madVR - high quality video renderer (GPU assisted) [Archive] - Page 649

HelmedHorror

5th August 2015, 23:23

Hi, I have a 2.37:1 monitor (as opposed to 1.77:1, aka 16:9) which has a resolution of 2560x1080. When I play Blu-ray content on my computer this creates a unique sort of problem. You see, most movies are shot in 2.39:1 ("anamorphic") aspect ratio, which ends up being 1920x800 resolution on Blu-ray (it rounds to 2.4:1 instead of 2.39:1 which would be 1920x803). But there's one caveat: for some stupid reason, Blu-ray video content has hard-coded letterboxes (usually 140 pixels each in height) to make the rendered image equal 1920x1080 and thus fit most monitors which are 16:9. (In reality, media players automatically intelligently add their own letterboxes/pillarboxes thus negating the need for hardcoded letterboxes, but that's a separate gripe.)

So, as you may by now be able to imagine, my problem can be demonstrated with this illustration (http://i.imgur.com/bcEpE1u.png) I made.

As you can see, Blu-ray's hardcoded letterboxes are treated as content to render as if it's part of the movie itself. Of course, any human knows that the letterboxes aren't part of the movie itself, but the media player doesn't know that. So my media player (MPC-HC) treats the video as if it's 1920x1080 because it's counting the two 140 pixel letterboxes as part of the video. Since my screen itself is only 1080 pixels high, the media player thinks it can't automatically zoom in to fill the screen with the full 2.40:1 movie.

By itself, this isn't a big deal. I can just manually zoom in with MPC-HC. Since my monitor is roughly the same aspect ratio (2.37:1) as anamorphic movies (2.40:1) a zoomed in Blu-ray is almost a perfect fit with my monitor, resulting in almost no black bars! Fantastic, right? Well, this is where madVR comes in.

MadVR doesn't seem to have an option to force image doubling. MadVR only allows image doubling "if needed" or "if at least 1.2x upscaling". The problem is that madVR doesn't understand that my image is being upscaled. Why? Because, again, it thinks the hardcoded letterboxes are part of the movie itself and so it treats the whole image as 1920x1080 when in reality it's 1920x800.

The reason this is not a problem for people with 16:9 monitors is because they really can't zoom in like I can, at least not without cutting off a huge chunk of the side of the video. But not all of us have 16:9 monitors.

Is there any way I can work around this problem other than being forced to disable image doubling?

dansrfe

5th August 2015, 23:29

@HelmedHorror

The letterbox black borders are part of the video frame. There's no thinking going on. It's a direct consequence of the rigidity of the Bluray spec's resolution requirement.

Second, if you want it to double on an (approximate) source/monitor matching aspect ratio then you can specify that via an image doubling profile.

HelmedHorror

5th August 2015, 23:40

@HelmedHorror

The letterbox black borders are part of the video frame. There's no thinking going on. It's a direct consequence of the rigidity of the Bluray spec's resolution requirement. I know, that's exactly what I said. Multiple times, in fact...

Second, if you want it to double on an (approximate) source/monitor matching aspect ratio then you can specify that via an image doubling profile.

Can you elaborate? I only see the option to double the image if it can be upscaled (which it is when I zoom in, but again madVR doesn't know that).

MistahBonzai

6th August 2015, 00:13

HelmedHorror

6th August 2015, 00:19

Perhaps OT but I'll chime in nonetheless. There are apps which can make decisions regarding 'black bars' versus content and crop the bars as needed. SVP comes to mind although I seriously doubt you would be interested in the frame smoothing. The point being that the crop functionality can be incorporated with playback *if* you are one who doesn't mind going the AviSynth path...

But, correct me if I'm wrong, there's no way to remove the black bars as part of the encoded video without re-encoding the video and thus losing quality. I shouldn't have to lower the quality of my Blu-rays just to get madVR to double the image resolution. Indeed, that would probably make the picture look worse than before, even with image doubling.

huhn

6th August 2015, 00:57

But, correct me if I'm wrong, there's no way to remove the black bars as part of the encoded video without re-encoding the video and thus losing quality. I shouldn't have to lower the quality of my Blu-rays just to get madVR to double the image resolution. Indeed, that would probably make the picture look worse than before, even with image doubling.

ffdshow can do it. just run it as an image processor and try to get the black bar removing working.

a blu ray frame is 1920x1080 so the issue is the BD disc nothing else.

vivan

6th August 2015, 01:09

The problem is that madVR doesn't understand that my image is being upscaled. Why? Because, again, it thinks the hardcoded letterboxes are part of the movie itself and so it treats the whole image as 1920x1080 when in reality it's 1920x800.Em... What? If you select "Video Frame->Touch Window From Outside" MPC-HC will tell madVR to use 0, -360, 2560, 1440 window (2560x1440 window that is shifted up by 360 pixels - essentially cropped by 360 pixels from top and the bottom).

So, can you select this option, enable OSD (Ctrl+J), take an screenshot and explain what is exactly wrong with what you're getting?

Is there any way I can work around this problem other than being forced to disable image doubling?Image doubling only tells madVR how to scale, it doesn't affect output window size in any way.

UPD: AFAIK automatic cropping is planned, I remember madshi was collecting samples some time ago.

huhn

6th August 2015, 01:29

there is no guarantee a 21/9 frame on a 16/9 frame has to be centred directly in the middle it's know this isn't the case always so touch from outside is not a same "fix".

MistahBonzai

6th August 2015, 01:36

Ah yes.. I had forgotten about the MPC-HC scaling options. There is also a zoom to Ultra Wide Screen option. Not sure of the aspect ratio but it's considerably wider than 16:9.

HelmedHorror

6th August 2015, 01:39

Em... What? If you select "Video Frame->Touch Window From Outside" MPC-HC will tell madVR to use 0, -360, 2560, 1440 window (2560x1440 window that is shifted up by 360 pixels - essentially cropped by 360 pixels from top and the bottom).

So, can you select this option, enable OSD (Ctrl+J), take an screenshot and explain what is exactly wrong with what you're getting?

Hmm, you're totally right. That's weird, because I (thought) I clearly remember it not working when I tried it a few days ago. I even went to all this effort to draw that illustration and register and wait 5 days and make my post here. Derp.

Okay, well then, ignore my complaint :p

But there's another question I have, now that I've gotten image doubling to work. I know it works because the render time goes up by an order of magnitude when I image double. The thing is, even with 2560x1080 pixels and staring right at the screen in a dark room in a perfect Blu-ray rip of a still scene, I can't discern any difference in quality with image doubling on. Is it designed more for low quality content? The only thing that seems to change is that it seems to remove the upscale refinement sharpening that I also set. So, in that sense, it looks worse than with image doubling off. Am I missing something, or is image doubling just not really made for +1/3rd upscaled 25Mbps 1080p Blu-ray content?

Image doubling only tells madVR how to scale, it doesn't affect output window size in any way.

UPD: AFAIK automatic cropping is planned, I remember madshi was collecting samples some time ago.

Yeah, I understand, that's totally fine because I can just resize the output window on my own. I was just (wrongly) concerned that it messed up madVR's ability to image double.

Thanks for setting me straight!

there is no guarantee a 21/9 frame on a 16/9 frame has to be centred directly in the middle it's know this isn't the case always so touch from outside is not a same "fix".

Yeah, in cases like that I just use the Numpad keys to manually zoom in to the closest I can. Numpad9 to zoom in, Numpad1 to zoom out.

Ah yes.. I had forgotten about the MPC-HC scaling options. There is also a zoom to Ultra Wide Screen option. Not sure of the aspect ratio but it's considerably wider than 16:9.

I just checked, and it's weird. It sets it to a 2:1 aspect ratio, specifically 2974x1492. Really odd...

foxyshadis

6th August 2015, 01:40

Off-topic (and dangerous) Windows 10 crap moved (https://forum.doom9.org/showthread.php?t=172441).

jobypeter

6th August 2015, 02:32

use copyback 10 bit native DXVA is not really working with directshow.

SO You mean that if I played 10BIT video with any codec the MADVR is not functional in DXVA mode(copyback & native)?

vivan

6th August 2015, 03:15

But there's another question I have, now that I've gotten image doubling to work. I know it works because the render time goes up by an order of magnitude when I image double. The thing is, even with 2560x1080 pixels and staring right at the screen in a dark room in a perfect Blu-ray rip of a still scene, I can't discern any difference in quality with image doubling on. Is it designed more for low quality content? The only thing that seems to change is that it seems to remove the upscale refinement sharpening that I also set. So, in that sense, it looks worse than with image doubling off. Am I missing something, or is image doubling just not really made for +1/3rd upscaled 25Mbps 1080p Blu-ray content?Well... Image doubling is not the point of that menu section. It's the opposite - the point is to be able to use algorithms that can't scale to arbitrary resolutions.

madVR has several scaling algorithms - most of them can be used to scale to any resolution. However some algorithms are special (nnedi and that new one, suber-xbr) - they could only scale in one direction and only to the double of the resolution.

NNEDI was designed to be used as spatial deinterlacer. In interlaced video one field has only half of the pixels - 1st line of the pixels, then 3rd, then 5th and so on. To deinterlace video you have to find missing pixels rows - effectively doubling the resolution of the field, to match frame size. Which is just scaling it vertically. In the end we got suprisingly effective algorithm that just destroyed everything else, not just as spacial deinterlacer - but also as upscaler.

But problem remains - it can only double resolution in one direction. Okay, you can use it twice. But what about when you need smaller scaling factor? You can double resolution and then downscale. Which is what madvr does.
NNEDI is very, very slow and using it all the time doesn't worth it (with small scaling factors improvement over other scalers is not noticeable). That's why we have options to decide when to use it.

NNEDI does shine on high quailty SD content, but only because that content have to be scaled by a very large factor.
Does doubling 1080p and downscaling it to 1440p worth it (33% upscale)? Probably not.
Some comparisons: http://forum.doom9.org/showthread.php?p=1726378#post1726378

karl_lillevold

6th August 2015, 03:16

I apologize if this has been asked and answered in one of the preceding 1622 pages.. I searched via Google, but did not read all pages..

After I installed Windows 10: whenever the screensaver has gone on, and I take MPC-HC with latest madVR full-screen, screen goes black, and I have to go back to normal size, and then back to full-screen. The problem is only with exclusive mode, so I can always switch it off - I am not even sure exclusive provides much benefit on my system. I have just always been using it.

*) I have a plasma HDTV, which is why my screen saver goes on after just 1 minute... and I see this problem frequently.

huhn

6th August 2015, 06:13

chros

6th August 2015, 09:29

madVR has several scaling algorithms - most of them can be used to scale to any resolution. However some algorithms are special (nnedi and that new one, suber-xbr) - they could only scale in one direction and only to the double of the resolution.

NNEDI was designed to be used as spatial deinterlacer. In interlaced video one field has only half of the pixels - 1st line of the pixels, then 3rd, then 5th and so on. To deinterlace video you have to find missing pixels rows - effectively doubling the resolution of the field, to match frame size. Which is just scaling it vertically. In the end we got suprisingly effective algorithm that just destroyed everything else, not just as spacial deinterlacer - but also as upscaler.

But problem remains - it can only double resolution in one direction. Okay, you can use it twice. But what about when you need smaller scaling factor? You can double resolution and then downscale. Which is what madvr does.
NNEDI is very, very slow and using it all the time doesn't worth it (with small scaling factors improvement over other scalers is not noticeable). That's why we have options to decide when to use it.

Thank you for the explanation I didn't have a clue about these things.

Manni

6th August 2015, 09:36

I'm here to report a bug: madVR isn't working with JR Media Center since a couple of versions. I'm trying to track down the issue to the version when it started crashing. It crashes just when you hit play button. No problem with MPC-HC so far. Thanks in advance.

I noticed a similar problem using post .16 MadVR builds on post 14.12 AMD drivers with MPC-Be (using KCP Black 0.6, which includes .16 MadVR, and updating manually up to and including .21).

MPC-BE crashes almost every time with 15.7 or 15.7.1.
Reverting to 14.12 (clean install) lowers the amount of crashes significantly, but they are still present.
Reverting to MadVR .16 eliminates almost completely the crashes (I still get an occasional one from time to time).
Playing mostly BD Folder on Win 8.1 HD7870 on 1080p display, so only using NNEDI3 chroma upscaling 32 neurons.

I'm not using FSE in case this matters (takes too long to switch to/from it on my setup) but driver and MadVR set to 10bits (the X500 has true 10bits panels apparently) and d3d11 so I can test 10bits output simply by enabling/disabling FSE. To date, even when I could see banding on screen, I haven't found a situation where 10bits would improve over 8bits on my display so I routinely use 8bits for the faster interface.

Apart from the above crashes (which aren't necessarily induced by MadVR), I have a small request for Madshi: as there are always a couple of dropped/repeated frames or glitches on start, would it be possible to add an automatic reset after about 1s play? That way, when we look at the info, we'll know if the issues are real or "normal", even if we forget to reset the stats right after the start of playback.

One last thing, I've noticed lately, using .16, that some frames drops aren't reported. I could clearly see some significant drops while playing a 1080p BD Folder and the stats were clean. Is this a known issue? By the way, I almost never have drops with .16 (I have more with more recent versions), so the drops are a one-off issue, not a constant problem, at least with .16.

Keep up all the good work, the improvements/new features in the last few months are really impressive, hopefully I'll be able to use post .16 builds without any issue on my setup soon and catch up with the latest. As .21 is required in Calman now, it would be good to be able to do so.

blindbox

6th August 2015, 13:34

there is no guarantee a 21/9 frame on a 16/9 frame has to be centred directly in the middle it's know this isn't the case always so touch from outside is not a same "fix".

SVP Manager's avisynth scripts have an auto-black border removal, so that might help you as a starting point. Perhaps there are other AVISynth scripts that would do only black border removal. Since MadVR operates after ffdshow raw video, it should work out fine. I haven't done anymore research on this, since hardcoded black borders rarely happen to me.

MSL_DK

6th August 2015, 14:33

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.

I don't know if this has something to do with my problem, and i can't check it, as my monitor is at service.

https://www.thurrott.com/wp-content/uploads/2015/07/display-settings.jpg

Mine was scaled to 150%

huhn

6th August 2015, 15:13

SO You mean that if I played 10BIT video with any codec the MADVR is not functional in DXVA mode(copyback & native)?

DXVA native has an issue with 10 bit. and only 1 card can decode anything with 10 bit and that's HEVC 10 bit nothing else works with 10 bit and copyback is used.

madVR doesn't care about input bit deep it supports up to 16 bit and all colorspaces lavfilter can output. the issue is DXVA >native< DXVa native has a different way to deliver the data to the renderer and 10 bit doesn't really work with directshow in general.

just use DXVA copyback and forget about native. DXVA native is currently reduction picture quality on nvidia it not supposed to do that but that's currently the case.

and there a ways to encode 8 bit h264 in a way DXVA will not work.

SVP Manager's avisynth scripts have an auto-black border removal, so that might help you as a starting point. Perhaps there are other AVISynth scripts that would do only black border removal. Since MadVR operates after ffdshow raw video, it should work out fine. I haven't done anymore research on this, since hardcoded black borders rarely happen to me.

ffdshow has an option for this too like I said earlier

XMonarchY

7th August 2015, 01:13

NVidia 353.71 Windows 10 drivers for laptops came out. Is there a way to check whether D3D9-OpenCL interop is supported without modifying and installing these drivers on my desktop GTX 980?

karl_lillevold

7th August 2015, 02:19

exclusive mode is not needed for most user you should be totally fine without it.

could you post the used GPU, driver and general madVR settings like overlay mode d3d11 mode and things like that so people can try to reproduce the issue?
Thanks for your reply! You are right, playback without exclusive mode is pretty great - I am unable to tell any difference (no detailed analysis).

GPU: AMD Radeon 6450 (this is a quiet fanless HTPC). YCbCr 4:4:4. Catalyst 15.7.1.
All my general, windowed, exclusive mode settings are default, AFAIK.
General: only "Use a separate device for presentation" is checked.
"Use Direct3D 11 for presentation" unchecked.
I am not well versed in video rendering details, so I don't know which settings are most relevant..

EDIT: to make sure all my settings were in fact default, I ran the madVR .bat script to reset to default. Except for my scaling options, the only settings that changed were : exclusive mode was re-enabled, and "Use separate device for presentation" became un-checked. I don't know how that had become enabled in my settings, but in any case, the problem I described remained the same, with exclusive mode enabled. Not a big deal - as exclusive is not needed, but it is still the default.

huhn

7th August 2015, 05:11

Thanks for your reply! You are right, playback without exclusive mode is pretty great - I am unable to tell any difference (no detailed analysis).

GPU: AMD Radeon 6450 (this is a quiet fanless HTPC). YCbCr 4:4:4. Catalyst 15.7.1.
All my general, windowed, exclusive mode settings are default, AFAIK.
General: only "Use a separate device for presentation" is checked.
"Use Direct3D 11 for presentation" unchecked.
I am not well versed in video rendering details, so I don't know which settings are most relevant..

EDIT: to make sure all my settings were in fact default, I ran the madVR .bat script to reset to default. Except for my scaling options, the only settings that changed were : exclusive mode was re-enabled, and "Use separate device for presentation" became un-checked. I don't know how that had become enabled in my settings, but in any case, the problem I described remained the same, with exclusive mode enabled. Not a big deal - as exclusive is not needed, but it is still the default.[/QUOTE]

there shouldn't be a visual differences. FSE is a good way to avoid issue with the windows desktop composition which was an huge issue with windows 7 which is still the most used OS so it's good that exclusive mode is default.

the best mode by far is overlay but AMD can't do this...

there is usually nothing to win by using YCbCr as output. madVR is feeding the driver RGB anyway. so you lose at least the potential for less conversations.

leeperry

7th August 2015, 07:09

there is usually nothing to win by using YCbCr as output. madVR is feeding the driver RGB anyway. so you lose at least the potential for less conversions.
It would be great if we could check somehow coz after comparing DVI & HDMI straight off my 7850 using 13.12 on W7, I strongly suspect that RGB 4:4:4 isn't quite as pass-through over HDMI as it is over DVI.

Of course I was comparing a poorly shielded 8ft HDMI cheapo OEM cable against a quad shielded very stiff 4ft branded DVI cable made of solid copper wires, but cable build quality would supposedly not matter with TMDS :devil:

kopija

7th August 2015, 07:32

FSE is a good way to avoid issue with the windows desktop composition which was an huge issue with windows 7 which is still the most used OS so it's good that exclusive mode is default.

My understanding is that Aero is gone in W10, so there is no need for disabling desktop composition in madVR?

Sunset1982

7th August 2015, 08:02

One last thing, I've noticed lately, using .16, that some frames drops aren't reported. I could clearly see some significant drops while playing a 1080p BD Folder and the stats were clean. Is this a known issue? By the way, I almost never have drops with .16 (I have more with more recent versions), so the drops are a one-off issue, not a constant problem, at least with .16.

I also noticed Framedrops since MadVR .16 and later. They're clearly visable but not shown in OSD. They are random and I can not reproduce them.

Using Windows 7 (x64), AMD R9 380, Catalyst 15.7.1, MPC-HC, LAV, MADVR ans XYSubFilter (all x64)

MadVR is running Fullscreen Exclusive Mode, Needi3 32, both D3D 9 and 11 show the glitches.

Same with my nVidia GTX970.

I'm now back to 0.87.21 and everything works great...

Manni

7th August 2015, 08:44

I also noticed Framedrops since MadVR .16 and later. They're clearly visable but not shown in OSD. They are random and I can not reproduce them.

Using Windows 7 (x64), AMD R9 380, Catalyst 15.7.1, MPC-HC, LAV, MADVR ans XYSubFilter (all x64)

MadVR is running Fullscreen Exclusive Mode, Needi3 32, both D3D 9 and 11 show the glitches.

Same with my nVidia GTX970.

I'm now back to 0.87.21 and everything works great...

Same here as reported a few posts above. Didn't try to go back to 0.87.21, but might if unreported dropped frames persist, thanks for identifying the last build where the issue doesn't appear.

Do you have some MPC-HC crashes with 15.7.x? They mostly stopped here reverting to 14.12.

jobypeter

7th August 2015, 10:15

dxva native has an issue with 10 bit. And only 1 card can decode anything with 10 bit and that's hevc 10 bit nothing else works with 10 bit and copyback is used.

Madvr doesn't care about input bit deep it supports up to 16 bit and all colorspaces lavfilter can output. The issue is dxva >native< dxva native has a different way to deliver the data to the renderer and 10 bit doesn't really work with directshow in general.

Just use dxva copyback and forget about native. Dxva native is currently reduction picture quality on nvidia it not supposed to do that but that's currently the case.

And there a ways to encode 8 bit h264 in a way dxva will not work.

thanks

James Freeman

7th August 2015, 10:41

I also noticed Framedrops since MadVR .16 and later. They're clearly visable but not shown in OSD. They are random and I can not reproduce them.

I'm now back to 0.87.21 and everything works great...

87 or 88 ???

Manni

7th August 2015, 12:16

87 or 88 ???

Good question. I assumed it was 87.21 as there are still dropped frames in my setup (these ones reported) in 88.21, in fact more than in 88.16, but it would be good to know if the OP meant 87.21 as I assumed or 88.21.

Sunset1982

7th August 2015, 12:44

87 or 88 ???

87.

It seems that 88. Versions has introduced this unreportet stuttering, but I haven't tested it that much with lower 88... builds.

Do you have some MPC-HC crashes with 15.7.x? They mostly stopped here reverting to 14.12.

I have had NO crashes with version 15.7.1 and windows 10.

Before I upgraded to win 10 I had a setup with win 7 and Catalyst Omega 14.12 which worked flawless too.

Manni

7th August 2015, 13:44

87.

It seems that 88. Versions has introduced this unreportet stuttering, but I haven't tested it that much with lower 88... builds.

I have had NO crashes with version 15.7.1 and windows 10.

Before I upgraded to win 10 I had a setup with win 7 and Catalyst Omega 14.12 which worked flawless too.

Thanks for the confirmation. I missed you were on Win10 already. I guess there is no way back to 14.12 for you :).

I might give 87.21 a try if the next 88.x build doesn't solve the dropped frames.

chros

7th August 2015, 13:46

the best mode by far is overlay but AMD can't do this...
Can you tell me more about this? I use FSE for a long-long time.
Thanks

xabregas

7th August 2015, 14:12

the best mode by far is overlay but AMD can't do this...Can you tell me more about this? I use FSE for a long-long time.
Thanks

He must be talking about his own best mode, as i see from his gpu the best mode may be overlay, still with a good gpu nothing is better than Full Screen Exclusive mode and whoever says u dont need it is lying. :P

James Freeman

7th August 2015, 14:39

It seems that 88. Versions has introduced this unreportet stuttering, but I haven't tested it that much with lower 88... builds.

As far as I know, if you are using D3D9 nothing has changed between 87 and 88.

Manni

7th August 2015, 15:07

As far as I know, if you are using D3D9 nothing has changed between 87 and 88.

Thanks for this, I'll try disabling d3d11 with 88.16 and I'll report if it solves anything. It's indeed the main thing I've changed compared to pre-dropped frames.

If it does, I might even be able to come back up to 88.21.

The only reason I selected d3d11 in recent builds was to be able to enable/disable FSE (therefore 10bits) in one click. As I don't really use FSE/10bits, going back to d3d9 should be pretty painless.

huhn

7th August 2015, 16:30

It would be great if we could check somehow coz after comparing DVI & HDMI straight off my 7850 using 13.12 on W7, I strongly suspect that RGB 4:4:4 isn't quite as pass-through over HDMI as it is over DVI.

AMD is outputting 10 bit by default unlike DVI so there is some about this but still less conversation than YCbCr.

i hope you are aware of the disable dither and forced 8 bit output mode possible with newer driver. a RGB capture card should be able to verify this.

Of course I was comparing a poorly shielded 8ft HDMI cheapo OEM cable against a quad shielded very stiff 4ft branded DVI cable made of solid copper wires, but cable build quality would supposedly not matter with TMDS :devil:

the results of a bad signal are very visible

My understanding is that Aero is gone in W10, so there is no need for disabling desktop composition in madVR?

all windows version since vista have "aero" it is simply not called aero any more and forced since windows 8. it's just the desktop composition which is very important except for FSE and overlay they work totally fine without it.

He must be talking about his own best mode, as i see from his gpu the best mode may be overlay, still with a good gpu nothing is better than Full Screen Exclusive mode and whoever says u dont need it is lying. :P

so why i'm lying? why do all people need FSE or are lying?

overlay mode is error resilient and even in window mode and can make the best use of GPU gamma ramps.

there is nothing wrong with FSE in general. if you have 2 displays you will notice some issue with FSE in general and FSE only works in fullscreen unlike overlay.

Asmodian

7th August 2015, 19:08

Personally I like overlay mode the best too.

baii

7th August 2015, 19:22

I think it is GPU driver fault, but I do get presentation glitches and frame drop (any content) in d3d11 windowed. Where d3d9 windowed have weired high render time and frame drop. This is same on both 88.16 //.21. Perfect playback on FSE.

15.7 AMD driver

HelmedHorror

8th August 2015, 00:16

Small bug I just noticed: When LumaSharpen is enabled within image enhancements and you change the radius value of it, madVR does not allow the Apply button to be pressed, as if it thinks no change has taken place. However, if you change teh strength or clamp values it will notice the change and let you press Apply.

Sunset1982

8th August 2015, 07:13

got my 4K TV yesterday. I have tested some things and would like to ask some questions:

- madvr is only running in windowed mode instead of fullscreen exclusive, when used to upscale to UHD, whatever I set it up to.

- is there a way to get the stats windows bigger when using 4K mode?

- what do I need to get 10 Bit mode working with 4k?

- since needi3 isnt working correct in win10 atm, what settings do you guys use to get best possible quality? Super-XBR S:100?

James Freeman

8th August 2015, 07:22

I also use Overlay mode.
It completely bypasses any Aero, Vsync, and Direct3D "collision" between software and OS.

Overlay: MadVR -> GPU -> Display.
No Overlay: MadVR -> GPU -> OS -> Display.

It resembles Exclusive mode with audio, in which the exclusive program takes control over the audio card and bypasses the OS mixer.
In Shared mode the OS mixes all the sounds from all software to the output.

In video Overlay the exclusive portion is only the size of the rendered video window.
That's is why when taking a screenshot there is a "hole" where the video should have been.

This is another plus for Nvidia users.

Akeno

8th August 2015, 07:49

- since needi3 isnt working correct in win10 atm, what settings do you guys use to get best possible quality? Super-XBR S:100?

Super-xbr combined with SuperRes appears to be the next best choice for upscaling. Some testing has shown that with SuperRes, super-xbr and nnedi3 look near identical while super-xbr is much lighter on the gpu.
Refer to this (http://forum.doom9.org/showthread.php?p=1730855#post1730855) post and the posts following for examples.

Sunset1982

8th August 2015, 09:09

Super-xbr combined with SuperRes appears to be the next best choice for upscaling. Some testing has shown that with SuperRes, super-xbr and nnedi3 look near identical while super-xbr is much lighter on the gpu.
Refer to this post and the posts following for examples.

thx for you info.

What settings in sharpness and superRes would you suggest?

Manni

8th August 2015, 09:25

Just a quick follow-up regarding my dropped frames.

I re-installed 88.21 and enabled FSE (10bits, d3d11).

I had no dropped (reported or not), duplicated frames of pres glitches. Playback was flawless.

This is only over a 2 hour movie, and I haven't tested if I still had MPC-BE crashes with this configuration.

If I do, I might try to disable FSE and switch back to D3D9 and see how it goes.

It would be nice to be able to use the latest build.

If either looks stable, I'll try to re-install 15.7.1.