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

[patul]

I get 164 fps for v11, AMD A6-5400K APU (HD 7540D). Haven't test on my laptop yet.

[Zachs]

Hi guys,

Thanks for all the help so far. I'd appreciate it if I could get you guys to run version 12 of the test app (http://www.zachsaw.com/downloads/mpdn_test/DxClTest12.zip) to see if it still runs and report back the new frame rate.

Thanks again!

Cheers.

[Anima123]

Test version 12 on Optimus (Intel HD 4600 + nVidia 880M), 249 fps, with previous version 9, the fps was 366.

[cyberbeing]

doesn't work with nvidia.

if i remember correctly someone has written a version for nvidia(was it cyberbeing? sorry i can't remember).
At the time, all I did compile a build with DirectCompute-only so I could see the results on my NVIDIA card. I never attempted to fix the OpenCL tests. It doesn't appear I ever shared the DirectCompute-only build back then, so here it is if anyone wanted it: https://www.mediafire.com/?u6x778w7odoaqdi

It is somewhat reassuring to see that NVIDIA has made some improvements to DirectCompute performance improvements since then though. A nice +65% increase (172fps->285fps) in DirectCompute Error Diffusion performance with 350.05 compared to the 321.10 driver I initially tested one year ago. Still quite poor compared to AMD cards though, considering even madshi's HD7770 could break 300fps in that DirectCompute test.

[Anime Viewer]

Hi guys,

Thanks for all the help so far. I'd appreciate it if I could get you guys to run version 12 of the test app (http://www.zachsaw.com/downloads/mpdn_test/DxClTest12.zip) to see if it still runs and report back the new frame rate.

Thanks again!

Cheers.

Version 12 gives 400-404 on my Optimus system.

[ryrynz]

Zach, you need to clear your inbox. :p

750 Ti 440 FPS build 12, we've dropped some performance.

[burfadel]

187.039 fps this time, AMD R9-280X.

[aufkrawall]

Gives me a performance gain with GTX 980, 1343fps.

[Zachs]

750 Ti 440 FPS build 12, we've dropped some performance.

On my GTX 560, it's gone up from 640 to 716fps. On the Intel, it's gone up from 315fps to 370fps.

[ryrynz]

On my GTX 560, it's gone up from 640 to 716fps. On the Intel, it's gone up from 315fps to 370fps.

*cough* 540 *cough*

[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]

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]

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 (https://github.com/zachsaw/MPDN_Extensions/blob/master/Examples/RenderScripts/Example.FrameRateHalver.cs). 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 (http://mpdn.zachsaw.com/Test%20Builds/)
Github nnedi3 branch (https://github.com/zachsaw/MPDN_Extensions/tree/nnedi3)

[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]

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]

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]

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]

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

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?p=1710018&postcount=1274

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

[Zachs]

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?p=1710018&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]

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]

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]

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.

[Zachs]

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.

The reason MPDN drops frames is because it's trying to sync to the audio clock. If the audio clock jumps around (which would be the case when audio skips / drops out), MPDN has to drop frames to stay in sync. It's just the way I've designed it. I believe the Asus sound card you're using is a very good card - I wasn't expecting it to have such problems at all.

[huhn]

The reason MPDN drops frames is because it's trying to sync to the audio clock. If the audio clock jumps around (which would be the case when audio skips / drops out), MPDN has to drop frames to stay in sync. It's just the way I've designed it. I believe the Asus sound card you're using is a very good card - I wasn't expecting it to have such problems at all.
but is works fine with EVR/madVR using direct audio with a 90 % buffer.

but i wouldn't be shock if the driver of this card are kind of "broken".
asus and creative doesn't really care about there driver.

the ac3 filter is load but not used i tried both 32 and 64 bit MPDN.

[Zachs]

but is works fine with EVR/madVR using direct audio with a 90 % buffer.

but i wouldn't be shock if the driver of this card are kind of "broken".
asus and creative doesn't really care about there driver.

the ac3 filter is load but not used i tried both 32 and 64 bit MPDN.

I don't think AC3 filter will do any good though - it's still using DirectSound in the end. What happens if you select WaveOut in MPDN's options dialog?

EDIT: I just checked my DirectSound audio filter props - it shows 30-40% fullness too but there's no breaks / discontinuities at all.

[Zachs]

256
950/1400 49% 31ms (close to the limited)


By the way, when you said it's close to the limit, do you mean it's close to video interval? It's only using 49% of your GPU though... Do you have a 720p clip you can test to see if you can get it to go over 90% GPU usage?

[huhn]

By the way, when you said it's close to the limit, do you mean it's close to video interval?
no video frame interval is 41.71. but you usually doesn't get 90% GPU usage an AMD and nnedi3
It's only using 49% of your GPU though... Do you have a 720p clip you can test to see if you can get it to go over 90% GPU usage?

this is from playing a 720p23 video at 1080 using 128 neurons.
http://abload.de/img/nnedi39cqcs.png
as you can see it's hard to judge the real GPU usages. rendertiems are 40.5 and sometimes frames are repeated or dropped.

and i can't test this for long my r9 270 is pretty hot 79 should be still fine but it is clocking is self sometimes to 900 hz from 925.

[Zachs]

no video frame interval is 41.71. but you usually doesn't get 90% GPU usage an AMD and nnedi3


this is from playing a 720p23 video at 1080 using 128 neurons.
http://abload.de/img/nnedi39cqcs.png
as you can see it's hard to judge the real GPU usages. rendertiems are 40.5 and sometimes frames are repeated or dropped.

and i can't test this for long my r9 270 is pretty hot 79 should be still fine but it is clocking is self sometimes to 900 hz from 925.

Why can't you get 90% GPU usage? This isn't using OpenCL at all. If you try 256 neurons, would that increase GPU usage?

[huhn]

Why can't you get 90% GPU usage? This isn't using OpenCL at all. If you try 256 neurons, would that increase GPU usage?

i get totally unreliable results. at the moment i got 128 running smoothly with 37-40 ms with an AVG GPU usage of ~70% at least for some mins. after some time it looks like the card is thermal throttling or something like that. looks like for some reason frames aren't rendered at all the render queue drop instantly to zero max render time jump to 120 ms.

with a GPU usage of 49-100% AVG 60-70% the GPU is at it's limited

[Zachs]

Yeah that sounds like thermal throttling to me too. How's it's performance vs MadVR?

[huhn]

Yeah that sounds like thermal throttling to me too.

i take that back. the GPU fan is at 60 % there is no way the GPU is thermal throttling without trying 100 % fan speed first. even through my r9 270 doesn't have a fan the GPU doesn't know that X-).
How's it's performance vs MadVR?

i set madVR to bicubic 75 for everything and disabled dither. i don't know what you use for downscaling i only know that your downscaling setting is not used for that.

720p23 -> 1080p23
64 neurons
madVR 25-26ms
MPDN 26-28ms
i got 23ms at first but this changes after some time to 26ms+

128 neurons
madVR 46ms +
MPDN 40-41 ms still not smooth at all after some mins

i would call that pretty much even.


and in this test i got frame drop problems again using MPDN the decoder queue is dropping to 0. but i can't reproduce this reliable... and i didn't used my headphones so not sure if it was the audio issue again. so these numbers aren't 100% reliable.

[Anime Viewer]

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

* 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: 48%/600Mhz, 32: 39%/600, 64: 54%/600Mhz, 128: 74%/850Mhz, 256: 74%/1200Mhz
3) 320x180
4) 1280x720


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)

Links (not compatible with MPDN v2.25.3):


I know you're mainly looking for AMD tests, but here is the results of my Nvidia Optimus test with the AnyCPU version:

specs:
1) Nvidia 680m GTX
2) 16: 44%/718.5Mhz(GPU core clock)900mhz(memory clock), 32: 63%/718.5(gpu)/900(mem), 64 and higher neurons result in dropped frames, etc.
3) 1280x720
4) 1920x1080
16=21-23ms
32=29-30ms
64+=N/A

1) Nvidia 680m GTX
2) 16: 24%/718.5Mhz(GPU core clock)900mhz(memory clock), 32: 32%/718.5(gpu)/900(mem), 64: 52%/718.5(gpu)/900(mem) 128: and higher neurons result in dropped frames, etc.
3) 848x480
4) 1920x1080
16=12ms
32=15-16ms
64=23-24ms
128+=N/A


GPUZ 0.78 used to get clocks and % data.

Not surprisingly my Intel HD 4000 GPU can't handle NNEDI3 in MPDN in any setting.

Things that aren't factored in - in the reported stats that I think could influence results:
Video source frame rate: 32.810Hz is probably going to be less taxing than say 60Hz videos.
Screen refresh rate and if it is matching the video Hz (or close to it), or if its different (ex: 32hz video and 60hz monitor/tv).
Video Renderer: General: Prefer Image Quality/Performance choice as well as FSE and queue settings.
Other programs/windows that may be running at the same time as the video/MPDN.

Like with madVR I don't see any noticeably benefit (visually) from using anything above 16 neurons.

[Zachs]

I have probably worded that post poorly but I'd like everyone to test it, not just AMD users.

EDIT: Thanks Anime Viewer! Forgot about frame rate, render quality etc. Will update the post.

[aufkrawall]

Will test tomorrow.
Btw: There's also most certainly a bug in the AMD drivers regarding reported GPU usage, at least with Hawaii GPUs: It shows inconsistent usage while in fact there should be permanent one.

[Anima123]

Well, the nnedi3 32-neurons works fine with 880m, rendering time is stable at around 21ms, when playing 576p -> 1080p, 23.976fps.

Edit1,
Actually, I am dreaming what's like with the combination of nnedi3 + SuperRes, hope there's no rendering time increasing issue.


Edit2,
Just test with 64 neurons, the rendering time around 29ms, not being able to fluent playing back with 128 though.

Maybe I should tried madVR's version with all other effects off to see if it works as fast.

[Scyna]

1) Amd 290
2) 16: 50%/553 32: 80%/601 64: 90%/603 128: 100%/722 256: 100%/914
3) 1280x720
5) Max Performance

2) 16: 74%/540 32: 64%/598 64: 90%/634 128: 100%/725 256: 100%/923
5) Image Quality

2) 16: 52%/464 32: 51%/500 64: 72%/491 128: 80%/541 256: 81%/623
3) 720x392
5) Max Performance

2) 16: 40%/470 32: 54%/501 64: 57%/511 128: 64/550 256: 73%/620
5) Image Quality

Strange after finishing this I can't seem to get my subtitles to work even though I got it checked on. I never bother to check if subtitles worked on this version

[Zachs]

OK guys, I've made another version to see if it's faster.
You'll need to overwrite your .cso files (in the Extensions\RenderScripts\NNEDI3 folder) with these ones (test2) (http://www.zachsaw.com/downloads/MediaPlayerDotNet/nnedi3_test2.zip).

I'd like to know if it's faster or slower for you at the various nnedi3 neuron settings.

While you're at that, give test3 (http://www.zachsaw.com/downloads/MediaPlayerDotNet/nnedi3_test3.zip) and test4 (http://www.zachsaw.com/downloads/MediaPlayerDotNet/nnedi3_test4.zip) a try too! :)

Thanks everyone!

[Anima123]

original 64-neurons, rendering time is 29.50 ms or so.
test2 better with 64-neurons on nVidia 880m Optimus, 27.xx ms, 576p -> 1080p, 23.976 ftp
test3 27.60 ms or so, not as fast as test 2
test4 faster than test3, not as fast as test 2

Conclusion: test 2 is the fastest for me with 64-neurons, at least for 576p -> 1080p.

Edit: Just had a quick test with test 2, 32-neurons, which behave similar to 64-neurons, faster than the original version.

[Zachs]

Can you try comparing 128 and 256 neurons too? Doesn't matter if it drop frames. I just want to know the render times.

[Anima123]

For 128
Original: 62.60 ms
test2: 68.xx ms
test3: 67.xx ms
test4: N/A

The original works faster than others.

Edit: OK, I admit, I just can't resist, here's my test results for 256-neurons:

Original: 118.50 ms
test2: 119.90 ms
test3: 116.90 ms
test4: N/A

[Scyna]

do u want the same information as test 1? people are mentioning render times should i include those too? btw my subtitles work again on 2.25.3.

[Zachs]

I've uploaded MPDN build 3066 to the test builds folder. It fixes a problem in 3065 where scaling luma wasn't skipped even when render script told it to, so it was unnecessarily scaling luma with MPDN's internal scaler even when it's already being scaled by nnedi3. So it should be make things a little faster.

I've also updated the nnedi3 script on GitHub - it now has an extra option called "Path" which presents you with 3 options: Prefer Scaler, Prefer Vector or Avoid Branches. Each GPU+driver works differently (as I've gathered from the tests you have done so far) so try out each path for each neuron setting, you'll find one that's fastest.

Same as before, I'd like feedback to be in the following format.

1) GPU name
2) 16/32/64/128/256 Neurons - Selected Path (Scalar/Vector/NoBranch) - GPU load / clock speed when it's running + render time (optional)
3) Source clip resolution / frame rate
4) Target resolution (must be bigger than source clip res)
5) Render quality settings (preferably "Image Quality" which is the MPDN default)

* 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/Scalar: 41%/600Mhz, 32/Scalar: 34%/600, 64/Scalar: 47%/600Mhz, 128/Scalar: 72%/800Mhz, 256/Scalar: 71%/1150Mhz
3) 320x180 23p
4) 1280x720
5) Max Performance

2) 16/Scalar: 42%/600Mhz, 32/Scalar: 35%/600, 64/Scalar: 49%/600Mhz, 128/Scalar: 74%/800Mhz, 256/Scalar: 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 = 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 v2.25.4.3066 (http://mpdn.zachsaw.com/Test%20Builds/3066)
Github nnedi3 branch (https://github.com/zachsaw/MPDN_Extensions/tree/nnedi3)