Media Player .NET (MPDN) - D3D HQ GPU Video Renderer [v2.49.0/v1.31.0 27 Dec 2018] - Page 129

Dark Eiri · 29th June 2015, 18:39

Is there any way to support GPU acceleration via integrated GPUs like I do with MPC and LAV? On MPC I set LAV to use QuickSync instead of CUDA, that makes my GTX 770 sleepy and silent at 135 MHz Core and uses my HD4600, which has no monitor output (I believe you guys call that "headless GPU"?), to decode everything, . Any way to do that with MPDN so the GTX770 won't go full clock and noisy every time I boot it up? The iGPU doesn't even show up on the Adapters list.

huhn · 29th June 2015, 18:50

the adapter is chosen for presentation not for decoding. you can just choice quicksync under lavfilter and MPDN will use it.

Belphemur · 29th June 2015, 20:54

I did think of file hashing instead of Timestamp comparison (as how the uninstaller is done for now). Until now I couldn't find a plugin/script for NSIS to generate hash. Now I found a basic CRC32 that should be enough to check for modification.

I can later on make an installer for the source version, but I think this would get a little complicate for the user (and also could lead to some conflict when the framework get changed/refactored).

Anyway I'm open to any ideas !

huhn · 29th June 2015, 22:38

Quote:

Originally Posted by huhn

i get high numbers in madVR and foobar too (~4000). the only difference is that MPDN has audio issues and the rest not. so the driver are kinda broken or this software doesn't work properly on windows 10.

not sure if it is good to waste time on it.

i disabled the asus sound card and used the AMD GPU as a sound card.

i saw again high interrupt to process values so the asus soundcard isn't the problem. but the real issue is that the values are way way higher with MPDN.

and this is on a completely reinstalled system.

the next thing i will try is switch out the GPU and have a look at the results.

Magik Mark · 30th June 2015, 01:26

Zachs how do we activate the dxvahd render?

Zachs · 30th June 2015, 04:08

I've just pushed the DXVA HD Scaler render script to github. This is unreleased yet but I'll be making a new MPDN_Extensions release shortly.

For more info, see https://github.com/zachsaw/MPDN_Exte...DXVA-HD-Scaler

Anyway, not sure why anyone would want to use it, but you're the boss!

Zachs · 30th June 2015, 04:09

Quote:

Originally Posted by huhn

i disabled the asus sound card and used the AMD GPU as a sound card.

i saw again high interrupt to process values so the asus soundcard isn't the problem. but the real issue is that the values are way way higher with MPDN.

and this is on a completely reinstalled system.

the next thing i will try is switch out the GPU and have a look at the results.

It sounds like your AMD driver is causing high DPC latency and MPDN may have uncovered a bug in their driver. Is it a beta version?

Dark Eiri · 30th June 2015, 05:16

Quote:

Originally Posted by huhn

the adapter is chosen for presentation not for decoding. you can just choice quicksync under lavfilter and MPDN will use it.

It is. And still makes my GTX 770 spike all the way to 1045 MHz.

Zachs · 30th June 2015, 05:25

Quote:

Originally Posted by Dark Eiri

It is. And still makes my GTX 770 spike all the way to 1045 MHz.

Do you see only one adapter on your drop down list? You should see 2. On my Win7 machine, I have the iGPU setup as a headless output and I can switch between the two easily. Are you sure MPC even supports using the headless iGPU to render your video? I certainly can't find an option to allow that.

Magik Mark · 30th June 2015, 06:07

Quote:

Originally Posted by Zachs

I've just pushed the DXVA HD Scaler render script to github. This is unreleased yet but I'll be making a new MPDN_Extensions release shortly.

For more info, see https://github.com/zachsaw/MPDN_Exte...DXVA-HD-Scaler

Anyway, not sure why anyone would want to use it, but you're the boss!

Thanks you! I'm now able to use file sharp & adaptive sharpen wihout any problem in GTX 550ti. Most of the videos I play are 8bit including blue ray

I have activated debanding and hopefully it would address any banding issues

Zachs · 30th June 2015, 06:12

Quote:

Originally Posted by Magik Mark

Thanks you! I'm now able to use file sharp & adaptive sharpen wihout any problem in GTX 550ti. Most of the videos I play are 8bit including blue ray

I have activated debanding and hopefully it would address any banding issues

On GTX 550ti, you should just use hardware bilinear for better quality. DXVA HD will actually produce worse quality and is slower at the same time!

mrcorbo · 30th June 2015, 06:23

Quote:

Originally Posted by Belphemur

I did think of file hashing instead of Timestamp comparison (as how the uninstaller is done for now). Until now I couldn't find a plugin/script for NSIS to generate hash. Now I found a basic CRC32 that should be enough to check for modification.

I can later on make an installer for the source version, but I think this would get a little complicate for the user (and also could lead to some conflict when the framework get changed/refactored).

Anyway I'm open to any ideas !

When I have some time I'll try to put together a more complete outline of how I think this might work, but mainly it centers on keeping a manifest of all of the extension files with a hash value for each file, the date modified file property as read at the time the current hash value was generated to detect when a file needs to be re-hashed and an additional property to allow users to mark certain extension files as manual update only. This last basically covering cases where the user has modified an extension and doesn't want the updater nagging them about it.

Magik Mark · 30th June 2015, 06:28

Quote:

Originally Posted by Zachs

On GTX 550ti, you should just use hardware bilinear for better quality. DXVA HD will actually produce worse quality and is slower at the same time!

How do I activate hardware bilinear? I could only see bilinear in chroma/luma upscaling/downscaling. It is not found in image scaling

Zachs · 30th June 2015, 06:46

It's under "Scaling performed by GPU hardware" --> Bilinear.

nevcairiel · 30th June 2015, 07:16

Does AMD implement DXVA HD now? A few years back when I investigated that API, they didn't offer DXVA HD at all.

Zachs · 30th June 2015, 07:24

Well it doesn't work on my old 4350 and I haven't bought an AMD product since so I'm not sure really.

EDIT: Anyway, the DXVA HD scaler seems to only benefit Intel GPUs. They can do Bicubic 100 no AR with a bit less power than MPDN (by exactly 0.2W on both up and downscaling on my Intel HD 3000). Even Nvidia is simply using bilinear when scaling and is identical to MPDN's hardware bilinear.

huhn · 30th June 2015, 09:20

madVr had problems using the better DXVA scaler too.

intel should be like lanczos 3 ar (8 bit) and amd and nvidia should be better than bilinear too even through amd is bilinear in madVR for me and something better in EVR. but i'm not even sure if you try to use the same API and i don't think they are useful too.

my GPU driver are software suite 15.20-150522a-184226e. the default WDDM 2.0 driver. these are not the 15.20 driver that promise a huge performance boost. the audio problem was older than preview windows 10 so not sure if a driver can be the reason.

huhn · 30th June 2015, 09:37

Quote:

Originally Posted by Dark Eiri

It is. And still makes my GTX 770 spike all the way to 1045 MHz.

maybe your MPDN settings force your GPU to do this for rendering. quicksync headless mode is about decoding only and decoding with quicksync and presentation with EVR has title to no effect on the GPU unlike most better scaler in MPDN.

Quote:

Originally Posted by Zachs

Do you see only one adapter on your drop down list? You should see 2. On my Win7 machine, I have the iGPU setup as a headless output and I can switch between the two easily. Are you sure MPC even supports using the headless iGPU to render your video? I certainly can't find an option to allow that.

intel quicksync headless mode is window 8+ only and is only about quicksync decoding. on windows 7 you need a fake monitor to access intel quicksync without putting an display to the GPU so this is no true headless mode.
at least he was talking about decoding not rendering. in headless mode the video is still rendered by the other GPU but EVR doesn't have an impact on them.

Zachs · 2nd July 2015, 02:49

Quote:

Originally Posted by huhn

intel quicksync headless mode is window 8+ only and is only about quicksync decoding. on windows 7 you need a fake monitor to access intel quicksync without putting an display to the GPU so this is no true headless mode.
at least he was talking about decoding not rendering. in headless mode the video is still rendered by the other GPU but EVR doesn't have an impact on them.

Headless mode isn't exactly a technical term - it simply means a display adapter's output that isn't connected to any displays. In both Windows 7 and 8, you can do what you call "fake monitor". There's no need to do that for Win8 if you're just trying to use QuickSync. However, for rendering (and I assume he wanted his iGPU to do the rendering since he asked if it's possible to get his NV card to stay at lowest clock speed) you need to still "fake" a monitor.

It would make more sense if Microsoft simply allowed you to choose an adapter that isn't connected to a display to do the rendering. That's simply how Optimus works anyway.

huhn · 2nd July 2015, 10:42

the comparison was with MPC-HC and there was no word about rendering what so ever.

in the end very high settings are needed to get a 770 gtx to high power state and the intel wouldn't be able to handle it.

do you really think it is a good idea to render frames on a GPU that isn't presentation them.

just look at optimus system and there high present times.

and 8bit UHD@60 would be about 1.4 gigabytes per sec that needs to be transferred to the presentation GPU.

30th June 2015, 04:08	#2566 \| Link
Zachs Suptitle, MediaPlayer.NET Join Date: Nov 2001 Posts: 1,721	DXVA HD Scaler I've just pushed the DXVA HD Scaler render script to github. This is unreleased yet but I'll be making a new MPDN_Extensions release shortly. For more info, see https://github.com/zachsaw/MPDN_Exte...DXVA-HD-Scaler Anyway, not sure why anyone would want to use it, but you're the boss!

30th June 2015, 07:16	#2575 \| Link
nevcairiel Registered Developer Join Date: Mar 2010 Location: Hamburg/Germany Posts: 10,407	Does AMD implement DXVA HD now? A few years back when I investigated that API, they didn't offer DXVA HD at all. __________________ LAV Filters - open source ffmpeg based media splitter and decoders

30th June 2015, 07:24	#2576 \| Link
Zachs Suptitle, MediaPlayer.NET Join Date: Nov 2001 Posts: 1,721	Well it doesn't work on my old 4350 and I haven't bought an AMD product since so I'm not sure really. EDIT: Anyway, the DXVA HD scaler seems to only benefit Intel GPUs. They can do Bicubic 100 no AR with a bit less power than MPDN (by exactly 0.2W on both up and downscaling on my Intel HD 3000). Even Nvidia is simply using bilinear when scaling and is identical to MPDN's hardware bilinear. Last edited by Zachs; 30th June 2015 at 07:28.

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

29th June 2015, 18:39	#2561 \| Link
Dark Eiri Registered User Join Date: May 2006 Posts: 335	Is there any way to support GPU acceleration via integrated GPUs like I do with MPC and LAV? On MPC I set LAV to use QuickSync instead of CUDA, that makes my GTX 770 sleepy and silent at 135 MHz Core and uses my HD4600, which has no monitor output (I believe you guys call that "headless GPU"?), to decode everything, . Any way to do that with MPDN so the GTX770 won't go full clock and noisy every time I boot it up? The iGPU doesn't even show up on the Adapters list.

29th June 2015, 18:50	#2562 \| Link
huhn Registered User Join Date: Oct 2012 Posts: 8,623	the adapter is chosen for presentation not for decoding. you can just choice quicksync under lavfilter and MPDN will use it.

29th June 2015, 20:54	#2563 \| Link
Belphemur MPDN Extensions Dev Join Date: Apr 2015 Posts: 52	I did think of file hashing instead of Timestamp comparison (as how the uninstaller is done for now). Until now I couldn't find a plugin/script for NSIS to generate hash. Now I found a basic CRC32 that should be enough to check for modification. I can later on make an installer for the source version, but I think this would get a little complicate for the user (and also could lead to some conflict when the framework get changed/refactored). Anyway I'm open to any ideas !

30th June 2015, 01:26	#2565 \| Link
Magik Mark Registered User Join Date: Dec 2014 Posts: 667	Zachs how do we activate the dxvahd render?

30th June 2015, 06:46	#2574 \| Link
Zachs Suptitle, MediaPlayer.NET Join Date: Nov 2001 Posts: 1,721	It's under "Scaling performed by GPU hardware" --> Bilinear.

30th June 2015, 09:20	#2577 \| Link
huhn Registered User Join Date: Oct 2012 Posts: 8,623	madVr had problems using the better DXVA scaler too. intel should be like lanczos 3 ar (8 bit) and amd and nvidia should be better than bilinear too even through amd is bilinear in madVR for me and something better in EVR. but i'm not even sure if you try to use the same API and i don't think they are useful too. my GPU driver are software suite 15.20-150522a-184226e. the default WDDM 2.0 driver. these are not the 15.20 driver that promise a huge performance boost. the audio problem was older than preview windows 10 so not sure if a driver can be the reason.

2nd July 2015, 10:42	#2580 \| Link
huhn Registered User Join Date: Oct 2012 Posts: 8,623	the comparison was with MPC-HC and there was no word about rendering what so ever. in the end very high settings are needed to get a 770 gtx to high power state and the intel wouldn't be able to handle it. do you really think it is a good idea to render frames on a GPU that isn't presentation them. just look at optimus system and there high present times. and 8bit UHD@60 would be about 1.4 gigabytes per sec that needs to be transferred to the presentation GPU.