madVR - high quality video renderer (GPU assisted)

[Georgel]

Georgel, what setting do you use for 720p anime? Sunset1982 i will be waiting!

Sorry for late answer!

I am using a GTX 860m GPU so these settings are relevant for this GPU, maybe a weaker one, as with a stronger one, you can crank up SuperRes and many other settings!

Deinterlacing - all off

artifact removal - High / high

reduce ringing - on, without second option

all image enhancements off

chroma - Jinc / with AR without SuperRes because my GPU cannot do SuperRes

downscaling, SSIM, 1D - 100%, with soft AR, scale in linear light, AB at 150%

image doubling off

image upscaling - JINC with AR and Sigmoidal light

upscaling refinement - Sharpen edges at 4.0, Thin edges at 8.0, AR at 150 %, AB, Refinne once after upscale complete, all other off.

Render settings, both CPU and GPU maxxed out,

disabled all 3D options

disabled smooth motion

dithering- none

[Nevilne]

Can you even give a worse advice on purpose

[Warner306]

Can you even give a worse advice on purpose

Looks ok to me. What do you use?

[Georgel]

Can you even give a worse advice on purpose

Why do you say that?

The result image is very clean, smooth, crisp and has a lot of detail.

It looks very improved compared to the un-filtered version. It looks higher resolution, to put it bluntly.

Higher resolution, and without adding any aliasing or artifacts.

This is one reason for that I was unable (so far) to use anime + SVP, SVP just adds too much artifacts for me.

P.S. I would crank Super Res up, but the perceived quality gain from using edge thinning and edge sharpening is much higher than when using Super Res. This and for a GTX 860m, super res is a bit too stressful.

[ryrynz]

Can you even give a worse advice on purpose

On 720P anime which is already reasonably detailed these settings come across to me to be quite excessive and would likely be harming image quality considerably, debanding at maximum for 720P?
How bad are these videos?

The first thing you should be doing is viewing high bitrate videos, I keep mine on low, set a shortcut to change it if the source requires.

On 720P I usually just upscale with NNEDI3 (which doesn't make that much of a difference on a 1080 screen TBH) downscale with SSIM 1D without AB maybe do some reasonable SE and CE sharpening and that's it. Line thinning was kinda useful on some soft ~640x480 anime I had but it even at lower settings in some areas didn't produce good results which is why I asked madshi about adding warpsharpen to madvr which is considerably better for anime.


The result image is very clean, smooth, crisp and has a lot of detail.


Link me a sample file that you feel your settings look good on and upload your settings.dat and I'll do a screenshotcomparison vs what I would use.
I might be able to make you reconsider what is considered good image quality.



Madshi, would it be possible to add to madVR the ability for it to remember settings for certain videos?

For example if I watch a video that has poor color and lots of banding and I hit a couple of hotkeys to change these defaults, could another hot key be pressed to remember those changes for any future videos played within that directory? That way if watching a series the user doesn't need to perform these adjustments again and again.

[Ver Greeneyes]

For example if I watch a video that has poor color and lots of banding and I hit a couple of hotkeys to change these defaults, could another hot key be pressed to remember those changes for any future videos played within that directory? That way if watching a series the user doesn't need to perform these adjustments again and again.You can already do that by adding tags to the directory name. But I agree that it would be nice to have a "use these settings for all files in this directory" function that doesn't rely on renaming the directory (and a "remember these settings for this file").

[Asmodian]

Sorry for a delayed response.

I don't get any banding, since I am also using reduce banding artifacts with high / high on processing -> artifact removal.

I also use reduce ringing artifacts too from there!


Debanding as set there has nothing to do with the banding due to lack of dithering, in fact on high/high it would make it worse by blurring details that help disguise the banding caused by rounding down (instead of dithering) madVR's internal 16-bit data to 8-bit.

Are you watching close to a large low resolution display?

Looks ok to me. What do you use?

Debanding high/high? Thin edges 8.0? Sharpen edges 4.0? Dithering off?

I watch mostly anime and I would never use anything like that and strongly recommend no one use them unless testing for themselves. Definitely settings that should not be recommended generally.

[70MM]

Guys Ive started to get some very pleasing images now without using Image Doubling with my 1080 card through my JVC X9000 projector.

Question please...
I can easily do NNED13, 128 neurons from Chroma Upscaling, SR and many other features turned off except AR. No dropped frames and all the queues sit on 7-8/8

I can also move up to NNED13, 256 neurons with no dropped frames, the only thing that goes down is the present cue from 7-8/8 to 6-8/8.
What I want to know should I remain on the higher NNED13, 256 neurons since all looks fine, is it considerably better than the 128 neurons???

Its so hard to tell as the picture is remarkable... :)

Again thanks to all those that have helped me understand more about madvr...

[Asmodian]

Guys Ive started to get some very pleasing images now without using Image Doubling with my 1080 card through my JVC X9000 projector.

Question please...
I can easily do NNED13, 128 neurons from Chroma Upscaling, SR and many other features turned off except AR. No dropped frames and all the queues sit on 7-8/8

I can also move up to NNED13, 256 neurons with no dropped frames, the only thing that goes down is the present cue from 7-8/8 to 6-8/8.
What I want to know should I remain on the higher NNED13, 256 neurons since all looks fine, is it considerably better than the 128 neurons???

Its so hard to tell as the picture is remarkable... :)

Again thanks to all those that have helped me understand more about madvr...

While I do think 256 neurons looks better than 128 it is less significant than the difference between 128 and 64, which is less significant than 64 and 32, etc.
Edit: this is for Luna doubling, are you talking about chroma only? For chroma I never use above 32 neurons because I never could find a source where I could tell the difference.

Once present starts to go a little slow you are very close to the edge, I might turn it down to 128, or just use 32 neurons for chroma upscaling. Can you tell the difference between 32 and 128 neuron chroma upscaling? I cannot.

BTW, it is NNEDI3, Neural Net Edge Directed Interpolation, version 3. :)

[70MM]

While I do think 256 neurons looks better than 128 it is less significant than the difference between 128 and 64, which is less significant than 64 and 32, etc.

Once present starts to go a little slow you are very close to the edge, I might turn it down to 128, or just use 32 neurons for chroma upscaling. Can you tell the difference between 32 and 128 neuron chroma upscaling? I cannot.

BTW, it is NNEDI3, Neural Net Edge Directed Interpolation, version 3. :)

Ok nice to know, I will just leave it at 128 and I really cant see an improvement on 256....Many thanks for the correction NNEDI3:o

[leeperry]

The result image is very clean, smooth, crisp and has a lot of detail.
Happy Happy Joy Joy then, what matters is that you like what you see coz at the end of the day the only thing that matters is your own satisfactshun.

All this said I have no idea how anyone can bear having any of the two dithering sub-options enabled so maybe you could try vanilla ED2 with both of these pesky things disabled as one makes the noise dance around on top of the picture and the other creates very faint color fringing :angry:

[70MM]

Happy Happy Joy Joy then, what matters is that you like what you see coz at the end of the day the only thing that matters is your own satisfactshun.

All this said I have no idea how anyone can bare having any of the two dithering sub-options enabled so maybe you could try vanilla ED2 with both of these pesky things disabled as one makes the noise dance around on top of the picture and the other creates very faint color fringing :angry:

Are you mentioning here not to have ticked "Use coloured noise" and "change dither for every frame"

Im currently using Error Diffusion Option 2 with the two above ticked...

Is it cleaner without using them?

I only ticked them as a tutorial on the web said to tick both:(

[ryrynz]

Ok nice to know, I will just leave it at 128 and I really cant see an improvement on 256....Many thanks for the correction NNEDI3:o

Chroma isn't that big of a deal, personally I can't imagine ever using anything about 32 or 64 neurons if I choose to use NNEDI3. I feel super-XBR is a far better option. Why not use a few videos for comparison to evaluate the difference?

I did a test on my GTX 960 on a full HD video and it took my rendering time up from 11ms to 28ms.. if you're going to used NNEDI3 for chroma then consider 64 neurons, that was like 16ms here. I doubt you'd notice the difference between 64 and 128 neurons. If you're not using this performance for anything else... sure, but it's excessive and you wouldn't find me using it, even if I had a Titan X!

As always I would recommend users always evaluate the difference in picture before changing options, especially the high cost options. NNEDI3's sweet spot is 64 neurons as far as cost to upscaling improvements is concerned. I feel anything higher for chroma really is just wasting power. If you can't visually see a difference then there's no point in changing it if it costs more performance, more heat, more noise, more on your bill. Doesn't make much sense.

[70MM]

Chroma isn't that big of a deal, personally I can't imagine ever using anything about 32 or 64 neurons if I choose to use NNEDI3. I feel super-XBR is a far better option. Why not use a few videos for comparison to evaluate the difference?

I did a test on my GTX 960 on a full HD video and it took my rendering time up from 11ms to 28ms.. if you're going to used NNEDI3 for chroma then consider 64 neurons, that was like 16ms here. I doubt you'd notice the difference between 64 and 128 neurons. If you're not using this performance for anything else... sure, but it's excessive and you wouldn't find me using it, even if I had a Titan X!

As always I would recommend users always evaluate the difference in picture before changing options, especially the high cost options. NNEDI3's sweet spot is 64 neurons as far as cost to upscaling improvements is concerned. I feel anything higher for chroma really is just wasting power. If you can't visually see a difference then there's no point in changing it if it costs more performance, more heat, more noise, more on your bill. Doesn't make much sense.

Oh Gee!
I thought using the higher 128 neurons with NNEDI3 was going to be the icing on the cake, but it looks like its not really worth it. I did know that Luma is much more important than Chroma and I have noticed that the rendering does go up, yet I thought since I had the 1080 card I should be using the highest settings....But you are right, my eyes don't seem to see anything better with the higher chroma settings :(

I wish there was some Luma settings that I could use to bump that side up since they say that Luma has the more important.

I will drop the Chroma upscaling back to 64 neurons and that too will really lower the render times....

My screen is pretty large, 120" dia 16:9 and 146" wide scope. I have to take care with some of the settings as if they are bumped up too high it really shows up artefacts!

Im learning stuff from this thread and many of you experts have been very helpful.

BTW all my watching is mainly BD rips, don't game or show much animated progs. No TV or anything else like that....

However I have stopped using Image Doubling after many here said its not really worth it on 1080 > 1080 BD rips....

[chros]

The 25Hz rezolution is not supported by TV and so these movies are displayed at 50 Hz. madVR stats are reporting 1 frame drop every 8-9 hrs or even 1.2 days (i have very light up-scaling options) but shows a huge number of repeated frames and the counter does not stop.
That's strange, it shouldn't work like this.

Is it because the 25 fps movie in order to be displayed on 50 Hz screen the number of frames need to be doubled?
No. There's no doubling is used unless: do you have Smooth Motion enabled? If so, then disable it, you don't need it at all if display refresh rate is really close to the origin fps.

Is this something normal or am I doing something wrong here?
No, it's not normal, and you do everything well. :)
This is how I use 29.97/30 fps content with 60Hz (in PC mode: it treats chroma differently).

[Sunset1982]

LumaSharpen is good sharper and I prefer it over AdaptiveSharpen but saying that I'm not big fan of LumaSharpen as using low values does nothing for me and with bigger values I noticed it sharpens noise as well, so I prefer to use Crispen Edges with Enhance detail instead as it achieves better result for my taste without really amplifying noise\film grain


Ok, thank you for your explanation. What values are you using with Crispen Edges and Enhance Detail?

[robl45]

Hello,

Is there any guide to what the Intel HD 530 internal graphics can handle? I'd like to see if I can get better picture out of madvr. Have never really played with the settings.

[AntonP]

You can't, you get the 3d refresh rate that you get and for nvidia that is 23.971 or so. Reclock certainly works fine with this, I have used it for years, but you aren't going to get bitstreaming like that so I have switched to using the internal intel graphics. Heck of a waste on a GTX950, but oh well.

At any rate. Reclock will take and speed up the video to 24 fps and match the audio. You should set your refresh at the nvidia 23hz setting which is 23.971 and let reclock do its thing.


I cant agree with you. For perfect playback with no dropped or repeated frames monitor's refresh rate (from nvidia) and movie's fps should be exactly the same.
Reclock makes 24fps movie from 23.976 and with 24Hz (madvr 1080p24 mode) displays refresh rate the picture is perfect! With 1080p23 mode (23.971 from nvidia) there are dropped frames. That's all about 2D. In 3D nvidia panel or windows display properties has both 23p and 24p modes, but madvr doesnt let me choose! There is not such an option in display modes smth like "1080p3D24p" ((




PS Reclock works with bitstream audio with now problems!

[robl45]

You don't understand how reclock works. Your refresh rate has to be the same as the movie for perfect playback. AFAIK, only intel can do perfect 23.976. With nvidia, you get 24hz or 23.971. Reclock will speed the move up to 24 and match the audio. 23.971 or 24hz makes no difference, you still are not matching the video and reclock will do the work. In the madvr display, you will see one frame drop or repeat like once every 8 hours or more if reclock is doing its job.


I cant agree with you. For perfect playback with no dropped or repeated frames monitor's refresh rate (from nvidia) and movie's fps should be exactly the same.
Reclock makes 24fps movie from 23.976 and with 24Hz (madvr 1080p24 mode) displays refresh rate the picture is perfect! With 1080p23 mode (23.971 from nvidia) there are dropped frames. That's all about 2D. In 3D nvidia panel or windows display properties has both 23p and 24p modes, but madvr doesnt let me choose! There is not such an option in display modes smth like "1080p3D24p" ((




PS Reclock works with bitstream audio with now problems!

[ryrynz]

Hello,

Is there any guide to what the Intel HD 530 internal graphics can handle? I'd like to see if I can get better picture out of madvr. Have never really played with the settings.

No, you'll need to test that for yourself "playing with the settings" is kind of a requirement to get the best (in your opinion) from it.

[leeperry]

I'm currently using Error Diffusion Option 2 with the two above ticked...

Is it cleaner without using them?

I only ticked them as a tutorial on the web said to tick both

Did you miss the parts that say "subjectively increases noise" and "increases chroma noise"?

Tutorials are nice and all but we all run different rigs, got different visions(8% of men are colorblind at different levels) and personal tastes.

[AntonP]

You don't understand how reclock works. Your refresh rate has to be the same as the movie for perfect playback. AFAIK, only intel can do perfect 23.976. With nvidia, you get 24hz or 23.971. Reclock will speed the move up to 24 and match the audio. 23.971 or 24hz makes no difference, you still are not matching the video and reclock will do the work. In the madvr display, you will see one frame drop or repeat like once every 8 hours or more if reclock is doing its job.

I understand it )) afaik both ATI and intel (new models) do correct 23.976.

The only difference with 23.971 and 24 display's refresh rates is that with 24Hz + reclocked 23.976 video i have no visiable drops (one in 20-40 hours is not a problem for me).
I want to use the same trick in 3D, but i cant do this becouse of absense of 24 (48)Hz option.

[robl45]

No, you'll need to test that for yourself "playing with the settings" is kind of a requirement to get the best (in your opinion) from it.

Yes, but how do i know what the HD 530 graphics can handle? Is it just basically if the CPU gets maxed out?

[strumf666]

You mean the GPU? ;)
Usually at GPU load ~70% and above dropped frames start occurring; it's quite visible to the eye, but you can check the OSD (ctrl+j) for dropped frames counter.

[robl45]

I understand it )) afaik both ATI and intel (new models) do correct 23.976.

The only difference with 23.971 and 24 display's refresh rates is that with 24Hz + reclocked 23.976 video i have no visiable drops (one in 20-40 hours is not a problem for me).
I want to use the same trick in 3D, but i cant do this becouse of absense of 24 (48)Hz option.

You would have no visible drops with 23.971 with reclock either. I've been using it this way for years. You may have to clean out the timings in reclock config so it can learn the 23.971 refresh rate. As far as I know for 3d, you get 23.971 with Nvidia. I tried making a custom refresh rate with nvidia and the 3d got amazingly messed up. I gave up on the Nvidia at this point as I'd like to be able to try DTSX and Atmos and they need to bitstream with 23.976.

[robl45]

You mean the GPU? ;)

How would I know if the GPU gets maxxed out? I assume the CPU will start maxxing out if the GPU can't handle it? Sorry, I've never really played with this much even though I've been using madvr for years. I set it to some low settings as my old GT430 wasn't very powerful.

[strumf666]

I edited my first reply for more info&clarity. You can use a program like gpuz for gpu load or any other (there are many), but you don't actually need that, because massive frame drops are easily visible, for minor you can use the OSD (ctrl+j) which has a dropped frames counter.
Your assumption isn't entirely correct, because CPU and GPU aren't fully interchagable in regards to the work madVR requires and even then the GPU is usually much more powerful.

[Warner306]

I cant agree with you. For perfect playback with no dropped or repeated frames monitor's refresh rate (from nvidia) and movie's fps should be exactly the same.
Reclock makes 24fps movie from 23.976 and with 24Hz (madvr 1080p24 mode) displays refresh rate the picture is perfect! With 1080p23 mode (23.971 from nvidia) there are dropped frames. That's all about 2D. In 3D nvidia panel or windows display properties has both 23p and 24p modes, but madvr doesnt let me choose! There is not such an option in display modes smth like "1080p3D24p" ((




PS Reclock works with bitstream audio with now problems!

I believe ReClock works on the audio alone, speeding it up or slowing it down to match the display refresh rate. madVR can change the speed of a 25 fps video to 24 fps. ReClock would then slow down the audio by the same amount.

[robl45]

I believe ReClock works on the audio alone, speeding it up or slowing it down to match the display refresh rate. madVR can change the speed of a 25 fps video to 24 fps. ReClock would then slow down the audio by the same amount.

No reclock changes the video speed and then resamples the audio to match. Reclock by itself will slowdown 25 to 24p, I've used that feature many times. i'm not sure exactly what madvr is doing with that. I suppose that function is if you use madvr to automatically select the refresh rate.

[Stereodude]

Ok, last post on the gtx1060 / rx480 topic, let's get back to madvr: (you can write me some pm if you got questions)

here is my upgraded comparison:

http://www.file-upload.net/download-11785367/MadVRComparison.xlsx.html
So, is Pascal better in madVR from Maxwell or did the RX 480 take a step back from the previous AMD hardware? Or, maybe both?

In the past an AMD card that got beat by a given Nvidia card in games could equal or beat it with madVR. Like a GTX 960 had about 10% better render times in madVR than a R7 260x, but the GTX 960 would be like 50% ahead in games compared to the R7 260x. The AMD cards tended to have a lot more shader/compute power for the price that didn't seem to get reflected in game performance. Now, the RX 480 and the GTX 1060 are effectively tied in madVR (averaging your results for all test cases), and their gaming performance is fairly comparable with DX11 leaning GTX 1060 and DX12/Vulkan leaning RX 480.

The GTX 960 had about 10% more FLOPS than the R7 260x and was about 10% faster in madVR. But, the RX 480 has 50% more FLOPS than the GTX 1060, but isn't any faster at madVR? :confused:

I was hoping to buy a RX 480 to upgrade my HTPC from an R9 380, but I'm not sure it's really much faster.

[Asmodian]

Are you mentioning here not to have ticked "Use coloured noise" and "change dither for every frame"

Im currently using Error Diffusion Option 2 with the two above ticked...

Is it cleaner without using them?

I only ticked them as a tutorial on the web said to tick both:(

I also much prefer them both disabled but that is on a very different screen, 27" 2560x1440 or a 1080p Plasma, so you will have to judge for yourself. I quite dislike chroma noise though.

I agree with ryrynz too, max NNEDI3 64 for chroma. I actually use Reconstruction soft + SR1 for chroma, or sometimes NNEDI3 32.

[robl45]

I edited my first reply for more info&clarity. You can use a program like gpuz for gpu load or any other (there are many), but you don't actually need that, because massive frame drops are easily visible, for minor you can use the OSD (ctrl+j) which has a dropped frames counter.
Your assumption isn't entirely correct, because CPU and GPU aren't fully interchagable in regards to the work madVR requires and even then the GPU is usually much more powerful.

Yes, this is the problem. I found a couple of pages that show settings and I tried those and the GPU was just maxed at 100% and the video wouldn't play. So I reset to default. Thats why I was asking what are recommended settings for Intel HD 530.

[imhh11]

Hi everyone.
Is it normal that the only way I can play ''Sony_4K_HDR_Camp'' in HDR is when i set madvr chroma upscaling to ''Bilinear'' ? Everything else doesnt work.
My gpu is the new GTX1070. Had the same problem with my GTX960

Other HDR demo like '' Eldorado_HDR1000_PQ_P3_4K.mp4 '' work without issue.

https://s32.postimg.org/5hgl8ax2d/Untitled_1.jpg

[ShiftyFella]

Ok, thank you for your explanation. What values are you using with Crispen Edges and Enhance Detail?
For movies\tvshows CE@0.6-0.7 ED@0.8-0.9, anime\cartoons usually go with CE@1.5-2 ED@1.2 and chroma upscaling changed to nnedi32 with SR for chroma turned off, sometimes I turn on Thin Edges, in particular for SD cartoons\anime but it depends on the file and how i feel how it looks

[ryrynz]

I agree with ryrynz too, max NNEDI3 64 for chroma. I actually use Reconstruction soft + SR1 for chroma, or sometimes NNEDI3 32.

NNEDI3 for chroma is nice indeed, 32 or 64 neurons are good choices if you have the performance to spare.

Hopefully not to bring begin a chroma discussion yet again here.. but I'd totally recommend you evaluate super-xbr 100 and 125 AR in comparison to Recon soft + SR1. I've never been a fan of any of the Reconstruction options especially all the sharp and placebo options (which I think should be be removed because of how bad they are) but Recon even with the improvement of SR1 still looks rather messy in comparison to super-xbr. Not only that, changing will save some performance too. If you find anything you feel Recon + SR1 looks better on than super-xbr please feel free to share it but personally I've found nothing other than NNEDI3 to rival it.

Also, I question allowing SuperRes up beyond a strength of two (or maybe even one) on chroma, it produces some fairly nasty artifacts even on the softest upscalers.
Perhaps madshi could consider having super-res just as a tick box with a strength setting of one? Does anyone actually use anything higher? If so perhaps I could assist you with considering IMO better options.

[70MM]

NNEDI3 for chroma is nice indeed, 32 or 64 neurons are good choices if you have the performance to spare.

Hopefully not to bring begin a chroma discussion yet again here.. but I'd totally recommend you evaluate super-xbr 100 and 125 AR in comparison to Recon soft + SR1. I've never been a fan of any of the Reconstruction options especially all the sharp and placebo options (which I think should be be removed because of how bad they are) but Recon even with the improvement of SR1 still looks rather messy in comparison to super-xbr. Not only that, changing will save some performance too. If you find anything you feel Recon + SR1 looks better on than super-xbr please feel free to share it but personally I've found nothing other than NNEDI3 to rival it.

Also, I question allowing SuperRes up beyond a strength of two (or maybe even one) on chroma, it produces some fairly nasty artifacts even on the softest upscalers.
Perhaps madshi could consider having super-res just as a tick box with a strength setting of one? Does anyone actually use anything higher? If so perhaps I could assist you with considering IMO better options.

Well Ive tested and tested so much material over the past weeks and only yesterday came to a final conclusion... And Im really happy!

Ive found so much stuff going on in the background of the image especially since mine is huge compared to a TV. 146" wide scope screen shows up everything...Ive tested so many features and always could see artifacts moving round in the background, I never knew what features were doing it as I had so many engaged. I thought since madvr was feature rich that the more you used the better the image might be, more fool me....:rolleyes:

I turned everything off and started from scratch, just one on at a time.... And watched up close to the screen....

I have finally decided (for me) that using a very large screen the less features one uses the cleaner the image..

So this is all Im going to use and the image now is as clean as a whistle, sharp and detailed with no artefacts moving round in the background.

Very simple settings for me are now the order of the day, even though Im using the 1080 card.

Image Enhancements:
sharpen edges 0.5
crispen edges 0.5
enhance detail 0.5
AR ticked

Chroma upscaling NNEDI3 64 neurons

Image upscaling Jinc AR and Sig light ON

Dithering Error Diffusion Option 2

Basically that's all Im using now, Ive turned everything else off and finally Im over the moon with a sharp, detailed and artefact free image, its wonderful. These settings make my JVC X9000 proj display a clean and really amazing image on BD rips.

[ryrynz]

These settings make my JVC X9000 proj display a clean and really amazing image on BD rips.

Sounds pretty good to me. Don't forget to uncheck all the trade quality for performance options, evaluate whether smooth motion helps at all, (probably not with a projector?) and look into whether debanding is useful for anything you watch (probably not) along with removing ringing artifacts (which I find especially useful with lower resolution anime)

[SpoCk0nd0pe]

The GTX 960 had about 10% more FLOPS than the R7 260x and was about 10% faster in madVR. But, the RX 480 has 50% more FLOPS than the GTX 1060, but isn't any faster at madVR? :confused:

I was hoping to buy a RX 480 to upgrade my HTPC from an R9 380, but I'm not sure it's really much faster.

GTX 960 is based on Kepler which is a really shitty architecture if you try anything other then rasterize polygons.

The stock RX480 throttles due to thermal issues like most recent blower cooled AMD cards.

nVidia is gaining ground in fixing their architecture because they noticed that neglecting shader power is a bad move for most workloads.

GCN's features are sadly pretty hard to use. GLSL and openCL do not provide all the features needed, so you waste some of those theoretical TFlops. Some of the GCN 1.2 features are probably not even beneficial for the madvr shaders.

I wouldn't upgrade, you probably won't notice too much of a difference. If you still want the extra power, wait for the aftermarket coolers. Sapphire has been consistently good with cooling AMD cards in the past few years, but there are small differences each generation. Stay away from Gigabyte, imho they apply too much voltage just for binning.

[ionutm80]

That's strange, it shouldn't work like this.


No. There's no doubling is used unless: do you have Smooth Motion enabled? If so, then disable it, you don't need it at all if display refresh rate is really close to the origin fps.


No, it's not normal, and you do everything well. :)
This is how I use 29.97/30 fps content with 60Hz (in PC mode: it treats chroma differently).

Hi, I have re-installed madVR from scratch, reset settings to default and now the stats are showing only 1 repeat frame every xx minutes but the counter for repeated frames is still increasing while the movie plays. See below:

http://i.imgur.com/8GUWUzp.jpg

And yes smooth motion is off.

I have also attached:

- display modes input in madVR:

http://i.imgur.com/S5B8inw.jpg

- screen refresh rates allowed by the GPU (Intel HD 44000):

http://i.imgur.com/jo1dvkr.jpg

- MediaInfo detail of the video stream I'm trying to play (MPEG2 stream in mkv container ripped from PAL DVD with makeMKV)

http://i.imgur.com/8mAgURa.jpg

Again for Blu-Ray rips (23,976 H.264 High 4.1 in mkv container also ripped with makeMKV) everything seems to be fine.

I really don't understand what is happening.

[huhn]

the 720x576 files says it is progressive but top field first.

and madVR get's the information of 50 fps but the file has only 25.
if it is deinterlanced first what deinterlancer is used it is clearly not madVr DXVA deinterlancing.

[ionutm80]

the 720x576 files says it is progressive but top field first.

and madVR get's the information of 50 fps but the file has only 25.
if it is deinterlanced first what deinterlancer is used it is clearly not madVr DXVA deinterlancing.

You're right I have also noticed the 50fps movie info in madVR but I did not understand its meaning ...
I've chosen Software Deinterlacing Weston Three Field (Simple) in LAV Video, shall I put it to none and let madVR do the job?

[chros]

You're right I have also noticed the 50fps movie info in madVR but I did not understand its meaning ...
I've chosen Software Deinterlacing Weston Three Field (Simple) in LAV Video, shall I put it to none and let madVR do the job?
Good question: you have to try it out.
Deinterlacers have usually 2 modes (as you can see on LAV video properties page):
- film: reserves the origianl fps (worse quality then the other one)
- video: doubles the original fps (better quality then the previous one, but madvr has to work double)

Also, there are sometimes wrong flags in header (as in your case), like with some TV captures.

Suggestion:
- try to disable deinterlacing everywhere at first, whether you will see any artifact at all
- if you do, then you have to experiment with the possible options

[huhn]

Good question: you have to try it out.
Deinterlacers have usually 2 modes (as you can see on LAV video properties page):
- film: reserves the origianl fps (worse quality then the other one)
- video: doubles the original fps (better quality then the previous one, but madvr has to work double)

that's not how it works.
film in lavfilter is just halve frame rate deintelerancing which is usually not recommand at all. because they are not restoring the orginal frame rate.

if the source is interlanced than it has 60/50 fields worth of 60/50 different frames so film mode in lavfilter is "just" throughing halve of them away.

than we have interlanced content that is PsF and don't need a deinterlancer.
and we have telecine which is 24 frames on 60 fields (and other stuff) the film mode in lavfilter is not restoring these the film mode in madVR is.

if the source is telecine a IVTC (film mode) algorithm is clearly better.

50i usually needs a deinterlancer or can be treated progressive.

[ashlar42]

But is madVR supposed to be better than Weston at properly deinterlacing or not?

[ashlar42]

I cant agree with you. For perfect playback with no dropped or repeated frames monitor's refresh rate (from nvidia) and movie's fps should be exactly the same.
Reclock makes 24fps movie from 23.976 and with 24Hz (madvr 1080p24 mode) displays refresh rate the picture is perfect! With 1080p23 mode (23.971 from nvidia) there are dropped frames. That's all about 2D. In 3D nvidia panel or windows display properties has both 23p and 24p modes, but madvr doesnt let me choose! There is not such an option in display modes smth like "1080p3D24p" ((




PS Reclock works with bitstream audio with now problems!

You don't understand how reclock works. Your refresh rate has to be the same as the movie for perfect playback. AFAIK, only intel can do perfect 23.976. With nvidia, you get 24hz or 23.971. Reclock will speed the move up to 24 and match the audio. 23.971 or 24hz makes no difference, you still are not matching the video and reclock will do the work. In the madvr display, you will see one frame drop or repeat like once every 8 hours or more if reclock is doing its job.

In my recent experience Reclock is most definitely not needed to achieve perfectly smooth playback. Check my guide here, I suppose you might be interested: http://forum.doom9.org/showthread.php?t=173571

[Stereodude]

GCN's features are sadly pretty hard to use. GLSL and openCL do not provide all the features needed, so you waste some of those theoretical TFlops. Some of the GCN 1.2 features are probably not even beneficial for the madvr shaders.

I wouldn't upgrade, you probably won't notice too much of a difference. If you still want the extra power, wait for the aftermarket coolers. Sapphire has been consistently good with cooling AMD cards in the past few years, but there are small differences each generation. Stay away from Gigabyte, imho they apply too much voltage just for binning.
The previous AMD cards have been very solid performers in madVR. The 290(x), 380(x), 390(x), etc. Sure they used a lot of electricity, but they were quite fast, especially for the price. Hence, the apparently poor performance of the RX 480 is surprising considering it's a continuation of the same architecture that did so well in the prior cards. If it delivered 390 performance in madVR for the price and power consumption it would be quite a nice upgrade. But, it doesn't. :confused:

[ashlar42]

Ok, last post on the gtx1060 / rx480 topic, let's get back to madvr: (you can write me some pm if you got questions)


here is my upgraded comparison:

http://www.file-upload.net/download-11785367/MadVRComparison.xlsx.html
May I suggest some sort of color coding for solutions that are not usable? There are several 60fps instances where the ms rendering time is above the needed 16.6ms.

[huhn]

But is madVR supposed to be better than Weston at properly deinterlacing or not?

madVR is using DXVa and DXVA is based on your GPU and GPU driver.

in the past nvidia and amd had good deinterlancer. there where a lot of problem with crimson but this should be fixed yet.

they should be using adaptive interlancing with field matching which should be better than weston and YADIF.

[ionutm80]

madVR is using DXVa and DXVA is based on your GPU and GPU driver.

in the past nvidia and amd had good deinterlancer. there where a lot of problem with crimson but this should be fixed yet.

they should be using adaptive interlancing with field matching which should be better than weston and YADIF.

But in my case we are not talking about interlaced video as the flag showed by MediaInfo clearly states 25fps progressive and I can confirm that is was ripped from a PAL DVD Disney movie.
Did LAV ignored the flag and forced a deinterlacing w/o being the case?

[pose]

If it delivered 390 performance in madVR for the price and power consumption it would be quite a nice upgrade. But, it doesn't. :confused:
How do you know that?