MPC-HC v2.2.1

[huhn]

yes they are both copyback and there is no reason to use them.

[NikosD]

Originally Posted by Nintendo Maniac 64

Do you know if black bar detection is supported with LAV's CUVID and QuickSync decoders?

Originally Posted by huhn

yes they are both copyback and there is no reason to use them.

CUVID is not copyback and it's as fast or even slightly faster than DXVA2 native.

Also CUVID supports in general even more HW decoding codecs than DXVA2 but not using LAV filters e.g VP8

[nevcairiel]

Originally Posted by NikosD

CUVID is not copyback and it's as fast or even slightly faster than DXVA2 native.

Thats not true. CUVID is very similar to DXVA2 Copy-Back, and will suffer from the same bandwidth related problems at like 4k 10-bit 60 fps that DXVA2 Copy-Back starts to run into issues with.
You *could* use it without Copy-Back, but of course no renderer supports that.

The native DXVA2 or D3D11 decoders are going to be more efficient, just because they don't have to send the image through system memory.

[NikosD]

Originally Posted by nevcairiel

Thats not true. CUVID is very similar to DXVA2 Copy-Back, and will suffer from the same bandwidth related problems at like 4k 10-bit 60 fps that DXVA2 Copy-Back starts to run into issues with.

But then how is it possible for CUVID to be faster or as fast than DXVA2 native even for 4K60fps 10bit clips or even 8K60fps clips ?

Where is the overhead of copy-back process hidden ?

[huhn]

for a UHD 60 fps 10 bit file i get.
real world playback the only thing that really matters here using mpcVR.
dxva2 copyback 2.1% at 1 core at ~1200 mhz or lower
cuvid 2.5% at 1-2 core changing heavily changing
dxva native up to 0.3% CPU at idle 300 mhz audio decoding i guess.

does CUIVD even support 10 bit now native or is 16 bit or dithering still needed?

[aufkrawall]

cuvid is dead anyway. While nvdec builds on cuda, it has superior parsing logic and likely more bug fixes and more supported formats these days. I think you can play any format the hardware supports with it in ffmpeg/mpv (also without copyback), e.g. HEVC 12 bit and probably also 4:4:4 video.
It forces the GPU into cuda pstate, which makes it burn lots of energy, but perhaps look better in some benchmarks. And there seems to be no way to use native GPU deinterlacing, apart from copyback or a ported yadif cuda filter in ffmpeg (which is extremely ugly to support at compile time).

[NikosD]

Originally Posted by huhn

for a UHD 60 fps 10 bit file i get.

real world playback the only thing that really matters here using mpcVR.

dxva2 copyback 2.1% at 1 core at ~1200 mhz or lower

cuvid 2.5% at 1-2 core changing heavily changing

dxva native up to 0.3% CPU at idle 300 mhz audio decoding i guess.

For the purpose of the current discussion, we care more about the GPU usage caused by using copy-back procedure, than CPU usage.

What is the GPU usage during normal playback of that file using the three modes (DXVA2 copy-back, DXVA2 native, CUVID) ?

[aufkrawall]

Why do you think cuvid could magically make the GPU faster? If you want to have high performance and a decent renderer at the same time, there is no way around mpv anyway.

[huhn]

if you care about GPU usage CUIVD already lost because it is pushing the GPU way to much.
GPU usages is hard to read when the GPU isn't using it's highest available clocks.

1060
DXVA2 native core 1050 mem 2003.4 load 10% TDP 23 %
DXVA2 copyback core 1544 mem 2003.4 load 41% TDP 27%
cuvid core 1544 mem 1901.2 load 41% TDP 27%

don't take this system to serious it goes pretty nuts without native decoding since i reinstalled it.

[nevcairiel]

Originally Posted by NikosD

But then how is it possible for CUVID to be faster or as fast than DXVA2 native even for 4K60fps 10bit clips or even 8K60fps clips ?

Benchmarking speeds are irrelevant because CUVID pushes the GPU to run at max clock at all times, which wastes a lot of energy.
During normal playback, the load on the system with any sort of copy-back implementation like CUVID is going to be worse then a native implementation.

Originally Posted by aufkrawall

And there seems to be no way to use native GPU deinterlacing

Thats only a FFmpeg limitation, and not an inherent problem.

I've been thinking about replacing CUVID with NVDEC in LAV, with built-in deinterlacing and either copy-back like today or interop with D3D11, so that it could be efficient like D3D11 native, just with deinterlacing, and potential 12-bit and maybe 4:4:4 support (the latter would likely need renderer support first)

[huhn]

if i would use a FHD 23p file or a DVD CUVID power consumption would look like a bad joke UHD 60p 10 bit is pretty much a best case scenario for CUVID.

[chros]

Originally Posted by nevcairiel

Thats not true. CUVID is very similar to DXVA2 Copy-Back, and will suffer from the same bandwidth related problems at like 4k 10-bit 60 fps that DXVA2 Copy-Back starts to run into issues with.
You *could* use it without Copy-Back, but of course no renderer supports that.

Thanks, I didn't know / remember that.
All I remember is that CUVID was a bit faster on my optimus laptop (just allowing to watch UHD 23p content that I couldn't with dxva copyback).

Originally Posted by huhn

1060
DXVA2 native core 1050 mem 2003.4 load 10% TDP 23 %
DXVA2 copyback core 1544 mem 2003.4 load 41% TDP 27%
cuvid core 1544 mem 1901.2 load 41% TDP 27%

I'll update this thread with the CUVID results, just out of curiosity.

Originally Posted by huhn

if you care about GPU usage CUIVD already lost because it is pushing the GPU way to much. GPU usages is hard to read when the GPU isn't using it's highest available clocks.

Originally Posted by nevcairiel

Benchmarking speeds are irrelevant because CUVID pushes the GPU to run at max clock at all times, which wastes a lot of energy.

Originally Posted by huhn

if i would use a FHD 23p file or a DVD CUVID power consumption would look like a bad joke

You are right, but since majority of the users here are using madVR it's a bit different story. E.g. I have profiles that max out my (slightly underclocked) GPU anyway, no matter what video file you throw at it: it always runs at its highest allowed state/frequency (1544 Mhz).
So this way it's easy to benchmark it.

Originally Posted by nevcairiel

I've been thinking about replacing CUVID with NVDEC in LAV, with built-in deinterlacing and either copy-back like today or interop with D3D11, so that it could be efficient like D3D11 native, just with deinterlacing, and potential 12-bit and maybe 4:4:4 support (the latter would likely need renderer support first)

That's a really good idea, thanks!
Maybe you could still keep CUVID as a separate option, but this decision is yours, of course (maintaining, dealing with possible issues, etc).

[sneaker_ger]

Originally Posted by nevcairiel

I've been thinking about replacing CUVID with NVDEC in LAV, with built-in deinterlacing and either copy-back like today or interop with D3D11, so that it could be efficient like D3D11 native, just with deinterlacing

Please think about throwing in debanding while you're at it.

[huhn]

this is not a madVR thread and i highly doubt that most mpc-hc user are using madVR.

NVDEC is CUVID or better an update all new version have this name so no need to keep both. NVDEC the real power of NVDEC is it's far better feature set and the fact that they don't need to wait and hope windows will add codec X with profile Y support.

[aufkrawall]

Originally Posted by nevcairiel

Thats only a FFmpeg limitation, and not an inherent problem.

But why do D3D11VA and VAAPI inside ffmpeg don't have such limitations? Nvidia could provide patches, couldn't they?

[nevcairiel]

Originally Posted by sneaker_ger

Please think about throwing in debanding while you're at it.

I would only expose deinterlacing because thats what the NVDEC API offers with like one simple call, no other manually applied video processing.

[olli66]

hi!
I experience a problem with fullscreen HDR since the latest Windows update. I use mpc hc plus mad vr. HDR content always worked. Now it doesn't anymore when the player goes into fullscreen, the colors are then washed out. I can't find a workaround or fix.
Any ideas? I have the latest Nvidia drivers installed. I am pretty sure it is a Windows issue.

[clsid]

Let madVR do the tone mapping instead of your screen.

madVR > devices > hdr > tone map HDR using pixel shaders

[clsid]

Hi everyone,

Development of a dark theme for MPC-HC is almost finished. Please test it and let me know if you see any visual bugs, or bugs when interacting with the player GUI.

test build (x64)
test build (x86)

Dark theme can be activated through: menu > view > dark theme

Please test both the dark theme and the normal theme.

[stax76]

https://postimg.cc/Cd9vdpZJ