madVR - high quality video renderer (GPU assisted) [Archive] - Page 647

metadethrax

31st July 2015, 07:49

@metadethrax

you have to create a 23,976 custom resolution within the nvidia drivers to get as close as possible to 23,976.

I created a custom resolution as you instructed me, now the repeated frames times went up from 40 seconds to 16 minutes, so it's definitely a progress.

movies are usually 24000/1001 hz (it is called 23p) not 24 hz, you should use 23p not 24p for this file in question.

http://forum.doom9.org/showpost.php?p=1731949&postcount=32210

wasn't he clear enough?

Would you, please, be so kind to detail a little? Where should I input 23p instead of 24p? Thanks!

Sunset1982

31st July 2015, 09:13

I created a custom resolution as you instructed me, now the repeated frames times went up from 40 seconds to 16 minutes, so it's definitely a progress.

you have to try different settings, there is a option where you can set the pixels. I will search for a how to or a screenshot for you later...

ashlar42

31st July 2015, 11:49

That results in terrible aliasing. Basically you are wasting the entire potential of a 4K screen, might as well buy a 1080p screen then.

Good scaling algorithms will provide a much better image.You are right, I tried nearest neighbor to upscale 720p to 1440p (same logic as 1080p to 2160p aka 4k)... terrible aliasing.
So we add information and a good algorithm makes sure that it's *good* information.

Thanks.

ashlar42

31st July 2015, 11:51

I created a custom resolution as you instructed me, now the repeated frames times went up from 40 seconds to 16 minutes, so it's definitely a progress.

Would you, please, be so kind to detail a little? Where should I input 23p instead of 24p? Thanks!If you are on Nvidia, you might want to check this: http://www.hometheatershack.com/forums/computers-htpc-media-servers/62832-cant-hit-23-976-hz-nvidia-gts-450-panasonic-ae4000u-projector.html#post560082

Following this I managed to get to 23.97589 and/or 23.97599 which is really close to what we aim for.

Sunset1982

31st July 2015, 12:10

If you are on Nvidia, you might want to check this: http://www.hometheatershack.com/forums/computers-htpc-media-servers/62832-cant-hit-23-976-hz-nvidia-gts-450-panasonic-ae4000u-projector.html#post560082

Following this I managed to get to 23.97589 and/or 23.97599 which is really close to what we aim for.

yeah thats what i meant and what i was searching for.

huhn

31st July 2015, 12:30

I created a custom resolution as you instructed me, now the repeated frames times went up from 40 seconds to 16 minutes, so it's definitely a progress.

Would you, please, be so kind to detail a little? Where should I input 23p instead of 24p? Thanks!

use 23hz not 24hz.

a custom refreshrate overrides the 24hz so you can used that for 24 hz movies any more.

huhn

31st July 2015, 12:38

I always see 4k TVs in the brochures...

that doesn't change that most people didn't have one and that this will not change for some time.

don't get me wrong I order one my self month ago and stopped test madVR and wasn't using my HTPC any more but i order this TV at hardwareversand so i'm forced to wait a "bit" longer. the good news i don't need a 960 any time soon.

but most important AMD cards are ridiculously strong for openCL. nnedi3 is the only madVR feature for openCL now but that may change in the future. so openCL maybe has use for something else so removing it doesn't really help.

x7007

31st July 2015, 13:52

Any fix for the NNEDI3 with Nvidia card ? I can't enable it no matter what DXVA Write Back , Native , CUVID .

it always revert to JINC on upscaling

huhn

31st July 2015, 13:55

install 353.49.

aufkrawall

31st July 2015, 14:07

that doesn't change that most people didn't have one and that this will not change for some time.

Most people don't use a HTPC for madVR with their 4k TV. ;)

but most important AMD cards are ridiculously strong for openCL.
I don't think it has necessarily to do with OpenCL but with the kind of calculation that is executed.
In Anandtechs compute tests (OpenCL only), Fury X is sometimes much slower than Titan X (and vice versa):
http://www.anandtech.com/show/9390/the-amd-radeon-r9-fury-x-review/24
I don't think GCN looks great here, since it's just 6.2 TFlops for GM200 vs. 8.6TFlops for Fiji.

x7007

31st July 2015, 15:41

install 353.49.

It's impossible to play with this drivers because of Mouse issue nvidia fixed on the latter release 353.50+

aufkrawall

31st July 2015, 15:58

What kind of mouse issue?

nevcairiel

31st July 2015, 16:23

It's impossible to play with this drivers because of Mouse issue nvidia fixed on the latter release 353.50+

Then install Windows 8.1, it solves all your troubles! :)

aufkrawall

31st July 2015, 16:33

I'm not aware of any mouse issue with 353.49, nothing like that showed up here with GTX 980.
There was a mouse cursor issue with drivers post 353.49, but it's fixed in 353.62 (which has the OpenCL issue, of course).

I still wonder if NNEDI3 OpenCL is also broken with AMD in general on Windows 10 with most current drivers (15.7.1).

Ver Greeneyes

31st July 2015, 16:34

I'm holding off on Windows 10 until this gets resolved. I don't need it that badly. The questionable opt-out systems (sharing Wifi with all contacts, P2P updates..) are also making me less than enthusiastic about upgrading.

Edit: Apparently the Wifi sharing is opt-in per network, which isn't so bad.

aufkrawall

31st July 2015, 16:39

XMonarchY

31st July 2015, 17:20

You can basically disable everything privacy-related.
Only thing that can't be disabled is feedback about installed programs, for this you need Enterprise edition (or dirty hacks).

Edit: Is the extension still there on Windows 8.1 with 353.62?

You can disable them too, but you do need to make a LOT of changes. Check out this guide I copied from a forum (NOT my guide) - https://dl.dropboxusercontent.com/u/93831976/The%20Guide.pdf

It has 90% of what everyone needs to do after Windows 10 installation to get it REAL fast, smooth, compatible, and with Feedback disabled for good!

I do not want to provide a link to the actual guide (on a forum) because I do not want to violate Doom9 forum rules, which are strict regarding certain aspects.

XMonarchY

31st July 2015, 17:21

I'm holding off on Windows 10 until this gets resolved. I don't need it that badly. The questionable opt-out systems (sharing Wifi with all contacts, P2P updates..) are also making me less than enthusiastic about upgrading.

If AMD also has D3D9 OpenCL interop messed up, then hopefully it can be re-worked using other ways. I am quite certain madshi would never let red pandas die on his watch!

noee

31st July 2015, 20:42

I still wonder if NNEDI3 OpenCL is also broken with AMD in general on Windows 10 with most current drivers (15.7.1).

Working here, watching right now...

Win10Pro, 15.7.1, NNEDI32, 720x352 encode => 1080P HDTV

aufkrawall

31st July 2015, 22:01

If the GPU of your signature applies, this doesn't have to mean much. That extension has not been removed with the newest Geforce drivers for Fermi GPUs either.

XRyche

31st July 2015, 22:29

Working here, watching right now...

Win10Pro, 15.7.1, NNEDI32, 720x352 encode => 1080P HDTV

Did you do the recent upgrade or are you one of the ones that have been using Windows 10 for a few months already?

noee

31st July 2015, 22:40

Did you do the recent upgrade or are you one of the ones that have been using Windows 10 for a few months already?

I just upgraded the other day on this tester machine, it's a madVR playback machine and also Linux dev box (dual-boot w/openSUSE). Went from Win7Ult to Win10Pro using the ISO. I had to reinstall jRiver and there is no NFS client, which is not a biggee.

I notice that FSE/Windowed mode switch in madVR is faster, but in jRiver, it seems to switch a few times unnecessarily compared to MPC-HC, not sure really.

I did use DDU to remove the old Win7 AMD driver and then installed 15.7.1 after the Win10 install was completed. No issues.

Only thing so far is no NFS client for Windows... ;(

huhn

31st July 2015, 22:56

If the GPU of your signature applies, this doesn't have to mean much. That extension has not been removed with the newest Geforce drivers for Fermi GPUs either.

r9 270 with 15.7.1 has working nnedi3 and intel has the extension too in there windows 10 driver.

so directx 9 <-> interop simply works under windows 10.
i don't know why they removed it but windows 10 supports directx 9 <-> openCL interop if they like or not.

zvans18

1st August 2015, 03:24

extremely minor, but can a scroll bar be added to the profile auto select rules section? i have quite a few and i have to use the arrow keys to navigate instead of mouse wheel

StinDaWg

1st August 2015, 08:34

Thank you for the in-depth sharpness comparisons everyone.
I had a look over everything posted over the last day or two, and did some testing of my own.

My concern is that most of the test images that people are using as a reference are already over-sharpened, containing pre-existing ringing artifacts and aliasing.
So in order to match the source image, you end up using overly sharp post-processing, which seems to be introducing a lot of artifacts around high-contrast edges.

I would focus more on good Blu-ray sourced images downscaled 50% and trying to match the original sharpness, than using photographs or content from what appears to be low-quality or re-compressed sources, if the goal is restoration rather than enhancement.
I do also wonder if people are checking the results on small PC monitors, or their tv/projection setups.
In what testing I have been able to do, SuperRes seems to be introducing aliaising and particularly ugly high-frequency ringing at all strengths.
Even with a strength of 1 and radius of 1.00 it's still adding aliasing - whether that is real-world images, or test patterns.
Of course if you're combining that with a scaler that already suffers from aliasing (basically anything other than Jinc or image doubling) perhaps the additional aliasing is not that noticeable?

I finally got around to trying out Super-XBR too, which seems to add a lot of low-frequency ringing with sharpness set above 25 - which you may not notice if you're looking at a poor quality source that already has ringing artifacts, rather than a good Blu-ray disc.
Combining Super-XBR with sharpening to achieve similar results to 16-neuron NNEDI3 ended up with higher render times and worse image quality than just using NNEDI3.

Here's one of the images that I was testing with which seemed to show aliasing quite well:
Source (http://abload.de/img/mononoke-source0lo7u.png)
Downscaled (http://abload.de/img/mononoke-ll1er51.png)
I'm not really happy with how that looks using any of the sharpening options, or scaling other than NNEDI3.
But perhaps I haven't spent enough time with the new options, I really haven't been able to do much testing with anything that's been introduced in the last couple of months at all.

I have to agree with this post. Here is an example from a 16 Mbps 720p MPEG2 ESPN recording. It was upscaled to 1080p with B75AR chroma+Lanczos3AR image+NNEDI3 32 neurons doubling+CR AR LL downscale.

Original image @ 720p | No sharpen | SuperRes 1.0 | SuperRes 4.0
http://t1.someimage.com/hRxNc44.jpg (http://someimage.com/hRxNc44) http://t1.someimage.com/jPNsJGf.jpg (http://someimage.com/jPNsJGf) http://t1.someimage.com/3dCTtzx.jpg (http://someimage.com/3dCTtzx) http://t1.someimage.com/yWWWBFc.jpg (http://someimage.com/yWWWBFc)

Look at the text. No sharpening looks best to me. SuperRes at 1.0 strength starts to alias. 4.0 looks downright comical.

I'm using the latest version of madVR.

MSL_DK

1st August 2015, 08:58

Ok . It's Direct3D 11 that's causing the problem. Newest driver from AMD on Windows 10 PRO x64 :(

No problem in windowed mode.

EDIT: Have to mention, no problem with win 7 on the same machine.

Same issue here on Win10Pro/CCC15.7.1, but I think it's cosmetic because I've watched 2 complete movies (almost five hours) and had only one dropped frame in each and that was at playback startup.

fwiw, in Windowed mode, it would drop a frame about every 3 minutes even though queue stats were "full" reported.

I don't think so. Look what happens when i move the mouse pointer to the bottom of the screen (https://www.dropbox.com/s/ydrp4yjhmxiwv4a/mvr.mkv?dl=0)

x7007

1st August 2015, 11:12

So everyone posted on nvidia forum and nvidia bug site about the 353.50+ driver that doesn't work with NNEDI3 . and the 353.49 not working with CUVID ?

I hope next driver they will fix it.

Soul

1st August 2015, 13:41

Hi, Is there any option to stop madvr from showing subtitle tracks? I want to set the default to "No subtitles" but it always pick English subtitle as the default and force it during playback.

Thanks

sneaker_ger

1st August 2015, 13:51

There is no madvr option for subtitle track selection. You have to do this in your player or splitter.

XRyche

1st August 2015, 15:22

I just upgraded the other day on this tester machine, it's a madVR playback machine and also Linux dev box (dual-boot w/openSUSE). Went from Win7Ult to Win10Pro using the ISO. I had to reinstall jRiver and there is no NFS client, which is not a biggee.

I notice that FSE/Windowed mode switch in madVR is faster, but in jRiver, it seems to switch a few times unnecessarily compared to MPC-HC, not sure really.

I did use DDU to remove the old Win7 AMD driver and then installed 15.7.1 after the Win10 install was completed. No issues.

Only thing so far is no NFS client for Windows... ;(

I'm going to try to delete my madVR directory and start from scratch later on. Let's see if that fixes my Nnedi3 issues and exceedingly long delayed playback on Windows 10.

XMonarchY

1st August 2015, 15:27

So everyone posted on nvidia forum and nvidia bug site about the 353.50+ driver that doesn't work with NNEDI3 . and the 353.49 not working with CUVID ?

I hope next driver they will fix it.

Yeah. CUVID is far less important than NNEDI3. CUVID puts your GPU in highest power state, which is unnecessary and only forces more GPU cycles. In addition, CUVID causes artifacts in some cases. It does improve image quality in any way.

Simply select DXVA2 (Copy-Back) in LAV Video decoder/filter and you'll get better performance and no artifacts.

sneaker_ger

1st August 2015, 15:28

I'm going to try to delete my madVR directory and start from scratch later on. Let's see if that fixes my Nnedi3 issues and exceedingly long delayed playback on Windows 10.
If you want a total reset you also need to delete the madvr keys from the registry. (Running "restore default settings.bat" should do that for you)

x7007

1st August 2015, 17:05

Yeah. CUVID is far less important than NNEDI3. CUVID puts your GPU in highest power state, which is unnecessary and only forces more GPU cycles. In addition, CUVID causes artifacts in some cases. It does improve image quality in any way.

Simply select DXVA2 (Copy-Back) in LAV Video decoder/filter and you'll get better performance and no artifacts.

Ye but stupid windows 10 always auto update the driver and it doesn't let me hide it.

Ver Greeneyes

1st August 2015, 17:29

Ye but stupid windows 10 always auto update the driver and it doesn't let me hide it.See How to temporarily prevent a Windows or driver update from reinstalling in Windows 10 (https://support.microsoft.com/en-us/kb/3073930).

This thread is really getting dominated by Windows 10 talk though. I think there's probably enough for madshi to work through already, so let's try to get back on track with upscaling refinement feedback.

XRyche

1st August 2015, 17:51

If you want a total reset you also need to delete the madvr keys from the registry. (Running "restore default settings.bat" should do that for you)

Thanks for reminding me. :thanks:

XRyche

1st August 2015, 18:05

Well that didn't change anything. Still getting inordinately long wait times for an image to show up. Still getting a brownish outline (almost like the emboss shader from MPC-HC) of everything in a video when in FS/FSE when image doubling with Nnedi3. When I take a screenshot using F5, the screenshot looks perfectly normal :shrugs: .

Pantara

1st August 2015, 18:52

@madshi

Hi,
I might have found a bug in madVR.I am using madVR v0.88.21 in combination with the 64-bit RC of MPC-HC+CCCP. Since some time (I first noticed it with v0.88.14) the video often pauses when you leave the fullscreen mode. I already checked the settings of madVR, as well as MPC but wasn't able to find a checkbox or anything similar (The option "delay playback start until render queue is full" is also inactive). The problem didn't occur with older versions of madVR (and the same version of MPC). It was already there in Windows 8.1 and now continues to occur since I upgraded to Windows 10. It also doesn't happen if you select a renderer other than madVR.
This is my setup:

MPC-HC (CCCP Build, 64-bit)
---------------------------

Build information:
Version: 1.7.8.162 (d12a1c4) (blitzkrieg) (master@46ce9d6)
Compiler: MSVC 2013 Update 4
Build date: May 4 2015 00:21:53

LAV Filters:
LAV Splitter: 0.65.0.8
LAV Video: 0.65.0.8
LAV Audio: 0.65.0.8
FFmpeg compiler: MinGW-w64 GCC 4.9.2

Operating system:
Name: Windows NT 10.0 (build 10240)
Version: 10.0 (64-bit)

Hardware:
CPU: Intel(R) Core(TM) i5-2430M CPU @ 2.40GHz
GPU: NVIDIA GeForce GT 540M (driver version: 10.18.13.5362)

EDIT:
I turned on debug OSD and it's always printing "Resetting Direct3D device failed (8876086c)" when you leave the fullscreen mode and the video stops.
There's also sometimes a "dxva proccessing failed" message (when you open up a file, etc.)

omarank

1st August 2015, 19:56

My render, upload and present queue doesn't get filled (1-3/16) when using 1080p60 no problem if i choose 1080p23 (FSE) in win 10 using AMD card no problem in windowed mode.

Same issue here on Win10Pro/CCC15.7.1, but I think it's cosmetic because I've watched 2 complete movies (almost five hours) and had only one dropped frame in each and that was at playback startup.

fwiw, in Windowed mode, it would drop a frame about every 3 minutes even though queue stats were "full" reported.
I have the same problem on a new laptop with AMD graphics (13.12 driver) and Windows 8.1 64 bit. My TV supports Full HD 23p, 24p, 50p, 59p and 60p refresh rates. When using D3D11 exclusive 10 bit mode, render queue and present queue don’t get filled if the selected refresh rate is 24p or 60p. The queues fill fine with other supported refresh rates though.

Although queues don’t get filled with 24p and 60p refresh rate, there are no reported or perceptible dropped frames. The queues fill properly in windowed mode or D3D9 output.

Warner306

1st August 2015, 21:32

For what it's worth, NNEDI3 is still working for me with Windows 10 Nvidia drivers and a GTS 450.

huhn

1st August 2015, 23:20

please name the drivers version it's know to work up to driver version 353.49.

AFAIK fermi didn't have an update proper wddm 2.0 driver yet

aufkrawall

1st August 2015, 23:50

Yes. Most current driver for Fermi has the same version number, but lacks WDDM 2.0 and still has D3D9 interop.

x7007

2nd August 2015, 01:48

I take quote from nvidia

I, and then ManuelG already addressed this.

But where as he cannot say so, I can and will:

MadVR needs to update to support the requirements/Changes in Windows 10 DXGI/DXVA, meaning they need to implement and use the necessary opencl/d3d interop as required: On windows 10, they need to use cl_nv_d3d10_sharing.

ManuelG-Customer Care- DirectX 12 support for Fermi is still being finalized and will be available in a future Game Ready driver later this year before the first DX12 games are released. DX9 interop is not supported on win10 so that is why the file was removed from driver 353.62. Not clear what the plan for support for this is as I am still researching this with the software team.

Madvr won't be able to add cl_nv_d3d10_sharing or the alter that should work instead of the interop d9 ?

aufkrawall

2nd August 2015, 03:23

James Freeman

2nd August 2015, 04:10

Is there anything else effected by Windows 10, or mainly NNEDI3?
Can I upgrade from 7 without any ill effects to HTPC (mpc-hc, lav, madvr) functionality?

ryrynz

2nd August 2015, 04:46

Is there anything else effected by Windows 10, or mainly NNEDI3?
Can I upgrade from 7 without any ill effects to HTPC (mpc-hc, lav, madvr) functionality?
FWIW I haven't encountered any issues except for the OPENCL issue which was fixed by reverting to the older driver and disabling third party updates in Windows. Otherwise it's been smooth sailing.

Asmodian

2nd August 2015, 05:29

Has anyone else, except this random guy of the forums, said this?
https://forums.geforce.com/default/topic/860152/geforce-drivers/official-windows-10-353-62-game-ready-display-driver-feedback-thread-7-29-15-/post/4618423/#4618423
He may just consider himself as too important.

The Nvidia rep, ManuelG, did state that the dx9 extension was removed because it was incompatible with Windows 10, which does imply madVR would need to use the dx10 extension, but that doesn't make sense given that earlier Windows 10 drivers had it and it works well.

Is there anything else effected by Windows 10, or mainly NNEDI3?
Can I upgrade from 7 without any ill effects to HTPC (mpc-hc, lav, madvr) functionality?

My experience with Windows 10 matches ryrynz'es; other than Nvidia's issues with OpenCL, Windows 10 has been excellent for HTPC use.

My NNEDI3 performance is better than ever; I can double the luma of 1080p30 to 4K with NNEDI3 128 but I have to use pretty fast options, e.g. Jinc chroma scaling and ordered dithering, for the other settings. My GPU doesn't even clock all the way up using NNEDI3 64 doubling for 1080p24 and downscaling to 2650x1440. For 720p30 to 2560x1440 I can use NNEDI3 256 with performance to spare, I really did not expect that.

I am using a 980 Ti with the 353.30 drivers.

ShadyCrab

2nd August 2015, 06:14

The Nvidia rep, ManuelG, did state that the dx9 extension was removed because it was incompatible with Windows 10, which does imply madVR would need to use the dx10 extension, but that doesn't make sense given that earlier Windows 10 drivers had it and it works well.

For many of us, if you look back through the thread, there were numerous issues with NNEDI3 on Win 10 NVidia, through many drivers. I personally had many issues with it I didn't have on Win 8.1, like green video occasionally, requiring me to change either dimension of the video to resize it and somehow trigger it back out of greenscreen. Many, many crashes too, and other issues.

If that truly is the reason its gone (it seems like they could put some effort into possibly fixing it, unless its some underlying WDDM thing), then I understand where they are coming from. It's still a shame, as NNEDI3 is the cleanest algo so far with sufficient neurons.

I dont know how high it is on madshi's priority list, but DX9 is irrelevant now, and as a minimum DX10/HLSL 4.0+ should be standard now. XP's days are over now, and Win 7 has replaced it as the long-term OS. It would be a lot of work due to DX10's fundamental changes, but hopefully in the future we can have a full D3D10/11 rendering pipeline, from start-present.