madVR - high quality video renderer (GPU assisted)

[Damien147]

Cpu is an old 4 core i5 760 OCed to @3800.Even with bilinear chroma I can't play high bitrate 4k with Madvr.You see the problem?I can play 4k sources with low bitrate and NGU med chroma though.(GPU is RX 470)


Edit:

The 4k file in the example is ~50Mbps.If nothing happens I guess I can settle to 30Mbps and less with ngu med on chroma.
Generally speaking it's not a universal solution.Solution would be dxva(native) support.You paid for your card to do HEVC decoding.

[CruNcher]

Nope that System wont cut the minimum requirements we talking 2nd Generation Core I5 at least aka Sandy Bridge IPC + Instruction set.
Also overall MadVRs Shader code is known not to be as efficient on AMDs Driver and Sahader Compiler as it is on Nvidias Driver and Shader Compiler per cycle but also here the RX 470 is overall to weak

30 Mbps is a much more sane max depending on the temporal complexity and decoder efficiency

50 Mbps with even higher Temporal based peaks is hardly anymore in the balpark for that IPC with only 4 Cores.

[Damien147]

With the OC I get 2nd Generation Core I5 performance.I don't have problem anywhere else.I can play everything I throw there from hd to 30Mbps 4k with NGU med.I need a better reason to upgrade cpu.
I am not complaining in the example for the load,I am complaining for the dropped frames and also for the NO benefits in comparison with EVR.Useless high cpu load.

[madshi]

Originally Posted by ryrynz

0.2 looks a bit heavy handed, 0.1 is okay (after upscale) still not quite as sharp as NNEDI3 256 in some areas but TBH I don't run anything over NNEDI3 64 neurons with the GTX 960 anyway, so likely this is sharper than that. Almost feel there needs to be another sharpener that can pick up on stuff other than just the more major lines.. but I guess that's what I've got Avisynth for.

Well, I had done a very quick test and NGU pixart 3 + AdaptiveSharpen 0.1 produced results that were a lot sharper than NNEDI3. Probably using very low LumaSharpen or "crispen edges" settings would also work. You could try different sharpeners to see which one produces the best results. But honestly, the sharpness difference between NGU pixart 3 and NNEDI3-256 with the Suzumiya image is really low. Almost *any* sort of post-scaling sharpening should remedy the softness difference.

Originally Posted by ryrynz

Wasn't really judging what was better as such, just my preference and I wanted to focus on what I felt was the most important considering NGU newpix had closed the gap in all the other areas.

Ok, that's fine. So does that mean you're not sure, anymore, if you like NNEDI3 better than NGU pixart 3 overall? FWIW, I'm still fine tuning pixart, but I doubt it's gonna become much sharper.

Originally Posted by Damien147

Can you do something with the black screen and HEVC in dxva(native)?

What black screen? Last time I checked it with my NVidia GPU, it worked just fine. I don't have a Polaris, so I can't test that.

Originally Posted by Damien147

User is gonna install mpc hc,madvr and is gonna be looking what's goin' on with the black screen as it is the default.
I had hardware decoder to none as suggested but I found problem with high bitrate HEVC in 4k.My pc couldn't handle it(cpu load,dropped frames) and I lowered chroma even to Bilinear.
I can run low bitrate HEVC in 4k with NGU med chroma but for high bitrate it's a no go.It seems that dxva(native) could solve that.
At the current state with high bitrate HEVC and hardware decoder to none it's not worth it to use Madvr at 4k.For everything else it's ''godlike'' but I found this small ''hole''.
I know that your priorities are different now but keep it in mind please.


Madvr with bilinear chroma and hardware decoder to none.

EVR(custom presenter) with YV12 chroma upsampling shader and hardware decoder to dxva(native).

You get this ridiculously high cpu load leading to dropped frames with no real life image quality benefits.

Let's set some facts straight:

High bitrate 4K HEVC decoding via CPU is *very* demanding. Every current CPU will struggle with that. Maybe the fastest can do it (barely), but I would expect CPU usage to be near 100%.

If you have near 100% CPU usage when playing a 4K HEVC file, then most probably you're using software/CPU decoding. If you have near 0% CPU usage, you most definitely are using hardware decoding. So if you look at your madVR vs EVR screenshots, you can see that your madVR screenshot must be from software decoding and your EVR screenshot must be from hardware decoding.

My recommendation is to use "DXVA copyback". But "DXVA native" should work fine, too.

Decoding a video file and *processing* it runs through totally different hardware circuits, so it doesn't really have anything to do with each other. If the decoder doesn't manage to decode the video fast enough, then even Bilinear won't help, because Bilinear vs NGU is running on the Pixel Shaders, while video decoding has nothing to do with Pixel Shaders.

[pose]

Damien147, which MPC version are you using? Try latest nightly build. I also have old i5 and RX 470. I will test some high bitrate 4K video and report back.

[CruNcher]

Originally Posted by madshi

Well, I had done a very quick test and NGU pixart 3 + AdaptiveSharpen 0.1 produced results that were a lot sharper than NNEDI3. Probably using very low LumaSharpen or "crispen edges" settings would also work. You could try different sharpeners to see which one produces the best results. But honestly, the sharpness difference between NGU pixart 3 and NNEDI3-256 with the Suzumiya image is really low. Almost *any* sort of post-scaling sharpening should remedy the softness difference.


Ok, that's fine. So does that mean you're not sure, anymore, if you like NNEDI3 better than NGU pixart 3 overall? FWIW, I'm still fine tuning pixart, but I doubt it's gonna become much sharper.


What black screen? Last time I checked it with my NVidia GPU, it worked just fine. I don't have a Polaris, so I can't test that.


Let's set some facts straight:

High bitrate 4K HEVC decoding via CPU is *very* demanding. Every current CPU will struggle with that. Maybe the fastest can do it (barely), but I would expect CPU usage to be near 100%.

If you have near 100% CPU usage when playing a 4K HEVC file, then most probably you're using software/CPU decoding. If you have near 0% CPU usage, you most definitely are using hardware decoding. So if you look at your madVR vs EVR screenshots, you can see that your madVR screenshot must be from software decoding and your EVR screenshot must be from hardware decoding.

My recommendation is to use "DXVA copyback". But "DXVA native" should work fine, too.

Decoding a video file and *processing* it runs through totally different hardware circuits, so it doesn't really have anything to do with each other. If the decoder doesn't manage to decode the video fast enough, then even Bilinear won't help, because Bilinear vs NGU is running on the Pixel Shaders, while video decoding has nothing to do with Pixel Shaders.

You also starting to bring the Driver into trouble with 100% Utilization on all 4 Cores (not only Video also Audio) and at the same time getting a stable Audio and Video Rendering without Frame Droops you have to be very precise scheduling depending also on the Underlying OS and also here overall the CPU architecture plays a big role how good it will manage that timer and frequency switching situation

If your code isn't excellently Multithreaded and every cycle optimized you should never try to overdo that and go above ~80%

But overall it still depends on the whole System configuration and latency and you have to find out what your system is capable of at that Edge and how well you tuned it

[Damien147]

@Madshi

I pointed the software and hardware decoding thing.
I am not complaining for the high cpu load specifically but for the whole result.If you suggest DXVA copyback then ok,I'm gonna use that but:
With dxva(native) and 10 bit HEVC I get black screen.With EVR and dxva(native) for the same file I don't get black screen and file plays well.
With dxva(copyback) and 10 bit HEVC I get dropped frames and picture slows down and ''skips''.
Yes it's a polaris gpu but if I am the only one ignore me.

@pose

Using latest official mpc-hc.Gonna try nightly to check.Try higher than 30Mbps 4K HEVC .

[madshi]

Originally Posted by Damien147

With dxva(copyback) and 10 bit HEVC I get dropped frames and picture slows down and ''skips''.

Then please post a screenshot of your OSD (Ctrl+J) while these slow downs and skips happen.

[Damien147]

Here you are.dxva(copyback) http://imgur.com/a/doUFv
I can play the file well with ''none''.

[madshi]

Are you sure that's really dxva copyback? Because the decoder queue is almost empty, which means that the decoder is not fast enough. How high is CPU usage in this situation? Does DXVA copyback run smoothly when using EVR?

[Damien147]

Yes,sure.Here is my cpu usage with DXVA copyback http://imgur.com/a/X4Px4 .DXVA copyback with EVR seems better.
With hardware decoder to none I can play well this file with Madvr and same settings.It's a lower bitrate 4K HEVC.

[pose]

Damien147, just tested 4K 60FPS 75.6 Mb/s bitrate HEVC video. Latest MPC-HC nightly build. Latest GPU drivers. Win7 64bit. Intel i5 ivy bridge 3.6 Ghz + RX 470.

Hardware decoder: none (software). 100% load on 4 cores.
Hardware decoder: DXVA2 (copy-back). 35% load on 4 cores.
Hardware decoder: DXVA2 (native). Black screen.

Judging by my OSD stats i can tell, that if you have 100% load on all of your CPU cores then it will BOTTLENECK madVR playback. If your CPU has some room then VIDEOCARD is going to bottleneck madVR playback.

I also noticed that you have SMOOTH MOTION ON. Try disabling it. On my 60FPS sample it made big difference in rendering times.

I also need to remind you that you need to RESTART your video player after you change LAV's hardware decoder settings or changes wont apply.

[madshi]

Originally Posted by Damien147

Yes,sure.Here is my cpu usage with DXVA copyback http://imgur.com/a/X4Px4 .DXVA copyback with EVR seems better.

How do you define "better"? Do you get perfectly smooth playback with EVR and DXVA copyback, same as with low-bitrate software decoding? Or not?

[Damien147]

Originally Posted by pose

Damien147, just tested 4K 60FPS 75.6 Mb/s bitrate HEVC video. Latest MPC-HC nightly build. Latest GPU drivers. Win7 64bit. Intel i5 ivy bridge 3.6 Ghz + RX 470.

Hardware decoder: none (software). 100% load on 4 cores.
Hardware decoder: DXVA2 (copy-back). 35% load on 4 cores.
Hardware decoder: DXVA2 (native). Black screen.

Judging by my OSD stats i can tell, that if you have 100% load on all of your CPU cores then it will BOTTLENECK madVR playback. If your CPU has some room then VIDEOCARD is going to bottleneck madVR playback.

I also noticed that you have SMOOTH MOTION ON. Try disabling it. On my 60FPS sample it made big difference in rendering times.

I also need to remind you that you need to RESTART your video player after you change LAV's hardware decoder settings or changes wont apply.

I don't think that smooth motion is the problem.
Tried mpc-hc nightly and I get same result.So for now I'm gonna stick with hardware decoder to none and low bitrate 4k HEVC.....
dxva(copyback)plays well with you?No slow downs and ''skips''?

Originally Posted by madshi

How do you define "better"? Do you get perfectly smooth playback with EVR and DXVA copyback, same as with low-bitrate software decoding? Or not?

Seems smooth with some slow down ''spikes'' I think.Frame rate gets slowed to 21 from 23.976,so it's not ''smooth''.

[madshi]

Originally Posted by Damien147

Seems smooth with some slow down ''spikes'' I think.

The key question is if the decoder is decoding in full speed or not. So do you get the same fluid playback with software decoding and DXVA copyback decoding, when using EVR? Or not?

[pose]

Originally Posted by Damien147

I don't thing that smooth motion is the problem.
Tried mpc-hc nightly and I get same result.So for now I'm gonna stick with hardware decoder to none and low bitrate 4k HEVC.....

Set hardware decoder to DXVA2 (copy-back) -> restart MPC -> disable smooth motion -> screenshot OSD and post here please.

Originally Posted by Damien147

dxva(copyback)plays well with you?No slow downs and ''skips''?

Plays perfectly.

[QBhd]

@Damien147,

Why is defrag running while doing this? If that is running on the drive you are trying to play back videos from it can cause all sorts of problems. I'm not suggesting it is the problem, but you really should be trying to troubleshoot without that running.

QB

[Damien147]

Originally Posted by madshi

The key question is if the decoder is decoding in full speed or not. So do you get the same fluid playback with software decoding and DXVA copyback decoding, when using EVR? Or not?

Not same,with EVR and DXVA copyback it's worst than setting decoding to none but better than madvr.

Originally Posted by pose

Set hardware decoder to DXVA2 (copy-back) -> restart MPC -> disable smooth motion -> screenshot OSD and post here please.


Plays perfectly.

Oh great,now I get black screen in fullscreen.Here's one without fullscreen and smooth motion off.http://imgur.com/a/X40TS
If it plays perfectly with same gpu then I am the problem but I don't know where....

Originally Posted by QBhd

@Damien147,

Why is defrag running while doing this? If that is running on the drive you are trying to play back videos from it can cause all sorts of problems. I'm not suggesting it is the problem, but you really should be trying to troubleshoot without that running.

QB

Defrag is paused,it doesn't do anything.

[Damien147]

Guys,till now I was avoiding Display Driver Uninstaller but I am gonna use it now because @pose with same gpu has smooth playback with DXVA(copyback).I am gonna report back.

[xyndv]

Originally Posted by Warner306

Have you tried using a lower value of anti-bloating (e.g. 25%)?

Yes, but I assume that would also apply less of the anti bloating effect.

Originally Posted by madshi

Which sharpness algos are we talking about? And which maximum limit would you like to have?

I'd like to have a maximum of 10 for AdaptiveSharpen and LumaSharpen (some headroom added)
Although, personally I'd prefer all values to be as unlimited as possible, if only for experimental purposes.

Thanks!