Media Player .NET (MPDN) - D3D HQ GPU Video Renderer [v2.49.0/v1.31.0 27 Dec 2018]

[burfadel]

Originally Posted by burfadel

187.039 fps this time, AMD R9-280X.

Please note that this is on build 10051 of the Windows 10 TP, running Windows 8.1 drivers. It would be interesting to see whether those running the Windows 10 drivers (which are feature incomplete) have any better performance.

I'm not using the Windows 10 drivers as I require the feature in the CCC where you set the screen resolution and overscan etc. Without this the image has a small black border around it (only on the connected TV) which is annoying!

The driver does seem to be quite 'different' in many ways.

[huhn]

this value will not change. PCIe 3.0 can't do a lot more.
the windows 10 feature have nothing to do with directx9EX openCL interop so it is very unlikely AMD will change things in the openCL and directx9 part..

[burfadel]

Originally Posted by huhn

this value will not change. PCIe 3.0 can't do a lot more.
the windows 10 feature have nothing to do with directx9EX openCL interop so it is very unlikely AMD will change things in the openCL and directx9 part..

Maybe there could be a WDDM 2.0 specific capable player version sometime in the future . I'm guessing that is restricted to when SharpDX is updated, which going by the commitlog has had some major rework done in preparation for version 3.

Then there's this:

Projects that are still requiring to use .NET 2.0 or .NET 4.0, or API like Direct3D10 will have to continue using SharpDX 2.6.3.

The support for these functions could be provided by a 'support pack', to be installed in a 'Legacy' folder or something. Just an idea .

[Zachs]

Hi all,

Thanks again for helping to test DX-OpenCL interop on the various GPUs. MPDN v2.25.0 has been released with OpenCL support via RenderScript.

I've added an example OpenCL script on GitHub that functions as a blue tint filter. If you want to test it out, simply copy the RenderScripts folder in Examples to your MPDN Extensions folder.

I'd be interested to see if it works on the GPUs you have tested DxClTest on.

Cheers.

[Zachs]

Originally Posted by RenderGuy2

Unfortunately I'm ignorant to the possibilities of render scrips. Would it be possible to write a render script that would take a 1920x1080@47.952 fps source and stack even frames over odd frames with a 45 pixel gap/border between frames? The resulting output should be 1920x2205@23.976 fps. Currently I accomplish this by performing these operations with Avisynth and just play the .avs with the player. Seems like these sorts of things could be done faster by a GPU. The point of all this is to produce an HDMI 1.4a compliant frame packed stream I can play on a 3d television. Thanks for the advice.

This is now possible with MPDN v2.25.0. I've added an example that performs frame rate halving on GitHub. RenderScript now has the capability to do what you're after, but someone would still need to write a script that does what you described.

MPDN now also allows you to access its render queue in RenderScripts (spatiotemporal image processing is now possible). Or in your case, you'd probably want to use it as a scratch pad for the odd frames before you stack the corresponding even frame on top of it.

[patul]

I get blank screen if I activate any of those RenderScript under Examples namespace except FrameRateHalver. Tried all versions of blue tint scripts to the same result.

Intel HD4000

[Zachs]

Make sure you copy all the files and folders under that folder. Sounds like you are missing the corresponding hlsl and cl files.

[patul]

Whoops, moving Example.*.cs into Renderscripts folder as well as Examples folder which contains the .hlsl and .cl solves the problem. Thanks. I put Example.*.cs into another Example folder under Renderscripts folder previously. All four of blue tint scripts work.

[Zachs]

Does anyone have the nnedi3 weights handy?

[Scyna]

with test 12 amd 290 189/194

[Zachs]

Hi everyone,

Shiandow has kindly ported NNEDI3 to MPDN. It's currently in my nnedi3 branch on GitHub so if anyone wants to give it a test, feel free to do so. We need AMD card users especially to test it as we don't have an AMD card to work with and would appreciate your feedback on the following. It currently only upscales luma and uses MPDN's upscaler for chroma.

1) GPU name
2) 16/32/64/128/256 Neurons - GPU load / clock speed when it's running
3) Source clip resolution / frame rate
4) Target resolution (must be bigger than source clip res)

5) Render quality settings

* Note1: If you start dropping frames, don't include the result with that neuron setting.
* Note2: We would appreciate if you could test it on sources with different resolutions too.

e.g.

1) Intel HD P4600
2) 16: 41%/600Mhz, 32: 34%/600, 64: 47%/600Mhz, 128: 72%/800Mhz, 256: 71%/1150Mhz
3) 320x180 23p
4) 1280x720
5) Max Performance

2) 16: 42%/600Mhz, 32: 35%/600, 64: 49%/600Mhz, 128: 74%/800Mhz, 256: 73%/1150Mhz
5) Image Quality

To get consistent comparisons, please set your MPDN to the following,

1) Disable fluid motion
2) Use 'no dithering'
3) Presentation Direct3D 9Ex
4) Video out bitdepth = 8
5) Disable subtitles
6) No other render scripts
7) Luma/Chroma upscaler = Bicubic (No AR sharpness 75)
8) Luma/Chroma downscaler (not used by anyway) = Bicubic (No AR sharpness 75)

9) Make sure no other apps are running - treat it like a benchmark run

Links (not compatible with MPDN v2.25.3):

MPDN test build v2.25.4
Github nnedi3 branch

[huhn]

r9 270

640x480 -> 1440x1080
256
950/1400 49% 31ms (close to the limited)
128
950/1400 32% 22ms
64
950/1400 21% 18ms
32
can't tset on my AMD you need a dual monitor system to test this. the GPU clock is totally unstable.

BTW. i still didn't get proper audio/stable playback with this player.

[Zachs]

Originally Posted by huhn

r9 270

640x480 -> 1440x1080
256
950/1400 49% 31ms (close to the limited)
128
950/1400 32% 22ms
64
950/1400 21% 18ms
32
can't tset on my AMD you need a dual monitor system to test this. the GPU clock is totally unstable.

BTW. i still didn't get proper audio/stable playback with this player.

Isn't your r9 270 an AMD card?
And what do you mean you can't get proper audio/stable playback? Does audio stutter or something?

[huhn]

Originally Posted by Zachs

Isn't your r9 270 an AMD card?

yes why?

And what do you mean you can't get proper audio/stable playback? Does audio stutter or something?

audio drops frame drops the audio buffer never fills up. it's extreme with a 23p source at 23p.

[Zachs]

Originally Posted by huhn

yes why?



audio drops frame drops the audio buffer never fills up. it's extreme with a 23p source at 23p.

You said you couldn't test your AMD card because it needed dual monitor?

Did you have the audio problem from the very first version or was it a recent release that caused this?

[huhn]

Originally Posted by Zachs

You said you couldn't test your AMD card because it needed dual monitor?

Originally Posted by huhn

64
950/1400 21% 18ms
32
can't tset on my AMD you need a dual monitor system to test this. the GPU clock is totally unstable.

i can't test 32 neuron and below. it did a real world fullscreen test. that's why you don't get numbers for this.

Did you have the audio problem from the very first version or was it a recent release that caused this?

maybe never really used it for real playback. i already reported this some time ago.

http://forum.doom9.org/showpost.php?...postcount=1274

the DWM problem is gone i'm not on windows 7 anymore.

[Zachs]

Originally Posted by huhn

i can't test 32 neuron and below. it did a real world fullscreen test. that's why you don't get numbers for this.

maybe never really used it for real playback. i already reported this some time ago.

http://forum.doom9.org/showpost.php?...postcount=1274

the DWM problem is gone i'm not on windows 7 anymore.

What CPU and sound card are you using?

And what do you mean it did a real world fullscreen test? Are you saying 32 neurons and below don't work at all?

[huhn]

Originally Posted by Zachs

What CPU and sound card are you using?

i3 4130
asus essence stx

And what do you mean it did a real world fullscreen test? Are you saying 32 neurons and below don't work at all?

i can't get proper numbers from GPU-Z it was working fine. after leaving fullscreen my GPU changes the powerstate a couple of times so the numbers are worthless.
nothing for you to bother simply no 32 neuron and lower numbers that's all. your program is not the problem here.

[Zachs]

Originally Posted by huhn

i3 4130
asus essence stx

i can't get proper numbers from GPU-Z it was working fine. after leaving fullscreen my GPU changes the powerstate a couple of times so the numbers are worthless.
nothing for you to bother simply no 32 neuron and lower numbers that's all. your program is not the problem here.

Ah understood.

Have you tried using MPC Audio Renderer with WASAPI instead just to see if it helps? Just trying to isolate the problem.

[huhn]

Originally Posted by Zachs

Ah understood.

Have you tried using MPC Audio Renderer with WASAPI instead just to see if it helps? Just trying to isolate the problem.

the mpc-BE audio renderer with WASAPI has kind of the same problem for me. no frames are dropped when using madVR but some audio drops/discontinues. i don't know if the audio buffer are not filled up with mpc-be wasapi there is no way to check.

i try AC3 filter.