madVR - high quality video renderer (GPU assisted)

[bcec]

I suspected that SM wouldn't have enough of an impact to meaningfully affect nnedi3 scaling capability. As to the other point I guess I don't know how nnedi3 upscaling works then. Isn't 3840x2160 exactly 4 times the resolution of 1920x1080?

After the latest nVidia updates, my gtx970 handles 1080p->2160p @24Hz no problem with NNEDI3@64 (at around 35ms).

[vomanci]

Wht is more important (correct) regarding MadVR, setting my tv at 60hz with SM and have 4:4:4 or using 23hz, 24hz etc without SM but 4:2:2 ?

[huhn]

Wht is more important (correct) regarding MadVR, setting my tv at 60hz with SM and have 4:4:4 or using 23hz, 24hz etc without SM but 4:2:2 ?

that what looks best to your eye. like always.

if you don't have any problems using SM i would say that's a lot better choice in this case.

[iSunrise]

The ol' resolution debate..

64 is the best choice performance/quality wise. Going beyond 64 doesn't yield you much but uses a fair bit more processing power. 32 neurons is just okay IMO, so fair call wanting 64, it's what I'd gun for.
I don't want to split hairs, but going NNEDI3 with 16 neurons already gives more improvement/performance than anything else. 64 neurons is a very heavy additional investment, with still visible additional benefits (only under closer inspection), but it's also _a lot_ heavier on the GPU.

Maybe that has changed with the newest drivers, but when I last did my test, 64 neurons was maxing out my overclocked GTX 580 all by itself, when I could easily go 16 neurons for luma/chroma and have additional HQ settings enabled, which I definitely prefer.

Of course when going 1080p->2160p 64 neurons could be more of a sweet-spot, but it doesn't change the fact that I rather have NNEDI3 with 16 neurons, than no NEEDI3 at all.

[nevcairiel]

NNEDI3 16 can make the "special" NNEDI artifacts kind of obvious, which is why I don't like running it on a bunch of content. With 32 its all much better.

[iSunrise]

NNEDI3 16 can make the "special" NNEDI artifacts kind of obvious, which is why I don't like running it on a bunch of content. With 32 its all much better.
At least on the pixel-sensitive tests I've done (e.x. the screenshots that madshi provided) it was visible, but i was not able to discern differences at my typical viewing distance when viewing live content (amiga recordings, etc.), practically invisible with real movie content.

Of course, opinions and setups vary, just my 2 cents.

If you have the GPU power to spare, going 64 neurons definitely doesn't hurt. But doing 1080p to 2160p doubling with 64 neurons should be very heavy even for actual GPU configurations (highest TDP, GPU always under very heavy load).

[ADude]

I've been away from the thread for a long time.

Has the small change - to not upscale 1916x1076 and 1912x1072 to 1920x1080 (perhaps optionally), been implemented ?

If not, is this likely to be done in the near future (it is the only small change that would be very helpful) ?

Thanks !

[huhn]

I've been away from the thread for a long time.

Has the small change - to not upscale 1916x1076 and 1912x1072 to 1920x1080 (perhaps optionally), been implemented ?

If not, is this likely to be done in the near future (it is the only small change that would be very helpful) ?

Thanks !
not added. if you mean if there is an option to not scaled at all.

[Desbreko]

You can already do that in MPC-HC, at least, by using View > Video Frame > Normal Size.

[QBhd]

I've been away from the thread for a long time.

Has the small change - to not upscale 1916x1076 and 1912x1072 to 1920x1080 (perhaps optionally), been implemented ?

If not, is this likely to be done in the near future (it is the only small change that would be very helpful) ?

Thanks !

That's the job of the media player. madVR only does what it's told by the player.

QB

[Fullmetal Encoder]

No, it is exactly 2 times the resolution (https://forum.doom9.org/showthread.php?p=1710665#post1710665). ;)

I would wait if you need a GPU that can do NNEDI3 64 neuron doubling on 1080p, I think a Titan X would be able to do it but even that would be close. Why do you need NNEDI3 64 instead of 32 neurons?

I knew what I'd done there even before I got to your link. Semantics aside though what's relevant is an understanding of how NNEDI3 defines resolution for it's purposes. And it's good to know that for my purposes I won't need to be so demanding of it after all.

I've seen some of the screen shots that madshi has posted in the past of upscaling 1080p content with 64 neurons. And if I'm going to be investing so heavily in a 4k monitor I better be able to get at least that much quality out of it.

[XMonarchY]

there is something not working as intended. hard to say what but this shouldn't happen.
can you try the old FSE mode?

you get the old FSE mode by disabling present several frames in advance under exclusive mode settings.

That worked! So does that mean the new FSE feature "Present Several Frames in Advance" is bugged? This feature defaults to 8 frames ahead and considering my heavy settings, my rig cannot handle that many frames ahead. I assume other rigs can? Why would mine not be able to do? I have a very clean and optimized PC with nothing in background when I play films. All drivers are properly set up, etc.

[huhn]

That worked! So does that mean the new FSE feature "Present Several Frames in Advance" is bugged? This feature defaults to 8 frames ahead and considering my heavy settings, my rig cannot handle that many frames ahead. I assume other rigs can? Why would mine not be able to do? I have a very clean and optimized PC with nothing in background when I play films. All drivers are properly set up, etc.

it should work so maybe this is a bug. check if you run out of Vram first.

[Fullmetal Encoder]

After the latest nVidia updates, my gtx970 handles 1080p->2160p @24Hz no problem with NNEDI3@64 (at around 35ms).

That's interesting. How does it fare at 60hz?

What I find so attractive about this particular upgrade path is that it doesn't necessitate any further manipulations. The target RESOLUTION is an exact even multiple of the source. I currently have an R9 290 but by the time the monitor I'm interested in becomes a viable purchase the next generation of cards will probably be available as well.

[Asmodian]

A 35 ms rendering time means it is not fast enough for 30 fps content, let alone 60.

Smooth motion for 24 fps @ 60 Hz wouldn't add enough rendering time you would need to worry for 24 fps content (you have another 5 ms to work on each frame).

[flashmozzg]

Smooth motion for 24 fps @ 60 Hz wouldn't add enough rendering time you would need to worry for 24 fps content (you have another 5 ms to work on each frame).

I depends. It's barely (~1-2ms) noticeable for me. But I have very good cpu/gpu setup.

[Asmodian]

I depends. It's barely (~1-2ms) noticeable for me. But I have very good cpu/gpu setup.

Yes but we were taking about a GTX 970 so smooth motion at 60 Hz would be very easy for it. :)

[ADude]

(In regards to a discussion earlier today in the thread) Previously, madshi said he was going to support this at some point in the future, which is why I asked if it had been implemented, and if not, was it still planned for the near future:

I think this is a valid suggestion.

madVR should of course still respect the size requested by the player. Doing padding instead of resizing could be beneficial for quality (and performance). It could be a "Trade quality for performance" option to do padding when size difference is <16 pixels in both dimensions.

I haven't said this yet, but I did already plan to support something like this in a future version, together with some other related changes.

[bcec]

A 35 ms rendering time means it is not fast enough for 30 fps content, let alone 60.


Yup. I don't have any 1080p60 content, but for 1080p29.97, I have to use NNEDI3@32, which is at around 27ms. Note that 35ms for NNEDI3@64 and 27ms NNEDI3@32 both use otherwise the same settings: Jinc3AR chroma upscaling, Lancsoz4AR image upscaling, error diffusion 2 (and all trade quality for performance options unchecked).

It's worth mentioning that switching from ED2 to Ordered reduces rendering times by roughly another ~3-4ms, so you can reach to ~32ms for NNEDI3@64 but that gives me dropped frames for 29.97 content every once in a while, as it is a tad too close to the limit than I'd like. You can probably squeeze few frames more by switching to L3AR, and turning on some performance options, but I haven't really bothered as even at NNEDI3@32, 1080p->2160p upscaling looks fantastic.

[Xaurus]

Right, so I am still confused as to whether one should double chroma resolution or not. I am setting up 6 different profiles now and all of them have these settings in common:
chroma upscaling: jinc3 ar
image upscaling: lanczos4 ar
error diffusion 1
double luma always

But the question is, do I simply push luma doubling as high as possible and ignore chroma doubling, or do I enable both?

This is all for 720p & 1080p -> 2160p.

[bcec]

But the question is, do I simply push luma doubling as high as possible and ignore chroma doubling, or do I enable both?

I wouldn't worry about chrome doubling. If I remember correctly, even madshi mentioned that he might remove chroma doubling option, as the benefit of turning that on is quite moot and not worth the tax.

[Warner306]

Right, so I am still confused as to whether one should double chroma resolution or not. I am setting up 6 different profiles now and all of them have these settings in common:
chroma upscaling: jinc3 ar
image upscaling: lanczos4 ar
error diffusion 1
double luma always

But the question is, do I simply push luma doubling as high as possible and ignore chroma doubling, or do I enable both?

This is all for 720p & 1080p -> 2160p.

The result of that chain should be pretty impressive. But Luma Doubling is only needed for 2x scaling factors (1080p -> 2160p) and Luma Quadrupling for 3x or more (720p -> 2160P, which is 3.5x).

[Xaurus]

Thanks guys. I will ignore the chroma doubling.

Also, will check out Luma quadrupling for 720p and lower. :)

[Xaurus]

As for the whole reclock or not debate, I have been trying without it to run 64-bit madvr, but whatever I do, I get a frame drop at an interval equal to the vsync which is roughly every 41.67ms.

Can anyone explain how they can run without reclock, even when the monitor refresh rate is basically equal to the movie fps? As it sure doesn't work here, even if the refresh rate is just 0.002125% higher than the frame rate as seen in the below picture.

http://home.halden.net/mordor/madvr2.jpg

[huhn]

As for the whole reclock or not debate, I have been trying without it to run 64-bit madvr, but whatever I do, I get a frame drop at an interval equal to the vsync which is roughly every 41.67ms.

Can anyone explain how they can run without reclock, even when the monitor refresh rate is basically equal to the movie fps? As it sure doesn't work here, even if the refresh rate is just 0.002125% higher than the frame rate as seen in the below picture.

http://home.halden.net/mordor/madvr2.jpg

your clock deviation is 0.1+ % is this screen from 1 sec after start so saying nothing?
the file says it is true 24 HZ that pretty rare in my experience.
are you sure this file isn't 23p(24000/1001) with a wrong flag. getting a repeated frame every 42 sec would be totally normal in this case.

[iSunrise]

I wouldn't worry about chrome doubling. If I remember correctly, even madshi mentioned that he might remove chroma doubling option, as the benefit of turning that on is quite moot and not worth the tax.
While that was true for the past, it was quite a while ago and I can't remember seeing anyone doing appropriate tests with very high scaling factors, since 4K wasn't even available back then. We might have to revisit that.

Chroma upscaling indeed did have a lot higher impact though, due to the way the upscaling pipeline works.

[Xaurus]

your clock deviation is 0.1+ % is this screen from 1 sec after start so saying nothing?
the file says it is true 24 HZ that pretty rare in my experience.
are you sure this file isn't 23p(24000/1001) with a wrong flag. getting a repeated frame every 42 sec would be totally normal in this case.
I have no idea if that file was true 24 or not. The picture was taken after 1 minute (hence the frame drop).

In any case, here's a new picture (taken after 5 minutes) from a 23.976 file played when the screen is set at 23.976 as well.
http://home.halden.net/mordor/madvr4.jpg

So I'm getting a frame drop every ~1 hour, that is not going to cut it for me unless there's something I can do to increase it to "days" like when I use Reclock? If not, are people just tolerating the unavoidable frame drop at a certain intervals when not using Reclock?

[jmartinr]

So I'm getting a frame drop every ~1 hour, that is not going to cut it for me unless there's something I can do to increase it to "days" like when I use Reclock? If not, are people just tolerating the unavoidable frame drop at a certain intervals when not using Reclock?
If you think you can notice one frame drop per hour, I think you're overestimating yourself ;)

[nevcairiel]

So I'm getting a frame drop every ~1 hour, that is not going to cut it for me unless there's something I can do to increase it to "days" like when I use Reclock? If not, are people just tolerating the unavoidable frame drop at a certain intervals when not using Reclock?

You should actually watch a movie and judge based on the final drop count. The estimate given by madVR is nothing but that, a estimate, it has no guarantees of perfect accuracy.

[x7007]

HI! I'm getting video freezes when using MPC-HC 1.7.8 and MadVR latest build to scale to UHD. Video will play to a random point then freeze, but the audio continues to play. I try to engage task manager but all I have after a control+alt+delete is a blank screen. I try to use the control+alt+delete to shutdown the system but I get the same blank screen & no shutdown. I have to do a hard restart with the power switch.

My gear:
Windows 7 64 bit
16GB RAM
i7 3820 CPU
X79A-GD45-PLUS Motherboard
R9 280x AMD Video card

I've managed to rule out MPC-HC directly causing the error as I have switched renderers to EVR CP and played several long video files. I also do not get freezes when using WMP to playback files. My Madvr settings are set to NNEDI3 32 neurons for image doubling, nothing checked for quadrupling, Lancoz for image upscaling (down from Jinc 3 taps to see if that would prevent the freeze- it didn't) bicubic for chroma. I have deblocking set to mild and I have engaged in luma the anti-ringing filter.

I have gotten this error with both 32 bit and 64 bit madvr and also other builds of madvr going back several iterations.

My *guess* is that I'm overpowering the GPU on the R9 280x by getting everything up to 3820x2160. I didn't know if any of you have had similar issues.


I have similar issue with potplayer randomly when using the Internal Ffmpeg decoder, but I think the same happens randomly with Lav Filters Cuda .
The other issue is for some reason even if I stop Potplayer it still uses Cpu Usage of all cores.
Dunno what is causing that. i'm using NNEDI3 64 with Luma always 64

[balkerman]

So I'm getting a frame drop every ~1 hour, that is not going to cut it for me

Have you actually watched a movie and noticed the frame drops or are you solely relying on what madvr estimates....

I stopped relock also with madvr 64 after years of use. I use to notice them years ago before using reclock, but not anymore. Plus without reclock you get tovhave untouched bitstreamed HD audio, yes I notice the difference. I stopped using reclock for blu-rays about 8 months ago....

[iSunrise]

So I'm getting a frame drop every ~1 hour, that is not going to cut it for me unless there's something I can do to increase it to "days" like when I use Reclock? If not, are people just tolerating the unavoidable frame drop at a certain intervals when not using Reclock?
Your example is already very close to perfect. What you could try is to tinker with custom refresh rates and try to increase it even further. You could also try multiples so you have a bit more flexibility in upper ranges, but that would probably make not much of a difference.

Make no mistake, though, when you are actually watching a movie, having only 1 or 2 frames drop in 2-3 hours is very hard to detect unless you really want to search for it. And since madVR only estimates (it's actually a very accurate assumption in my experience), you probably won't notice it at all.

So, try it and then afterwards judge if it annoys you. Not before.

[aufkrawall]

It would give me a bad feeling if I knew there might be frame drops which I could perceive.

[iSunrise]

It would give me a bad feeling if I knew there might be frame drops which I could perceive.
Yes, but do you actually perceive them?

Say, you don't have any knowledge of higher compressed content and visually, it looks perfect to your eyes. Now suddenly, someone tells you that it's highly compressed. Now you know, there was information lost in the process, but does it actually matter when there is no actual difference for your eyes?

I know what you mean though, too much information is not necessarily a good thing, rather the contrary.

[Luv]

7/32,ZP10,LAV 0.65,madVR 87.19,HD 7790,Omega driver 14.12.

Hi,
has anybody had problems with deinterlacing 1080i using HD 7790,R7 260 or 260x?
The present and render queues won't fill up and I have subsequently hundreds of dropped frames in a matter of seconds.
I previously used an HD 4770 + Catalyst 12.4 with the exact same software player config and never observed such a catastrophy.

[Xaurus]

Yes, the frame drops are VERY easy to spot because it makes movement skip "into the future". So I was right in my assumption.
Anyway, it's fine, there's no need to jump to 64-bit just yet. Thanks for all your opinions.

[XMonarchY]

it should work so maybe this is a bug. check if you run out of Vram first.

OK, some updates:
- Both displays are 1080p
- vRAM usage does not go over 1400MB regardless of whether I use the new or the old FSE mode.
- New FSE mode seems to be working fine on my 120Hz monitor, which stays at 120Hz during film playback. Max stats remain at 24.5ms 99% of the time, but 250ms+ appears once in a while, without creating frame drops. Frames are dropped on my 60Hz TV, which goes into 23Hz mode for film playback. Max stats remain at 250ms+ 90% of the time, but 24.5ms appears 10% of the time.
- GPU usage does not go over 16% on 60Hz TV @ 23Hz during playback. CPU usage for madVR stays @ 0% and for MPC-HC 32bit does not cross 6%, regardless of FSE mode.
- GPU usage does not go over 20% on 120Hz monitor @ 120Hz during playback. CPU usage for madVR stays @ 0% and for MPC-HC 32bit goes up as high as 20%, regardless of FSE mode.
- RAM usage stays at 241MB, regardless of display and FSE mode.


I have some side questions:
- Assume FSE is set to old mode or not enabled at all and there are no frame drops on either display.
- Since CPU, vRAM, and GPU usages stayed so low, wouldn't that mean my system can handle more robust settings? For example, higher neurons for Chroma Upscaling and Chroma/Luma Image Doubling/Quadrupling should be possible, should they not?
- They aren't on my 60Hz TV @ 23Hz, but they are on my 120Hz monitor @ 120Hz. If I push Chroma Upscaling from 64 neurons to 128 neurons for 1080p content on my 60Hz TV @ 23Hz, I get frame drops, regardless of FSE mode. The same goes for sub-1080p content and Chroma/Luma Image Doubling/Quadrupling on my 60Hz TV @ 23Hz.

Why not? Is it due to being limited as far as render times go @ 23Hz and less limited @ 120Hz? Should it not be the other way around??? 16ms @ 60Hz (not sure about ms @ 23Hz) vs. 8ms @ 120Hz? It doesn't make sense to me...

[XMonarchY]

Have you actually watched a movie and noticed the frame drops or are you solely relying on what madvr estimates....

I stopped relock also with madvr 64 after years of use. I use to notice them years ago before using reclock, but not anymore. Plus without reclock you get tovhave untouched bitstreamed HD audio, yes I notice the difference. I stopped using reclock for blu-rays about 8 months ago....

Some of us are sensitive and continue seeing de-synchronization between audio and video without ReClock or JRiver Media Center's similar feature. It does reduce immersion for me personally and prevents the perception of a polished playback. Videocards are the cause of the problem because they still operate on frequencies that are either slighter faster or slightly slower than Blue-Ray players and mastered films use, depending on whether you have an AMD card or a nVidia card.

JRiver Media Center sucks IMHO. Its very bulky with too many needless features (although you can disable most of them), slower performance, and harder to use than MPC-HC.

Does madVR 64bit rendering + 64bit MPC-HC + 64bit LAV Filters actually perform better? Does it allow, for example, higher neuron settings for Chrome Upscaling and/or Chrome/Luma Image Doubling/Quadrupling on the same hardware that displays frame drops when using x86 versions of madVR + LAV Filters + MPC-HC?

[XMonarchY]

OK, last post in a row:
- Is Catmull-Rom + Activate Anti-Ringing Filter + Scale in Linear Light still the best Image Downscaling method for the best quality? Lanczos shows better stats - less aliasing, but more sharpness. I am not going to argue with experts who say Catmull-Rom is better. I just wonder whether that has changed.
- Is Catmull-Rom also the best Image Downscaling method when Image Doubling/Quadrupling is used?
- Does nVidia DSR for desktop improve film playback quality? For example, I can downscale 4K DSR to my monitor's/TV's native 1080p resolution. I know games look better, but films are not games and this may not apply. Would DSR look better than NNEDI Image Doubling/Quadrupling, for example? I use 0% Smoothness.
- I just tried MPC-HC x64, but I am was not sure whether madVR x64 was utilized. Task Manager showed that "mad* home cinema control" service/process is 32bit and I couldn't tell whether LAV filters used are x64 from any menu and Task Manager. I used the all-inclusive LAV filter installation that carried both x86 and x64 versions.
- It looks like the latest builds of madVR are mostly bug fixes, which is awesome, but I am curious whether there are any image quality improving features being developed, although current image quality is already very good!

[tobindac]

Motion interpolation is like upscaling, it helps sometimes, but its tied to the quality of the source, you can't create information that isn't there.

I noticed SVP looks amazing on the better 1080p sources for some reason. On various 720p sources the artifacts in high motion scenes are more obvious (which is weird since resolution alone shouldn't do that).

I personally go even on bad sources with SVP since I can't stand the slideshow of low fps more than I can bear the artifacts in some rare scenes.

[XMonarchY]

I noticed SVP looks amazing on the better 1080p sources for some reason. On various 720p sources the artifacts in high motion scenes are more obvious (which is weird since resolution alone shouldn't do that).

I personally go even on bad sources with SVP since I can't stand the slideshow of low fps more than I can bear the artifacts in some rare scenes.

Bah. Artifacts are always there and easily noticed. madVR is all about high quality and SVP negates that IMHO. High refresh rate + light strobing on LCD's or high-end plasma TV's are the ways to go if you want smooth and fluid motion. I believe several of Sony's Smooth Motion TV's use light-strobing (a lot like on Eizo FG2421) instead of interpolation, but nothing beats a good plasma TV, except for maybe LightBoost/ULMB on TN panels with horrid image quality for film playback.

Only films recorded at higher framerate, such as The Hobbit trilogy, are worth watching at higher refresh rate. 60fps recording @ 60Hz is excellent, but 23fps recording + SVP @ 60Hz looks very unnatural.

[huhn]

OK, some updates:
- Both displays are 1080p
- vRAM usage does not go over 1400MB regardless of whether I use the new or the old FSE mode.
- New FSE mode seems to be working fine on my 120Hz monitor, which stays at 120Hz during film playback. Max stats remain at 24.5ms 99% of the time, but 250ms+ appears once in a while, without creating frame drops. Frames are dropped on my 60Hz TV, which goes into 23Hz mode for film playback. Max stats remain at 250ms+ 90% of the time, but 24.5ms appears 10% of the time.
- GPU usage does not go over 16% on 60Hz TV @ 23Hz during playback. CPU usage for madVR stays @ 0% and for MPC-HC 32bit does not cross 6%, regardless of FSE mode.
- GPU usage does not go over 20% on 120Hz monitor @ 120Hz during playback. CPU usage for madVR stays @ 0% and for MPC-HC 32bit goes up as high as 20%, regardless of FSE mode.
- RAM usage stays at 241MB, regardless of display and FSE mode.
it doesn't make any sense that you have higher CPU usage on the 120 Hz.
have you ever watched if your GPU clock changes in playback?
presenting frames in advance can take some more processing power from the GPU. so using new path(windowed and FSE) can result in higher GPU load on a 120 hz screen with a 23p source. but not by a lot.

I have some side questions:
- Assume FSE is set to old mode or not enabled at all and there are no frame drops on either display.
- Since CPU, vRAM, and GPU usages stayed so low, wouldn't that mean my system can handle more robust settings? For example, higher neurons for Chroma Upscaling and Chroma/Luma Image Doubling/Quadrupling should be possible, should they not?
- They aren't on my 60Hz TV @ 23Hz, but they are on my 120Hz monitor @ 120Hz. If I push Chroma Upscaling from 64 neurons to 128 neurons for 1080p content on my 60Hz TV @ 23Hz, I get frame drops, regardless of FSE mode. The same goes for sub-1080p content and Chroma/Luma Image Doubling/Quadrupling on my 60Hz TV @ 23Hz.

Why not? Is it due to being limited as far as render times go @ 23Hz and less limited @ 120Hz? Should it not be the other way around??? 16ms @ 60Hz (not sure about ms @ 23Hz) vs. 8ms @ 120Hz? It doesn't make sense to me...
the v-sync interval is not important. the 120 hz screen has a v-sync interval of 8.33ms but the important part is the video frame interval which is 41.71 wiht a 23p source.


OK, last post in a row:
- Is Catmull-Rom + Activate Anti-Ringing Filter + Scale in Linear Light still the best Image Downscaling method for the best quality? Lanczos shows better stats - less aliasing, but more sharpness. I am not going to argue with experts who say Catmull-Rom is better. I just wonder whether that has changed.
- Is Catmull-Rom also the best Image Downscaling method when Image Doubling/Quadrupling is used?
you don't ask for the best on this forum it's. there is no clear answer to that so that's the reason is not allowed in the first place. but people still like catmull-rom AR LL. let your eye decide that just copying stuff is a waste of time. and the madVR informaton are from madshi they are not absolute they are there to give you an idea what you can except from the scaler.

- Does nVidia DSR for desktop improve film playback quality? For example, I can downscale 4K DSR to my monitor's/TV's native 1080p resolution. I know games look better, but films are not games and this may not apply. Would DSR look better than NNEDI Image Doubling/Quadrupling, for example? I use 0% Smoothness.

most games down have a resolution they are theoretical unlimited.
videos have resolution so DSR does nothing it just force madVR to upscale by the power of x2 and your GPU is now suing crappy scaler to downscale is back to the native resolution of your screen.
not a good idea.
- I just tried MPC-HC x64, but I am was not sure whether madVR x64 was utilized. Task Manager showed that "mad* home cinema control" service/process is 32bit and I couldn't tell whether LAV filters used are x64 from any menu and Task Manager. I used the all-inclusive LAV filter installation that carried both x86 and x64 versions.
- It looks like the latest builds of madVR are mostly bug fixes, which is awesome, but I am curious whether there are any image quality improving features being developed, although current image quality is already very good!

when you use a 64 bit player and get madVR in the filter list it is 64 bit madVR 32 bit wouldn't work. it is that simple.
"*mad* home cinema control" has nothing to do with it.

Some of us are sensitive and continue seeing de-synchronization between audio and video without ReClock or JRiver Media Center's similar feature. It does reduce immersion for me personally and prevents the perception of a polished playback. Videocards are the cause of the problem because they still operate on frequencies that are either slighter faster or slightly slower than Blue-Ray players and mastered films use, depending on whether you have an AMD card or a nVidia card.
and some people hate resampled of audio.
the issue is not down to the GPU alone.
the issue is down to systemclock ,soundcard clock and GPU they are not the same. you can by 3 r9 270 and get 3 different clocks.

JRiver Media Center sucks IMHO. Its very bulky with too many needless features (although you can disable most of them), slower performance, and harder to use than MPC-HC.
that's a mean thing to say it uses lavfilter too...

Does madVR 64bit rendering + 64bit MPC-HC + 64bit LAV Filters actually perform better? Does it allow, for example, higher neuron settings for Chrome Upscaling and/or Chrome/Luma Image Doubling/Quadrupling on the same hardware that displays frame drops when using x86 versions of madVR + LAV Filters + MPC-HC?
madVR shouldn't be faster.
but things like lavfilter can be a lot faster.

[XMonarchY]

Thank you Huhn! I really appreciate your help in these matters. I believe you've answered all my questions! At this point I'll test GPU and CPU usage during playback with different FSE modes enabled. One way or another, the problem is solved - I can use the old FSE method and very happy with it.

I do have another small dilemma:
When Image Doubling uses only the scaling factor of 1.5x for 720p content, which results in 1080p image (native to my display), does it still double the image first? Meaning that even 1.5x factor for 720p content will still double the image to 1440p (2x720p) first, and then downscale the image to display's native resolution, which is 1080p in this case. Otherwise, if Image Doubling uses only a scaling factor of 1.5x (not 2.0) for 720p content, then there is no reason Image Downscaling should be used because in my case 1.5x720p=1080p (native resolution for my displays). When only Image Doubling is enabled for 720p content on 1080p displays, I press CTRL+J and under "Target Rectangle" I see:

Luma > NNEDI64 < Catmull-Rom AR
Chroma > Jinc3 AR

I assume the 2 lines above show current Scaling Algorithms applied for the content being displayed. It means Image Doubling always doubles the image by a factor of 2.0 as its first step, regardless of everything else, right?

Would it be possible (in the future) to multiply the image by correct factor (1.5x for 720p content in my case) that would result in the usage of display's native resolution, without having to go through Image Downscaling process, which is an extra step that probably reduces the quality of pure upscaling?

[nevcairiel]

Image doubling is called that for a reason, it can only ever double the image. Its a strict limitation of NNEDI3, the algorithm requires it.

Only "normal" scaling algorithms can do 1.5x, so either you upscale 1.5x with eg. Jinc directly, or you double and downscale. You should judge the quality of the result and decide for yourself.

[Anime Viewer]

OK, some updates:

- Frames are dropped on my 60Hz TV, which goes into 23Hz mode for film playback. Max stats remain at 250ms+ 90% of the time, but 24.5ms appears 10% of the time.

Some videos do not play well when screen refresh rates try to match video at 23hz. Have you tested with OSD (ctrl+J) using EVR custom (instead of madVR) to see if it shows the video fluctuating (variable frame rates)? You may get the best playback of 23hz videos by actually not having having 23hz or 24hz as supported frame rates in your supported display modes area in madVR settings. If your TV supports 60, 59, 30, 29, 24, and 23 try only with an entry like: "1080p60, 1080p59, 1080p30, 1080p29" and see how the video(s) preform.

[leeperry]

@madshi: could you please allow madhccctrl to recognize the "windows minimize" message? I got it mapped for all my apps on the left mousewheel button(right one is CTRL+F4) in the Logitech drivers and now that it's not running always on top anymore I'd like to keep it open in the background so I won't forget to uncheck "treat 25p as 24p" after I'm done watching a PAL sped up movie. I can always click on it on the taskbar and it'll minimize anyway but it'd be quite handy if I could just minimize it with a hotkey.

Thanks for all the updates BTW, watching movies has never been that enjoyable...gotta love when there's nothing to nitpick about anymore, I'm just bored standing in awe at what I see :D

[Anime Viewer]

Its interesting seeing the results of NNEDI3 testing that is currently going on with the MPDN player, and how different shader codes (optimizations) seem to make a significant difference at higher settings depending on what code is being used (Perfer Scalar, Perfer Vector, or Avoid Branches).

madashi: Out of curiosity which style of NNEDI3 is used with madVR currently (any of these, or something different)? Don't know how important NNEDI3 optimization in madVR is on your list of priorities, but something similar might be worth testing.

[QBhd]

7/32,ZP10,LAV 0.65,madVR 87.19,HD 7790,Omega driver 14.12.

Hi,
has anybody had problems with deinterlacing 1080i using HD 7790,R7 260 or 260x?
The present and render queues won't fill up and I have subsequently hundreds of dropped frames in a matter of seconds.
I previously used an HD 4770 + Catalyst 12.4 with the exact same software player config and never observed such a catastrophy.

I think I noticed something along these lines... I had all my queues set to 4, but with 1080i (film) they were not filling... All I did to solve this was bump only the CPU queue up to 5 and all was well again...

BTW, you might want to try the 13.12 AMD drivers. It's still is the best performer on my system (specs below). I periodically try all the latest Catalyst packages, but they always fail to render fast enough. I don't un-install the Catalyst package and re-install 13.12, I just roll-back the driver to this one:

13.251.9001.1001

http://img.prntscr.com/img?url=http://i.imgur.com/xoVhbya.png


QB

[khanmein]

OK, some updates:
- Both displays are 1080p
- vRAM usage does not go over 1400MB regardless of whether I use the new or the old FSE mode.
- New FSE mode seems to be working fine on my 120Hz monitor, which stays at 120Hz during film playback. Max stats remain at 24.5ms 99% of the time, but 250ms+ appears once in a while, without creating frame drops. Frames are dropped on my 60Hz TV, which goes into 23Hz mode for film playback. Max stats remain at 250ms+ 90% of the time, but 24.5ms appears 10% of the time.
- GPU usage does not go over 16% on 60Hz TV @ 23Hz during playback. CPU usage for madVR stays @ 0% and for MPC-HC 32bit does not cross 6%, regardless of FSE mode.
- GPU usage does not go over 20% on 120Hz monitor @ 120Hz during playback. CPU usage for madVR stays @ 0% and for MPC-HC 32bit goes up as high as 20%, regardless of FSE mode.
- RAM usage stays at 241MB, regardless of display and FSE mode.


I have some side questions:
- Assume FSE is set to old mode or not enabled at all and there are no frame drops on either display.
- Since CPU, vRAM, and GPU usages stayed so low, wouldn't that mean my system can handle more robust settings? For example, higher neurons for Chroma Upscaling and Chroma/Luma Image Doubling/Quadrupling should be possible, should they not?
- They aren't on my 60Hz TV @ 23Hz, but they are on my 120Hz monitor @ 120Hz. If I push Chroma Upscaling from 64 neurons to 128 neurons for 1080p content on my 60Hz TV @ 23Hz, I get frame drops, regardless of FSE mode. The same goes for sub-1080p content and Chroma/Luma Image Doubling/Quadrupling on my 60Hz TV @ 23Hz.

Why not? Is it due to being limited as far as render times go @ 23Hz and less limited @ 120Hz? Should it not be the other way around??? 16ms @ 60Hz (not sure about ms @ 23Hz) vs. 8ms @ 120Hz? It doesn't make sense to me...

i think ur monitor is actually 60Hz not 120Hz (due to OC)

[kalston]

The Eizo FG2421 is real 120hz.