madVR - high quality video renderer (GPU assisted)

[VBB]

That's something unrelated.

[B1gD4ddy]

ok, would be nice if that gets on the bug list.

[dbcooper]

Any issues reported with Windows 10 2004 update?

[VBB]

Been using it with madVR. No problems so far.

[Ænz]

Hi. Im having some trouble with the custom resolutions needed for playback without dropped frames on nvidia cards.
I have a custom resolution in madvr that works well, 1 dropped or repeated frame every 11+ hours. But it only works in mpc-hc. If I play a movie in mpc-be or kodi dsplayer, I get the dropped frame every 4,5 min. or so. I can not find the source of this problem. I have verified that madvr is used in the other players. I have tried different versions of dsplayer, witch I probably would prefer.

[huhn]

make sure the same audio renderer and audio output is used.

[Ænz]

make sure the same audio renderer and audio output is used.

Thanks. I just checked it. I use DirectSound bitstream to my Denon AVR. Have that set as output, passthrough device and audio renderer. Still the same, repeated frame every 4.5 min. But runs smooth in mpc-hc

[B1gD4ddy]

i added the hdr passthrough thing to the bug tracker and madshi actually assigned it... http://bugs.madshi.net/view_all_bug_page.php ...now wait.

[SigHunter]

i added the hdr passthrough thing to the bug tracker and madshi actually assigned it... http://bugs.madshi.net/view_all_bug_page.php ...now wait.

thank you very much! I hope madshi can fix this soon, I hate to be stuck to old graphics driver versions and this effectively breaks madvr for the future. 446.14 is the last one that works with our old madvr

[mogli]

Btw which scaler is Krigbilateral exactly?

[Alexkral]

You mean in madVR? You have some similar bilateral options for chroma resampling. KrigBilateral is a shader created by Shiandow for MPDN and also adapted by igv for mpv. It's used for chroma resampling though it could also be used for 2x chroma upscaling.

[huhn]

the bilateral in madVR should be from shiandow too. i can't recommend it. this scaler lives from the luma channel but the luma channel is not always available when chroma scaling is critical.

[deado]

Any issues reported with Windows 10 2004 update?

Fullscreen exclusive mode has stopped working for me (player is mpc-hc 1.9.6).

[Klaus1189]

@deado: What is exactly the issue?

[deado]

@deado: What is exactly the issue?

Strange.. I changed nothing, woke up today, and it's now working again. Couldn't get it to work last few days no matter what I tried (since upgrading to win10 2004).

Oh well good to know it's not an 'issue' generally.

[RXP]

Had a really frustrating problem now - and only realised what it was at the end of the movie. Display Identification got messed up so none of my profiles applied.

I've got all my profiles nicely setup in MadVR - everything "just works" recently, very happy with it. Then sat down for a re-watch of the Dark Knight with my wife and I noticed the colours were over-saturated. I didn't want to pause the movie and faf around with settings because I thought it was the tone mapping algo having a few issues and it's really frustrating to do that when you're watching with someone else.

Turns out for some reason MadVR had a different display device active, no EDID info just "generic PnP monitor" with a string of Windows info. So none of my profiles for the other EDID's were selected. I assume MadVR defaults to Rec709 so the tone mapping was trying to tone map to Rec709 and I'd flipped to the Rec2020 mode on my PJ. Hence over-saturation.

No idea why Windows/Nvidia/whatever decided to show my display as another display. 0 updates, 0 changes - I dual boot for this exact reason so I keep one boot perfectly stable. But I guess even that doesn't work. Sigh....

[Martin51]

Hello, I am following the tutorial for custom modes. Once optimization data is available, I see the "Test mode" button instead of "Apply". I tried changing the refresh rate, and restart with no luck. Any ideas?

[mclingo]

Hello, I am following the tutorial for custom modes. Once optimization data is available, I see the "Test mode" button instead of "Apply". I tried changing the refresh rate, and restart with no luck. Any ideas?

shouldnt be that hard but it doesnt always work but when it doesnt it can be a reall faff to get this work, sometimes, even though you've done it all correct it still does nothing, I have this issue with my wireless Arctis pro headphones, no matter what I do or set I get a frame drop every 20 mins, which isnt a disaster but I get several hours or days with my AMD HDMI audio without and special configs at all.

Quick quide.

Delete any you've already put in and reboot

click add - select your resolution and refresh rate and click apply.

Play a move for 20 mins.

Go back in and open the optimize tab and select one of the options that look good and apply.

Reboot and see what you get.

What are you currenting timings, post a screen shot of your control J screen after 5 mins playback in case you've set something up wrong, for instance you only need timings for NVIDIA cards usually, AMD card have impeccable timing and require not custom modes (usually).

[x7007]

hey guys, nvidia changed its hdr code in the 45x.xx drivers. because of that madvr hdr passtrough does not work anymore.

quote from guru3d forum: "HDR works fine, it's just madvr that doesn't know what to do with how Nvidia changed the HDR code in the drivers. If it's like the 451.22, NVHDR activates the Windows HDR option instead of running its own code."

from me: can confirm. mpc be madvr hdr passtrough does not work anymore. it just activates windows hdr but movie is still sdr. only help for now is to go back to 446.14.


so please update madvr and make hdr passtrough usable again. thx.
Yes works with potplayer x64 also.. Thanks

[Martin51]

mclingo, I followed the step you suggested

- Deleted everything, then rebooted
- Clicked edit on the custom mode I want to set
http://forum.doom9.org/attachment.php?attachmentid=17406&stc=1&d=1593620434
- Test mode, then rebooted
- Played video for about 20m (Switched automatically to 25 Hz)
- Stopped video (switched back to 60Hz)
- Back to the settings I see:
http://forum.doom9.org/attachment.php?attachmentid=17407&stc=1&d=1593620434
- Clicked the optimize button
http://forum.doom9.org/attachment.php?attachmentid=17408&stc=1&d=1593620434

=> There is no apply button

I played the video again a few min cause I forgot to make the screenshot:
http://forum.doom9.org/attachment.php?attachmentid=17409&stc=1&d=1593620434

[kuchikirukia]

I have a 60Hz and 120Hz monitor. When I'm extending my Windows 10 desktop between them I get 60 presentation glitches per second when MPC-HC w/ madVR is in windowed mode on the 120Hz monitor. (the counter increments even when the video is paused)
Sometimes the video looks fine, sometimes it looks like it's dropping frames (even though the dropped frames counter doesn't show it)
It doesn't have an issue when fullscreened or on the 60Hz monitor, or when Windows is set to only display on the 120Hz.

[huhn]

windows doesn't really like 2 refreshrates the desktop is running only at 1 of them newer version are more clever about this so it's not accidentally running at 60 HZ internally on a 120 hz screen.

overlay rendering avoids.

[biship]

Is there a work around for the HDR bug on Nvidia? My LG thinks non-HDR video's are HDR and so it switches into HDR mode :(
I am using Media Portal v1 + LAV + Madvr.

[Andy o]

Is there a work around for the HDR bug on Nvidia? My LG thinks non-HDR video's are HDR and so it switches into HDR mode :(
I am using Media Portal v1 + LAV + Madvr.

If you re-open the file (e.g. double click on it again in Explorer) without closing MPC-HC it will display HDR, but I now have the issue that it's crushing black levels in that situation. Of course if I switch to 16-235 in madVR it displays correctly, but then SDR has the wrong levels. This is with the GTX 1650 Super 451.48 drivers on Win 10 1909 madVR 0.92.17. RGB Full output.

[SamuriHL]

I went back to the previous studio driver. The latest stuff is a mess. Maybe it'll get fixed eventually but right now I'm sticking to working drivers.

[njfoses]

i went back to the previous studio driver. The latest stuff is a mess. Maybe it'll get fixed eventually but right now i'm sticking to working drivers.

442.19?

[Klaus1189]

Please see here:
AMD, Intel and Nvidia driver issues and last recommended version (https://forum.doom9.org/showthread.php?t=176013)
Last working GRD: 446.14
Last working SD: 442.92

[SamuriHL]

442.19?

Nope.

442.92-desktop-win10-64bit-international-nsd-dch-whql

[Martin51]

mclingo I have posted a reply a few days ago with screen shots but my post never appeared. I followed all the steps you suggested, but when clicking the optimise button I see "Test mode" instead of apply.

In fact, I was only interested in custom modes because I do experience some audio dropouts from time to time. I thought it had something to do with my old AV receiver, I now have a recent one but the problem is still present.
Can the dropouts be related to dropped or repeated frames?

[mclingo]

audio dropout are unlikely to be custom res, more likely something wrong in your hardware chain, perhaps one of your cables is not great and isnt handing the bandwidth very well or just a driver issue, i've had cabling issues before. Hard to say more without a comprensive list of your setup. Also, dont post pics on here, they take ages to show up, host them on ONEDRIVE or something like that, use a URL.

[VincAlastor]

Fullscreen exclusive mode has stopped working for me (player is mpc-hc 1.9.6).

here the same!

periodically dropped frames in DX11 fullscreen mode since windows 10 2004 update (Nvidia driver 451.48). No difference with on/off gpu scheduling

has anyone already found a solution to the problem?

[Klaus1189]

...

Please try older drivers: GRD 446.14 or SD 442.92. Does that eliminate the issue?

[VincAlastor]

Please try older drivers: GRD 446.14 or SD 442.92. Does that eliminate the issue?

Thank you very much.

GeForce 451.67 GRD WHQL also fixed the issue of frame drops.

[Klaus1189]

You‘re welcome. See the driver thread for issues:
AMD, Intel and Nvidia driver issues and last recommended version (https://forum.doom9.org/showthread.php?t=176013)

I am very active and update as fast as possible.

[orena]

What is the best MadVR settings for High End HTPC with Nvidia RTX 2080TI GPU?

[nussman]

https://forum.doom9.org/forum-rules.htm

[mclingo]

What is the best MadVR settings for High End HTPC with Nvidia RTX 2080TI GPU?

Thats like walking into a DIY store and asking a member of staff how to build the best house.

install it, play around a bit, do a bit of reading so you understand what each option does and then you'll be able to push each option a little to see how your PC performs.

There are no "best " options, everyone has different setups and requirements, this is why MADVR is so good, it caters from the noobs who want to just install and go to pro's and everyone in the middle.

I'm no longer a noob but a long way from a pro.

[Asmodian]

What are some good settings MadVR settings for High End HTPC with Nvidia RTX 2080TI GPU?

I have some settings.bin's linked at the end of the third post of the thread in my signature. They are what I use on a 2080 Ti plugged into a LG CX OLED TV. If you are willing to do a bit of reading that thread is also a good place to start understanding the settings so you can tweak to your preferences. :)

HDR is a bit troublesome now, since I use current Nvidia drivers, but I still use passthrough for HDR. It does require turning on Windows HDR manually but I don't watch that much HDR content.

[VBB]

As, did you replace your C9?

[Asmodian]

As, did you replace your C9?

Yes, it is still around but I have a small (48") CX now too. I no longer have a C7, just in time. ;)

[mzso]

Hello!

Which AMD driver version is considered the best (or least bad) for madVR right now?

[Klaus1189]

Latest 20.7.2 is fine.

[mzso]

Latest 20.7.2 is fine.

With the plethora of problems I have doubts. (Though Windows 10 might be garbage as well.)

[Klaus1189]

Which one do you prefer then?

[TheDarkTemplar]

Hello! I keep getting this crash in MPC-HC about 50% of the time I open and fullscreen a video:

https://drdump.com/UploadedReport.aspx?DumpID=49079987

It is apparently an issue in MadVR. Is there some workaround to this or will it be fixed at some point? It's not the worst issue ever since I can just relaunch again but it's a bit annoying.

I think it may be related to switching out of HDR since I am generally fullscreening non-HDR videos.

[mzso]

Which one do you prefer then?

Maybe all's garbage. And/or the OS is...

Starting to get fed up with everything always being sh*t.

[Klaus1189]

Starting to get fed up with everything always being sh*t.

Developers of graphic drivers don't seem to care about HTPC usage, for many years. None of the three teams is perfect, none will work as you want it to if you try every little thing out.
It is like finding the smallest evil :devil:

[Charky]

Is madvr known for causing a/v sync issues ?

I created a short movie using Adobe Premiere Pro CC 2020 (Vimeo 1080p23 preset, H264/AAC, MP4 container). In Premiere, everything is in sync. However, on my HTPC (MPC-BE, LAV, madvr v113, custom resolutions), I noticed the picture is slightly behind the audio, like 50 ms. With VLC ou MPC-BE using EVR, the sync is perfect, so I'm pretty confident that the issue lies with madvr itself.

I actually never noticed that before with "regular" content (i.e. bluray remuxes) but a 50 ms delay is not much and I probably haven't been paying as much attention as with my own movie, because I spent a lot of time lipsyncing dubbed dialogues and creating foley sounds and making sure everything is in sync.

I solved the issue by setting a 50 ms audio delay in LAV audio decoder (regular content now seem to be in perfect sync too), but am i missing anything ?

[Shiroe]

Maybe someone can assist me with a problem. In PotPlayer, when the video renderer is set to madVR, subtitles flicker when seeking.

Here is a demonstration:

https://e.pcloud.link/publink/show?code=kZXKkZJ5K6wzDUVAh8uzaE6vCEmfIODk8X

In those videos, I am jumping by frames. Subtitles disappearing is not limited to jumping by frames. When I jump backward by seconds, sometimes subtitles that had appeared during normal playback do not appear when the video automatically resumes. In that case, I can jump backward again, pause the video, and then play the video to make the subtitles visible. I recently discovered MPC-BE suffers from that same problem, i.e., subtitles disappearing when jumping backward by seconds, when the video renderer is set to madVR.

If I recall correctly, those problems were not present in PotPlayer when I installed it via LAV Filters Megamix, and as far as I know, those problems are also not present in MPC-HC, which I installed via K-Lite Codec Pack Standard.

[Alexkral]

Hi guys, I'm finally done with the Super-resolution implementation in AviSynth+ that I've been working on, you can try it here (https://github.com/Alexkral/AviSynthAiUpscale). Now I'm doing some tests, and since there was some interest, I started with one including all the NGU settings. For this I have used Set14 (https://github.com/xinntao/BasicSR/blob/a19aac61b277f64be050cef7fe578a121d944a0e/docs/Datasets.md#common-image-sr-datasets), consisting of 14 images, which with 16 NGU settings has forced me to capture 224 screenshots, so for larger datasets this will not be possible. The other algorithms tested are:

- The five new AiUpscale models.
- The three FSRCNNX models.
- Waifu2x-caffe (upconv_7_photo model).
- Lanczos3 (Matlab) for reference.

The FSRCNNX models were implemented in the same way as the AiUpscale models. The metrics used are:

- PSNR
- SSIM
- IFC
- PI
- SIS

PSNR and SSIM are well known. IFC (http://users.ece.utexas.edu/~gustavo/papers/de%20Veciana1251902020.pdf) (Information Fidelity Criterion) is a metric based on natural scene statistics that has been shown to have a high correlation with perceptual scores in the context of SR evaluation. PI (https://arxiv.org/pdf/1809.07517.pdf) (Perceptual Index) is a No Reference metric designed for the 2018 PIRM Challenge on Perceptual Image Super-resolution, which is also highly correlated with the ratings of human observers in a human-opinion study. SIS (https://core.ac.uk/download/pdf/187752279.pdf) is a method specifically designed for assessing the visual quality of super-resolved images, which also has a very high correlation with a human-opinion study carried out with a database built for the same purpose.

Downsampling was done with Bicubic (Catmull-Rom settings). All the metrics were computed on the Y channel. NGU Anti-Alias introduces a half pixel offset that needed to be corrected, which slightly worsens the scores (in the case of PSNR I'd say only a few hundredths). For the PI less is better, and since it is a No Reference metric, the reference is also included.

Model PSNR Model SSIM Model IFC Model PI Model SIS
__________________________ ________________ __________________________ ________________ __________________________ ________________ __________________________ ________________ __________________________ ________________

AiUpscale HQ 32.5251817903342 AiUpscale Medium 0.90942544644301 AiUpscale Medium 8.27967367353770 Reference 3.40104937646879 AiUpscale Medium 0.92414368352855
AiUpscale Medium 32.4232206599772 AiUpscale HQ Sharp 0.90843988514755 AiUpscale HQ 8.24574655763528 AiUpscale HQ Sharp 3.58668578180591 AiUpscale HQ Sharp 0.92382785147580
Waifu2x-caffe 32.3473914060035 AiUpscale HQ 0.90839669836945 AiUpscale Fast 8.16061456261743 AiUpscale Medium 3.65547901829712 AiUpscale HQ 0.92338216249267
NGU Sharp (very high) 32.2938176168901 AiUpscale Fast 0.90693736664703 AiUpscale VDSR 7.90344864991210 Waifu2x-caffe 3.67943591351007 AiUpscale Fast 0.92006283988828
AiUpscale VDSR 32.1939373807653 AiUpscale VDSR 0.90521242215438 AiUpscale HQ Sharp 7.73815420390795 AiUpscale Fast 3.71741685428603 AiUpscale VDSR 0.91742165939633
AiUpscale Fast 32.1863953769405 Waifu2x-caffe 0.90495347031380 Waifu2x-caffe 7.68228442626807 FSRCNNX_x2_8-0-4-1 3.81516969239280 Waifu2x-caffe 0.91636248251022
FSRCNNX_x2_56-16-4-1 32.1759503587084 NGU Sharp (very high) 0.90460008282567 FSRCNNX_x2_8-0-4-1 7.66717466529441 AiUpscale HQ 3.84336339556552 FSRCNNX_x2_8-0-4-1 0.91619538116915
FSRCNNX_x2_16-0-4-1 32.1484038971244 NGU Sharp (high) 0.90193258684294 FSRCNNX_x2_16-0-4-1 7.56245532653046 NGU Sharp (very high) 3.93987906364856 FSRCNNX_x2_16-0-4-1 0.91375067249656
AiUpscale HQ Sharp 32.1199740239966 FSRCNNX_x2_8-0-4-1 0.90107924389368 NGU Sharp (very high) 7.56185554352781 FSRCNNX_x2_56-16-4-1 3.96283639678717 NGU Sharp (very high) 0.91283152130317
NGU Sharp (high) 32.0745914126075 FSRCNNX_x2_16-0-4-1 0.90021250932189 FSRCNNX_x2_56-16-4-1 7.54238083322097 NGU Sharp (high) 3.96668297143171 FSRCNNX_x2_56-16-4-1 0.91250552777137
FSRCNNX_x2_8-0-4-1 32.0254211392229 NGU Sharp (med) 0.90008345962041 NGU Sharp (high) 7.10509959144021 FSRCNNX_x2_16-0-4-1 3.98720638993818 NGU Sharp (high) 0.90383627513313
NGU Sharp (med) 31.9361908894940 FSRCNNX_x2_56-16-4-1 0.89928737851361 NGU Sharp (med) 6.93150412700636 AiUpscale VDSR 4.04068349093325 NGU Sharp (low) 0.90309609902881
NGU Sharp (low) 31.7104151080959 NGU Sharp (low) 0.89500452165092 NGU Standard (very high) 6.79878287005572 NGU Sharp (med) 4.04345458415675 NGU Sharp (med) 0.90084060562877
NGU Standard (very high) 31.5974304026016 NGU Standard (very high) 0.89373264838815 Lanczos 6.78807153422038 NGU Sharp (low) 4.20257164181289 NGU Standard (very high) 0.89213869604695
NGU Standard (high) 31.4416756290765 NGU Standard (high) 0.89223745931491 NGU Sharp (low) 6.76680062532443 NGU Standard (very high) 4.27329141474409 NGU Standard (high) 0.88663713794603
NGU Standard (med) 31.3266464934018 NGU Standard (med) 0.89079871365700 NGU Standard (high) 6.65548044824456 NGU Standard (high) 4.34545507861726 NGU Standard (med) 0.88214872829691
NGU Standard (low) 31.1795059486520 NGU Standard (low) 0.88724999653284 NGU Anti-Alias (very high) 6.54847171682001 NGU Standard (med) 4.34697051435912 NGU Standard (low) 0.87988322632132
NGU Anti-Alias (very high) 31.0032971499628 NGU Anti-Alias (very high) 0.88462531208344 NGU Standard (med) 6.52544719594284 NGU Standard (low) 4.47568646264313 NGU Anti-Alias (very high) 0.87742671984170
NGU Anti-Alias (high) 30.9569815543731 NGU Anti-Alias (high) 0.88321233882888 NGU Anti-Alias (high) 6.42962113230166 NGU Soft (very high) 4.87787622216863 Lanczos 0.87560000704257
NGU Anti-Alias (med) 30.8970336747559 NGU Anti-Alias (low) 0.88065405839635 NGU Standard (low) 6.38911485598076 NGU Soft (med) 4.89969182841594 NGU Anti-Alias (high) 0.87458590111814
NGU Anti-Alias (low) 30.8339491643038 Lanczos 0.88061992663747 NGU Anti-Alias (low) 6.30590655792457 NGU Anti-Alias (very high) 4.91336898065817 NGU Anti-Alias (med) 0.87280516539207
Lanczos 30.6067648881095 NGU Anti-Alias (med) 0.88043102098649 NGU Anti-Alias (med) 6.18364967410367 Lanczos 4.92084403030973 NGU Anti-Alias (low) 0.87012785173923
NGU Soft (very high) 30.6012993686409 NGU Soft (very high) 0.87854781224895 NGU Soft (very high) 5.98479762818895 NGU Soft (high) 4.92618718914022 NGU Soft (very high) 0.85343713547148
NGU Soft (high) 30.4664433204248 NGU Soft (high) 0.87680636052519 NGU Soft (high) 5.92007932481318 NGU Anti-Alias (low) 4.92710046176526 NGU Soft (high) 0.85054526901960
NGU Soft (med) 30.4483124249698 NGU Soft (med) 0.87621006480048 NGU Soft (med) 5.89496906342640 NGU Anti-Alias (high) 4.94724879619383 NGU Soft (med) 0.84712992882588
NGU Soft (low) 30.2594794628864 NGU Soft (low) 0.87162551537298 NGU Soft (low) 5.71276482218494 NGU Anti-Alias (med) 4.95971449743798 NGU Soft (low) 0.83820122012128
NGU Soft (low) 5.07866068664092

As you can see, the NGU scores are completely ordered (Sharp > Standard > Anti-Alias > Soft) especially in terms of PSNR and SSIM, with some differences in the other metrics (e.g. AA low performs better than AA med for three metrics). NGU Soft is consistently below Lanczos, same as AA for all metrics except PSNR and SSIM, and even Standard for IFC. FSRCNNX_x2_8-0-4-1 performs better than the other FSRCNNX for all metrics except PSNR. Waifu performs well, and needless to say, I am very satisfied with the AiUpscale results. :)