madVR - high quality video renderer (GPU assisted)

[Stereodude]

As long as the average is below 1/refresh and you don't get dropped or repeated frames, it's fine. The max could be a problem if your buffers / queues are small.

[huhn]

It doesn't. All that it knows is the HDR peak brightness (of the display), as set during 3D LUT creation, then applies the BT.2390 rolloff + saturation adjustment, then maps into the display colorspace (with imo great results btw, my implementation has come some way in the past three weeks and will be available in the next version of DisplayCAL). Of course, a 3D LUT cannot react to dynamic metadata, which could be seen as a drawback. So far, I don't think it matters much, though.

using MPC-BE with internal codecs could work as a work around.

madVR should get the bt 2020 information but not the HDR infos and if i'm not mistaken that is what you want.

i will for sure try your new 3D LUT.

[fhoech]

using MPC-BE with internal codecs could work as a work around.

madVR should get the bt 2020 information but not the HDR infos and if i'm not mistaken that is what you want.
Good point, I will try that, thanks.

i will for sure try your new 3D LUT.
I'll likely post it in the madVR/Argyll thread on AVSforum when it becomes available. Currently the changes are only in version control.

[huhn]

Good point, I will try that, thanks.


I'll likely post it in the madVR/Argyll thread on AVSforum when it becomes available. Currently the changes are only in version control.

don't worry i will find it i just need my TV back first.

[wolfman2791]

Is it me or is there no difference between chroma upscaling with super xbr compared to nnedi... my gpu can handle both fine. But i don't see any reason to use up the additional resources. Am I wrong in my assessment?

[Warner306]

Is it me or is there no difference between chroma upscaling with super xbr compared to nnedi... my gpu can handle both fine. But i don't see any reason to use up the additional resources. Am I wrong in my assessment?

No, your assessment is accurate. You've discovered how difficult it is to detect changes to the chroma layer.

[Ver Greeneyes]

There are some clips where the difference is more visible, but it's usually really hard to tell, especially in motion. The only reason to use high quality upscaling for chroma is when you have a lot of power left over - chroma usually only has a quarter as many pixels as luma, so high quality upscaling is relatively inexpensive.

[Georgel]

You quoted me for something I didn't say. I ran both through a picture comparison tool and there's little change in detail nothing really noticeable.. It's just the lines which we already know are thinned too aggressively.

Sent from my Nexus 5 using Tapatalk

Oh, sorry for quoting you on the wrong reply.

At any rate, we both can agree that the thinning is very aggressive, that's totally true! I just happen to enjoy it with a lot of content :D





For some reason I keep going back to Jinc for all my scaling needs. It seems to look more natural and has almost no artifacts unlike the other scalers.

NNEDI3 128 does look quite good. 256 looks a tiny bit better, but my 1080 GTX tends to get hot with that setting and the render times are at about 39 ms, which is too close to the frame time of 24p content. Should I lower the setting if the max sometimes hits 45 ms, but the average is around 30 ms? I'd ideally want the max under the 41.7 ms of 24p content?

You're playing FHD content and getting these times?

I imagined that after getting an 1080 I would finally be able to do 4k + NNEDI 256 neurons :D

[ryrynz]

Is it me or is there no difference between chroma upscaling with super xbr compared to nnedi... my gpu can handle both fine. But i don't see any reason to use up the additional resources. Am I wrong in my assessment?

As Ver said the differences are extremely minor. Super-xbr is your best bang for buck and the main choice I think for most in the forum as it's relatively inexpensive in comparison to NNEDI3, just make sure to tick anti-ringing.

[ryrynz]

For some reason I keep going back to Jinc for all my scaling needs. It seems to look more natural and has almost no artifacts unlike the other scalers.

NNEDI3 128 does look quite good. 256 looks a tiny bit better, but my 1080 GTX tends to get hot with that setting and the render times are at about 39 ms, which is too close to the frame time of 24p content. Should I lower the setting if the max sometimes hits 45 ms, but the average is around 30 ms? I'd ideally want the max under the 41.7 ms of 24p content?

I wouldn't recommend 256 neurons it's really not worth it, don't let the max render time get that high and accept that extra heat and noise for something you won't even notice.
There's significant diminishing returns past 64 neurons.

[AngelGraves13]

Oh, sorry for quoting you on the wrong reply.

At any rate, we both can agree that the thinning is very aggressive, that's totally true! I just happen to enjoy it with a lot of content :D







You're playing FHD content and getting these times?

I imagined that after getting an 1080 I would finally be able to do 4k + NNEDI 256 neurons :D

My monitor is 1440p, so I can't really test 4K unless I downscale.

[chros]

I'll likely post it in the madVR/Argyll thread on AVSforum when it becomes available. Currently the changes are only in version control.
Cheers, I'm curious.
don't worry i will find it i just need my TV back first.
What happened to it? What model?

[AngelGraves13]

I wouldn't recommend 256 neurons it's really not worth it, don't let the max render time get that high and accept that extra heat and noise for something you won't even notice.
There's significant diminishing returns past 64 neurons.

That's why I'm pretty much sticking to Jinc for all my scaling needs.

[Betroz]

That's why I'm pretty much sticking to Jinc for all my scaling needs.

You can't see a difference in sharpness between Jinc and NNEDI3 64 neurons?

[AngelGraves13]

You can't see a difference in sharpness between Jinc and NNEDI3 64 neurons?

I can, but NNEDI3 has more artifacts. It tends to look more processed. Not a huge issue with chroma, but it doesn't look too good with image doubling.

[ryrynz]

Guys, Check your OSD after installing the Anniversary update as 14393.10 broke NNEDI3 on my GTX 960 and may likely bork other setups too.

NNEDI3 was working fine previously just prior to the update, I simply reinstalled the graphics drivers selecting a clean install and it's working fine now.

[Ver Greeneyes]

Thanks. Yeah, those big updates do tend to break drivers - it's pretty annoying.

[Betroz]

Guys, Check your OSD after installing the Anniversary update as 14393.10 broke NNEDI3 on my GTX 960 and may likely bork other setups too.

NNEDI3 was working fine previously just prior to the update, I simply reinstalled the graphics drivers selecting a clean install and it's working fine now.

It's because the Windows update deletes OpenCL driver. Happens for every big update to Win10.

[ianken]

Anyone else having problems with Pascal GPU (1070,1080) getting the proper P-state when using Madvr in mpc-hc or mpc-be?...

YES. MY findings:

OK. Here's my fun-times with MAD-VR and 4K + HDR. Running 0.90.23 in MPC-HC with a 12 core XEON and GTX 1080 on an HP DeamColor z32x 4K monitor.

Windowed: like butter. Even with the window nearly the full size of the 4K display. GPU clock :1.8 GHZ, load (with all HDR processing enabled: 75%) No dropped frames decoding 4k60p HDR.

Full screen exclusive: Drops frames like mad with both 4K24 or 4K60. GPU-Z shows the GPU clock is WAY lower at 835Mhz and fully loaded. Unusable. Even dialing back settings, frames still drop.

Really odd.

[digitech]

anyone uses Nvidia DSR with MadVR on MPC-HC with a 1080p display.. is it recommended for 2k or 4k then downscaling to native res?

[ryrynz]

anyone uses Nvidia DSR with MadVR on MPC-HC with a 1080p display.. is it recommended for 2k or 4k then downscaling to native res?

Doesn't work. Use NNEDI3 and your choice of downscaler, I'd recommend SSIM 1D/2D and Jinc.

[Xaurus]

YES. MY findings:

OK. Here's my fun-times with MAD-VR and 4K + HDR. Running 0.90.23 in MPC-HC with a 12 core XEON and GTX 1080 on an HP DeamColor z32x 4K monitor.

Windowed: like butter. Even with the window nearly the full size of the 4K display. GPU clock :1.8 GHZ, load (with all HDR processing enabled: 75%) No dropped frames decoding 4k60p HDR.

Full screen exclusive: Drops frames like mad with both 4K24 or 4K60. GPU-Z shows the GPU clock is WAY lower at 835Mhz and fully loaded. Unusable. Even dialing back settings, frames still drop.

Really odd.
Interesting. Actually my core clock is just fine, but the memory runs slower than usual, typical for the "2d" p-state used for blu-ray etc. I will do some testing and see if windowed makes any difference.

[Stereodude]

My GTX 1080 is supposed to arrive today and it's slated for my HTPC so I will report back after testing.

[ryrynz]

My GTX 1080 is supposed to arrive today and it's slated for my HTPC so I will report back after testing.

Give the leaked 369.09 drivers a shot too (new branch)

[sneaker_ger]

Yeah, if you have strange problems with video playback (glitches, frame drops, stuttering) and Pascal cards (GTX 1060/1070/1080) try newer drivers. There are reports about dpc latency problems. Unfortunately, many report even the Hotfix drivers don't fix them completely. (Don't know about 369.09 suggested by ryrnz.)

https://forums.geforce.com/default/topic/941579/geforce-1000-series/gtx-1080-high-dpc-latency-and-stuttering/1/
https://forums.geforce.com/default/topic/951723/geforce-drivers/announcing-geforce-hotfix-driver-368-95/

[jkauff]

Guys, Check your OSD after installing the Anniversary update as 14393.10 broke NNEDI3 on my GTX 960 and may likely bork other setups too.

NNEDI3 was working fine previously just prior to the update, I simply reinstalled the graphics drivers selecting a clean install and it's working fine now.
NNEDI3 broke on my machine, too. After reinstalling the drivers, the problems in MPC-HC playback went away, but it doesn't work in JRiver MC. Maybe a clean install isn't enough. I could try a deep uninstall and then repeat the clean install.

MPC-HC has a different madVR problem, though. I can't get out of FSE mode with the mouse, I have to use the keyboard. No idea what that's about.

[Stereodude]

Give the leaked 369.09 drivers a shot too (new branch)
The 368.81 drivers worked fine for me under 8.1 x64. It's 80% faster than my R9 380 in the 5 madVR scenarios I tested. I'm not 100% confident in those numbers because it looked like the card was clocking itself down some due to not being heavily loaded by the madVR settings. I will see if I can force the highest power state and try again.

[Karl78]

Hi guys. After win10 anniversary update, madVR crashes in full screen mode (exclusive or not is the same). I'm on laptop (graphic: Intel hd 530 with latest driver and nvidia 970m with 359.54 driver - I know, 368.81 are the latest) and jriver as media player. I've reinstalled jriver, madvr (0.90.23), both driver intel and nvidia, and nothing to do (no problem with standard red october). Anyone experienced that problem? Thanks

[har3inger]

Post-anniversary update, the performance drop between .90.20 and .90.23 has worsened. Performance in .90.20 is also worse.

I clean installed both my intel and AMD drivers, and issue persists.

Something new: in FS windowed mode, I get really bad horizontal striping artifacts. These stripes are long, thin rectangular blocks/fragments from frames that were already presented. Artefact issue goes away in FSE mode. Performance in FSE mode is now what it was in FS windowed mode before the win10 update, but the FS windowed mode performance is 10-20% slower and prevents smooth playback at old settings.

For now, changing playback to FSE mode and staying with .90.20 is the only way I have acceptable performance and reliability.

Jeez, just couldn't have had a smooth upgrade, hey?

[Stereodude]

This GTX 1080 is a lot of fun. I've got a 1080p display. I can play back p24 DVDs with NNEDI3-128 Luma doubling, NNEDI3-32 Chroma doubling, NNEDI3-64 Luma quadrupling, NNEDI3-32 Chroma quadrupling, SSIM2D 100AR downscaling, all the upscaling refinements enabled (with light settings) except for LumaSharpen and the render time is about 37mS. :devil:

[Patrik G]

this test file is kind of broken.

it is interlanced and madVr isn't deinterlancing it because it is saying that is has 60 FPS before deinterlancing.

when lavfilter is deinterlancing than madVr things this file is 120 hz.

ok that would explain it
thanks for the information

is anyone else using a 3D LUT calibration with madvr?
how many patches do you use for profiling?

[huhn]

is anyone else using a 3D LUT calibration with madvr?
how many patches do you use for profiling?

i use a 3D LUT and i use way to many patches about 12000.

choosing patches depends a lot on your screen. if your gamut is close to DCI P3 i would use more patches for a BT 709 3D LUT.

maybe you will liek this thread: http://www.avsforum.com/forum/139-display-calibration/1471169-madvr-argyllcms-146.html

[Georgel]

This GTX 1080 is a lot of fun. I've got a 1080p display. I can play back p24 DVDs with NNEDI3-128 Luma doubling, NNEDI3-32 Chroma doubling, NNEDI3-64 Luma quadrupling, NNEDI3-32 Chroma quadrupling, SSIM2D 100AR downscaling, all the upscaling refinements enabled (with light settings) except for LumaSharpen and the render time is about 37mS. :devil:

Now that sounds like so much fun! Makes us curious if you can post some example pictures, would be fun seeing the results :)

Can't wait for 1080 laptops to emerge! :cool:



BTW, with everyone complaining about performance drops after windows updates, is there a solution to it? It did not happen to me yet, but my windows build is custom made, just want to know for future references.

[Betroz]

I noticed a ~3ms higher rendertime after the Win10 update. Same nVidia driver and settings in MadVr.

[kalston]

This GTX 1080 is a lot of fun. I've got a 1080p display. I can play back p24 DVDs with NNEDI3-128 Luma doubling, NNEDI3-32 Chroma doubling, NNEDI3-64 Luma quadrupling, NNEDI3-32 Chroma quadrupling, SSIM2D 100AR downscaling, all the upscaling refinements enabled (with light settings) except for LumaSharpen and the render time is about 37mS. :devil:

Why not Luma 256? :D I watch DVDs on a 1440p display with luma 256/chroma 32 on a 980 ti and my render times are more than fine :) (image upscaling is JincAR).
I only use ordered dithering and don't bother with upscaling refinements though.

Such a joy to watch DVDs like this!
I remember before I found madvr how I hated watching DVDs on a computer screen @1080p and up.

[leeperry]

BTW I recently updated to the latest version of PotP and when I seek several times in a row(even in windowed mode) I get this box (http://www.imagebam.com/image/af9071498496016) showing up for a split second. Log file available here (https://mega.nz/#!Cxp11QQS!4P7JCT_utVmiv3CHoSc9MHEv64O8mETQrsgyboc4cXI).
00000787.958 Render Get3DOffset for frame 10: 0
00000787.979 Render CRendering::ExecuteStep(self: 0C86D888);
00000787.983 Render beginning texture resampling scene failed
Still running W7SP1/13.12 HD7850, any idea what's wrong please?

:thanks:

[Stereodude]

Now that sounds like so much fun! Makes us curious if you can post some example pictures, would be fun seeing the results :)
I don't think I can take screenshots since 10-bit mode only works in FSE.

BTW, with everyone complaining about performance drops after windows updates, is there a solution to it? It did not happen to me yet, but my windows build is custom made, just want to know for future references.
I'm using Windows 8.1 x64, so I don't have that issue.

Why not Luma 256? :D I watch DVDs on a 1440p display with luma 256/chroma 32 on a 980 ti and my render times are more than fine :) (image upscaling is JincAR).
I only use ordered dithering and don't bother with upscaling refinements though.
I figured quadrupling was better than using really high doubling settings. I haven't experimented that much yet. I only got the card yesterday. I'm using ordered dithering since I'm using 10-bit output. Upscaling refinement is very nice with DVD.

[Warner306]

Can someone explain how 3D LUTs interact with madVR's HDR10 mapping algorithms?

How to set up DisplayCAL to create a HDR to SDR 3D LUT (http://www.avsforum.com/forum/139-display-calibration/1471169-madvr-argyllcms-146.html#post45777857)

Should madVR's mapping be unchecked in this case?

[ryrynz]

I noticed a ~3ms higher rendertime after the Win10 update. Same nVidia driver and settings in MadVr.
That's meaningless without knowing what clock rate the card is operating at. FWIW I didn't see any change on my 960.

[fhoech]

Can someone explain how 3D LUTs interact with madVR's HDR10 mapping algorithms?

How to set up DisplayCAL to create a HDR to SDR 3D LUT (http://www.avsforum.com/forum/139-display-calibration/1471169-madvr-argyllcms-146.html#post45777857)

Should madVR's mapping be unchecked in this case?
They shouldn't be used both at the same time. The only way you can currently use a 3D LUT for HDR tonemapping is by using madVR 0.9.21 or earlier as the current version disables 3D LUTs when disabling HDR processing, or by removing the HDR metadata from HDR media (e.g. by re-encoding, this is what I did), or using a version of LAV filters that doesn't pass HDR metadata to madVR.

[CarlosCaco]

They shouldn't be used both at the same time. The only way you can currently use a 3D LUT for HDR tonemapping is by using madVR 0.9.21 or earlier as the current version disables 3D LUTs when disabling HDR processing, or by removing the HDR metadata from HDR media (e.g. by re-encoding, this is what I did), or using a version of LAV filters that doesn't pass HDR metadata to madVR.

Or enabling ffdshow between lav and madvr in the render chain às ffdshow don t read/pass hdr metadata

[har3inger]

A long time ago I found a bug where increasing present frame queue caused video to be delayed from audio proportional to how long the queue was. Today I found the fix: turn off "use separate device for presentation". I now have 12 frames in queue and no a/v de sync.

Not sure why this happens on my laptop, but if others run into this problem, this is a fix to try.

[ionutm80]

Hi guys, just found an annoying issue after the Win 10 anniversary update on an Intel NUC (HD 4400 Graphics, latest driver 15.40.25.4463 / 17.06.2016).
With D3D11: display rate is shown correct at 23,976002 but composition rate is 24,00 !? and I have heavy "hanging" / not watchable.
With D3D9: display again correct at 23,976002 and also composition rate is 23,976 with 1 frame drop every 1.2-3 hours.

I have re-installed LAV 0.68.1, madVR reset settings to default, re-instal Intel HD 4400 driver.

Any idea what's wrong with D3D11? Before updating Win 10 everything was ok.

Thanks in advance.

[ryrynz]

Any idea what's wrong with D3D11? Before updating Win 10 everything was ok.


Composition rate shouldn't change when switching between Direct3D versions.. does this happen with all content for you at 24Hz?

D3D11 has never been usable for me regardless of OS on my HD4000, I get glitches on the display every few seconds running the latest driver on W10A.
Perhaps the easiest solution for you is to simply revert to 10586.

[ionutm80]

Hi, yes all 23,976 content (h.264 blu-ray rips). Strangely, it does not happen with 25,000 content (mpeg2 DVD rips).
And it all started to happen after Win 10 update ...
The render time are also more than reasonable <5ms with spikes (?!) of 14ms.

[Siso]

Could someone explain to me what "present" really means, it is under rendering, with D3D9 it is around 0,08 ms, with D3D11 it is around 0,34 ms. Which one is better?

[chros]

And it all started to happen after Win 10 update ...
Off: can you disable auto update in Win10 completely?

[huhn]

0.08 ms is "better" but it doesn't matter.

it is the time that is needed to present the frame. so as long as you are way under the refresh rate it doesn't really matter.

[ryrynz]

Off: can you disable auto update in Win10 completely?

Just disable the service. Don't ask anything more about it here, if you don't know what I mean then Google it.

[Siso]

0.08 ms is "better" but it doesn't matter.

it is the time that is needed to present the frame. so as long as you are way under the refresh rate it doesn't really matter.

Thank you.