclock. I seem to have some problems with reclocks frame rate detection and the "assumed frame rate when not found" function. If I set "assumed frame rate when not found" to 23,976 it works fine for all material with this framerate. But if I then play material where the source filters says its 25p - the only way I can get Reclock to detect this probably is by setting "assumed frame rate when not found" to 25p. I figured out that it works for all kinds of material if I set "assumed frame rate when not found" to unknown. Then media detection info is on automatic and it auto detects correctly. Basically media detection info is set to the same as "assumed frame rate when not found" setting.

I then updated Lav filters and it suddenly worked. I could set "assumed frame rate when not found" to 23,976 and it would work for both 24p and 25p content. Now I reformatted to install Windows 10 and I’m back to square one.

Does any1 know anything about this?

[Bandito]

If we're using MPC-HC 64bit we should install the LAV 64bit also right? Can we install 32bit and 64bit together?

Also what LAV hardware acceleration should I use for Nvidia card? If I don't use Cuvid should I enable software deinterlacing for DXVA??

[vivan]

MPC-HC comes with LAV Filters - so you shouldn't.
But if you really want to, for some reason (and don't want to just replace files) - then yes. Also don't forget to select installed filters in External Filters.

[romulous]

Hi nev,

I have a sample file - download below - that has two subtitle streams (Dutch, and English). When LAV is the splitter, only one is exposed to the player (the Dutch). When using Haali's or MPC-BE's internal splitter, both streams are exposed. Here are a few simplified GraphStudio screenshots demonstrating.

One Haali graph, one LAV graph:
http://i.imgur.com/FDmftOF.jpg

Stream Selection for Haali Graph:
http://i.imgur.com/fs4ZDGB.jpg

Stream Selection for LAV Graph:
http://i.imgur.com/6H9LmQH.jpg

10MB Sample:
https://dl.dropboxusercontent.com/u/105555957/Subs.mkv

Would it be possible for LAV to expose both subtitle pins to the player in this case?

Thanks,

romulous

[nevcairiel]

Would it be possible for LAV to expose both subtitle pins to the player in this case?


The second track has the "Enabled" flag set to "no", and LAV obeys such flags, which causes the track to be hidden.
If you want a track to be available, I recommend you set Enabled to yes in the file.... :)

[romulous]

Interesting - is there a particular tool you use to see that flag? MediaInfo does not show it at all.

romulous

[nevcairiel]

mkvmerge-gui -> header editor

Someone sure must've gone out of their way to set it. ;)

[romulous]

Ok, thanks - that tool may come in handy for the future, I didn't know that flag was a thing (just assumed the only two subtitle flags were the two shown by MediaInfo, ie Default and Forced).

romulous

[videonerd]

Hi nevcairiel, I am using mpchc64/internal lav filters/madvr and with software decoding and notice for high bitrate files e.g. youtube 4k60p, the decoder queues are not filling up even though the CPU load in total and at individual core level (as measured by task manager / hwinfo64) indicate that it is only at circa 70%.

May I seek your insight on whether this is lav filters intentionally not maxing out the available cpu cycles or are there other factors in the chain that is preventing the cpu from maxing out?

Thank you!

[jebediah]

Just want to know why i don't get any sound from Cloverfield.2008.PROPER.720p.BluRay.x264-SiNNERS when i'm using LAV, but with ffdshow it works perfect. Any ideas?

[vivan]

Hi nevcairiel, I am using mpchc64/internal lav filters/madvr and with software decoding and notice for high bitrate files e.g. youtube 4k60p, the decoder queues are not filling up even though the CPU load in total and at individual core level (as measured by task manager / hwinfo64) indicate that it is only at circa 70%.

May I seek your insight on whether this is lav filters intentionally not maxing out the available cpu cycles or are there other factors in the chain that is preventing the cpu from maxing out?

Thank you!Try using more threads and increasing CPU queue in madVR.

[LigH]

Just want to know why i don't get any sound from Cloverfield.2008.PROPER.720p.BluRay.x264-SiNNERS when i'm using LAV, but with ffdshow it works perfect. Any ideas?

Because #6 (http://forum.doom9.org/forum-rules.htm).

[videonerd]

Try using more threads and increasing CPU queue in madVR.

The threads have already been maxed out and increasing CPU queues will not help as it is not even filling up - that is the problem.

[P.J]

LAV can't use more than 2 threads for decoding H.264 10bit 4:2:2, why? :confused:

[nevcairiel]

Some clips just don't scale well with multithreading.

[ryrynz]

Some clips just don't scale well with multithreading.

Does the code change enough to warrant it being looked into again in the future?

[foxyshadis]

Does the code change enough to warrant it being looked into again in the future?

Not unless memory bandwidth improves significantly.

[P.J]

Anyway to choose the Processor Device in LAV? :confused:


Not unless memory bandwidth improves significantly.

Isn't Dual DDR3 1600 enough?

[ryrynz]

Isn't Dual DDR3 1600 enough?

DDR is quite slow, there's considerable room for improvement. A large L4 could help a little, but I guess we're waiting for HMC which would deliver over 16x the performance of DDR4-3200.

[red5goahead]

Is it will possibile to have a parametric equalizer embedded directly into Lav audio to a precise calibration of the room, for example using Rew as free tool sources ?

[nevcairiel]

Is it will possibile to have a parametric equalizer embedded directly into Lav audio to a precise calibration of the room, for example using Rew as free tool sources ?

No, that is not something LAV will ever do.

[DragonQ]

I'm finding myself getting more and more annoyed at not being able to hear speech in films since I moved to a stereo setup (no room for my rear speakers). My AVR is set to stereo and so is Windows on my HTPC. I use LAV Audio so I thought I'd set it to downmix for me and boost the centre channel to 1.0, but it doesn't seem to do anything to help. Am I missing something or is there a better way to do this? Unfortunately my AVR doesn't let me change the way it downmixes.

[nussman]

You can try to boost centre channel in LAV and output multichannel PCM and let your AVR do the downmix.
Works only for multichannel sources and PCM output of course.

[clsid]

You must disable bitstreaming. You should then be able to hear the contents of the center channel. If not, double check that everything is really set to stereo.

If the volume is just too low, then enable volume normalization in MPC-HC, or enable Loudness Equalization in Windows.

[trandoanhung1991]

LAV Video decoding always show avcodec as the active codec regardless of which I choose: DXVA CB/Native/NVIDIA CUVID. I'm using a GTX 970 on Windows 10. Any ideas?

[sneaker_ger]

Not everything is compatible with DXVA/CUVID decoding. Post MediaInfo and/or sample of the file you are testing if you want more info.

[trandoanhung1991]

Not everything is compatible with DXVA/CUVID decoding. Post MediaInfo and/or sample of the file you are testing if you want more info.

On closer inspection it's a H264 Hi10 file. I thought LAV worked with 10 bit files?

[NikosD]

Yes, it works only using SW decoding because there is no GPU capable for HW H.264 10bit decoding.

[DragonQ]

You must disable bitstreaming. You should then be able to hear the contents of the center channel. If not, double check that everything is really set to stereo.

If the volume is just too low, then enable volume normalization in MPC-HC, or enable Loudness Equalization in Windows.
Ah, that's a good point. I think I have bitstreaming enabled for Dolby TrueHD to workaround a bug where the video gets very choppy. That may be the issue, which annoyingly means there's not really a way around it.

It's not the volume, it's just the volume of the centre channel relative to everything else.

[NikosD]

Not unless memory bandwidth improves significantly.


Isn't Dual DDR3 1600 enough?

DDR is quite slow, there's considerable room for improvement. A large L4 could help a little, but I guess we're waiting for HMC which would deliver over 16x the performance of DDR4-3200.

Optimized multi-threaded apps with vectored code (by intelligent compiler or assembly) are starved for memory bandwidth of System RAM.

The internal cache speed and the execution units of Haswell for example are monstrous, but the system RAM is a poor slow DDR3-1600.

A little survey regarding System RAM and cache speed of the last 7 years.

Core2Duo@2.83GHz (2008)

RAM: 7 GB/s
L2: 39 GB/s
L1: 90 GB/s

Sandybridge-Core i5@3.3GHz (2011)

RAM: 20 GB/s
L3: 138 GB/s
L2: 208 GB/s
L1: 400 GB/s

Haswell-Core i7@4.0GHz (2013)

RAM: 23 GB/s
L3: 195 GB/s
L2: 365 GB/s
L1: 950 GB/s (!)

The speedup of L1 cache is according to the speed of execution units that System RAM clearly can't follow.

[balkerman]

Ah, that's a good point. I think I have bitstreaming enabled for Dolby TrueHD to workaround a bug where the video gets very choppy. That may be the issue, which annoyingly means there's not really a way around it.

It's not the volume, it's just the volume of the centre channel relative to everything else.

I would send everything untouched to the avr and let that do the stereo downmix. That's what is was made for...

[huhn]

I would send everything untouched to the avr and let that do the stereo downmix. That's what is was made for...

that his issue the AVR isn't down mixing correctly if bitstreaming was used.

[The_Nephilim]

Hi All,

I have Stereoscopic player and downloaded the LAv Filter .065.. now I am trying to get the DTS HD Working on my setup but the movie does not play when in LAV Audio config set to DTS HD..

I ripped it so I can play in Stereoscopic player and got the DTS HD TRacks too. I ripped them to a HDD..

Now I had this working before but do not remember what I did to get it working..

here is my system I run it on:

Acer X1261P Projector NVidia certified 3D
Intel i5 3570K
eVGA 980GTX
8gb Corsair ram
Samsung Evo 840SSD
Built in Realtek soundcard 7.1
Buttkicker Simulation..

I just reinstalled windows and reinstalled Stereoscopic player and dl LAv Filters v65 and I need some help..

If you need more info Please lte me know what you need??

[Neroldy]

Hi nevcairiel,
I have a ts file played with mpc-hc using lav will display some block in the first few seconds. Like the picture in the attachments.
I used MPC-HC 1.7.9.145 x64 with LAV 0.65.0.49-git with DXVA2(naive/copy-back will be the same result) opened.
When I switch to the SW and everying is fine.
I also used the x86 MPC-HC and LAV, same result.
So I think it may be something wrong in the HW part.
My card is AMD Radeon HD 8650G + HD 8790M Dual Graphics
The driver is 15.20.1062.

Here is a 10 seconds sample
http://www.mediafire.com/download/f15ff925tlql7lm/sample.ts

thanks~

[IanD]

Any likelihood that LAV video decoder might include frame interpolation for H.264 (or legacy mpeg using motion vectors) in future?