madVR - high quality video renderer (GPU assisted)

[QBhd]

Originally Posted by bytheway

Is there a current favorite GPU for using madVR at "maximum" that is also quiet (doesn't have to be silent) and has a reasonable TDP? Priority is excellent 23.976p/24p on a Sony TV that supports 1080p/24p. Thanks.

My AMD R9 270X is super quiet in my enclosure (HAF 932)

https://www.asus.com/ca-en/Graphics_...9270XDC2T2GD5/

It's more than enough to get madVR going and still be quiet. Just be sure to use the 13.12 drivers though, they are still king!

QB

[MSL_DK]

Originally Posted by QBhd

Just be sure to use the 13.12 drivers though, they are still king!

QB

Why is that? In what way?

[a8213711]

What's the worst upscaling in terms of quality?
I want to use it to compare it with the one I use now (chroma and image up: lanczos 3 taps + AR).
For the moment I used bilinear + bilinear and these are the result with 2 different file:
http://www.mediafire.com/view/uebjbagjb9vnnf2/a%2C 1.png
http://www.mediafire.com/view/1asqfiruye3v7j2/a%2C 2.png
http://www.mediafire.com/view/1z9v92xqg20vtje/b%2C 1.png
http://www.mediafire.com/view/rzwfq7gatuzrdbw/b%2C 2.png
(used Win8's win+stamp because the screenshots in MPC have the resolution of the file instead of the output, don't even know if it's normal)
Maybe I have too much trade-off? Because the difference is barely visible (do you want to guess which one uses lanczos? )
Tradeoff I have: http://i.imgur.com/PLb5LCM.jpg
By the way these are the settings that were already checked right after a clean install...I suppose madVR has some pre-built profile for specific hardware, and I didn't changed them.

[nevcairiel]

Nearest Neighbour is the worst quality.

[meetajhu]

Originally Posted by djsolidsnake86

any chance to see madvr work with MPC 64bit?

Madshi has already stated before there are no plans for x64 builds currently until the x86 one becomes stable.

[madshi]

Originally Posted by makz

When I playing AVSHD 709 disk debug OSD say limited range

Weird. Ok, one more thought: Is it possible that your media player has color or saturation controls set to non-neutral values?

Originally Posted by 6233638

It may not be madVR's fault, but I've definitely had sync issues without the "delay playback" options enabled.
I have not tested it recently, but it seemed to mainly occur when going from windowed to fullscreen where the refresh rate changes.
I have madVR set to only change when entering full-screen, and use smooth motion when running in a window at 60Hz.

I tried disabling the option again back when I got a GTX970 because I really like how much more responsive it is when you're not waiting for 128 frames to queue up, but I ran into sync problems again.
I was never able to figure out if there was a specific delay or if it only happened some of the time with certain formats for example, but any time I don't have the "delay playback" options enabled, I end up running into sync issues that are corrected by re-enabling it.

As far as I see, there's no technical reason why you should have sync problems without delay, but no sync problems with delay. Actually the delay option is quite "hacky". So if there's a problem, I would have thought it more likely that the delay problem causes a problem. Anyway, after a couple of seconds there should be no difference between delay on/off, so I'm really confused as to where those problems could be coming from. You have the same problems regardless of rendering mode (FSE vs. windowed vs. overlay, "new" presentation path vs. "old")? I think the issue is worth looking into, at least if you can reliably reproduce it.

Originally Posted by ryrynz

Submitted a crash dump not long ago via MadVR, 0.87.14 is crashing in MPC-BE when subtitles are set to XYSubfilter upon playback of a file with subtitles, 0.87.13 doesn't crash.

Yeah, will try to fix that for the next build.

Originally Posted by anuswara

Hi, I 'd linke to know if this madVR can fix my issue.
Monitor. 60Hz (unpossible to change this value!).
In both interlaced and progressive files I am unable to read the credits at the end, crawl.

These credits are increditbly fast. Some displays show such credits better than others. You'll get the best results with a display which is able to interpolate intermediate frames. But then I wonder if those algorithms are able to handle such quickly moving credits. Don't really know. Your best bet is to make sure you're playing the movies at the correct refresh rate. If you do that, you can only hope that your display handles them well. If all else fails, you could try playing with SVP. That might help with these credits. But I wouldn't put too much weight on the credits alone. I wouldn't really activate motion interpolation just for those. If you like the look of motion interpolation / SVP, then it's quite alright to use it, of course.

Originally Posted by Asmodian

Edit: After making the chart I again thought it would make more sense to use chroma upscaling instead of image upscaling to scale up the chroma when only doubling luma. It is, at least, easier to explain.

Ideally I could set a second chroma scaling option in image doubling, including NNEDI3, similar to the chroma scaling settings page.

Well, the "chroma upscaling" section is only for getting from 4:2:0 or 4:2:2 sources to 4:4:4. Once we're at 4:4:4, everything else is considered "image" instead of "chroma". I understand that it's slightly confusing that when using doubling, I'm using the "image upscaling" algorithm for upscaling chroma. But that's just because NNEDI3 is so damn slow than it doesn't make sense to use it on the chroma channels. So the whole concept of going from RGB to YCbCr and scaling Y separately from CbCr is just due to NNEDI3's slow speed. I still consider the whole operation to be "image upscaling", even if madVR applies different algorithms to Y and CbCr in that situation. But I do understand the potential for confusion.

Of course I could add even more settings to allow the user to choose just another algorithm for CbCr upscaling when using NNEDI3 for Y. But really, there are already more than enough options there. Adding even more might make it more logical, but also more complex.

Originally Posted by FireFreak111

Well, news on the OpenCL front for Nvidia, 349.72 is coming, primarily for Windows 10 with the new WDDM 2.0 driver model support, DirectX 12 and, surpisingly, OpenCL 1.2.

Oh, that would be cool!

Originally Posted by ryrynz

Madshi could you edit the change log as well?

I don't find any "Strg" in changelog.txt?

Originally Posted by Pat357

I have a problem that I can't see or track my mouse movements in FSE mode.
There is no typical mouse "arrow" visible any more as soon as FSE is enabled.
When moving my mouse blindly, I see the typical MadVR FSE time-bar, and clicking starts seeking.
So this part seems to work, but I can't choose to what time I seek because the mouse is completely invisible.

Strange, nobody else seems to have this problem. Could you try different media players, or maybe a different MPC-HC build to check if that might make a difference?

Originally Posted by bytheway

Is there a current favorite GPU for using madVR at "maximum" that is also quiet (doesn't have to be silent) and has a reasonable TDP? Priority is excellent 23.976p/24p on a Sony TV that supports 1080p/24p. Thanks.

No GPU today can run NNEDI3 at 256 neurons for 1080p -> 4K upscaling. It will probably a while before that's possible. But it's probably also not necessary. The increase in image quality when doubling neuron count is there, but not very high. And the cost doubles every time.

So what am I saying? The key thing is that you need to make up your mind what you understand as "maximum" settings. (Note: I don't really want an answer to the following questions). Which scaling algorithms do you want to use? When using NNEDI3, how many neurons do you want to use? And do you only watch 24fps movies? Or also 60i/p content? Which resolution does your display have? And which resolution your movies? Only with all that information we could maybe provide you with good advice. But even then, I personally wouldn't really know which GPU to recommend. There are pros and cons for most of today's GPUs. E.g. the 750 (Ti) is a nice GPU, but has no HEVC decoders and not as fast as the 960. The 960 is nice, and has HEVC decoders, but is not as fast as the 970/980. The 970/980 are quite fast, but they don't support HEVC hardware decoding. AMD also has nice GPUs which have their own list of pros and cons. So which should I recommend? I've no idea.

Originally Posted by tfouto

There will be a 64 bit version of madvr?

With HEVC decoding, 64 bit is really important...

Originally Posted by djsolidsnake86

any chance to see madvr work with MPC 64bit?

Originally Posted by Asmodian

madshi has said one day there may be a 64-bit madVR but not soon.

Originally Posted by meetajhu

Madshi has already stated before there are no plans for x64 builds currently until the x86 one becomes stable.

Well, I said that at a time when there was no big benefit using x64. Recently priority has changed a bit because the x64 HEVC decoder is clearly faster than the x86 decoder. So what does that mean? I don't know, but x64 support has at least moved up a bit in my priority list.

[MS-DOS]

Originally Posted by madshi

x64 support has at least moved up a bit in my priority list.

Wish I could add a few more bits there =) Currently, the lack of 64 bit madVR is the only reason why I'm still using the 32 bit versions of MPC, LAV and some other stuff.

[Asmodian]

Originally Posted by madshi

Well, the "chroma upscaling" section is only for getting from 4:2:0 or 4:2:2 sources to 4:4:4. Once we're at 4:4:4, everything else is considered "image" instead of "chroma". I understand that it's slightly confusing that when using doubling, I'm using the "image upscaling" algorithm for upscaling chroma. But that's just because NNEDI3 is so damn slow than it doesn't make sense to use it on the chroma channels. So the whole concept of going from RGB to YCbCr and scaling Y separately from CbCr is just due to NNEDI3's slow speed. I still consider the whole operation to be "image upscaling", even if madVR applies different algorithms to Y and CbCr in that situation. But I do understand the potential for confusion.

After a better understanding of what is going on I am more comfortable with the way it is now.

It does take a little explaining but it makes sense to keep all the algorithms the same after passing the first conversion to RGB. They would have to be the same if not using image doubling. If you listened to me I would probably come back and ask for a way to use the alternative chroma scaling even when not using image doubling.

Originally Posted by madshi

Well, I said that at a time when there was no big benefit using x64. Recently priority has changed a bit because the x64 HEVC decoder is clearly faster than the x86 decoder. So what does that mean? I don't know, but x64 support has at least moved up a bit in my priority list.

for the response.

[bytheway]

Originally Posted by madshi

No GPU today can run NNEDI3 at 256 neurons for 1080p -> 4K upscaling. It will probably a while before that's possible. But it's probably also not necessary. The increase in image quality when doubling neuron count is there, but not very high. And the cost doubles every time.



So what am I saying? The key thing is that you need to make up your mind what you understand as "maximum" settings. (Note: I don't really want an answer to the following questions). Which scaling algorithms do you want to use? When using NNEDI3, how many neurons do you want to use? And do you only watch 24fps movies? Or also 60i/p content? Which resolution does your display have? And which resolution your movies? Only with all that information we could maybe provide you with good advice. But even then, I personally wouldn't really know which GPU to recommend. There are pros and cons for most of today's GPUs. E.g. the 750 (Ti) is a nice GPU, but has no HEVC decoders and not as fast as the 960. The 960 is nice, and has HEVC decoders, but is not as fast as the 970/980. The 970/980 are quite fast, but they don't support HEVC hardware decoding. AMD also has nice GPUs which have their own list of pros and cons. So which should I recommend? I've no idea.

Thanks, to clarify, I'm mostly interested in Blu Ray image quality (most are 1080p24) on a 1080p TV (no 4k) and second priority would be Australian TV which is 576i50, 720p25 and 1080i50. For Blu Ray I would like to be able to match the latest Oppo Blu Ray player. Any recommendations for a GPU and CPU for this? If it has low TDP that's a bonus. Thanks again.

[bytheway]

Originally Posted by QBhd

My AMD R9 270X is super quiet in my enclosure (HAF 932)



https://www.asus.com/ca-en/Graphics_...9270XDC2T2GD5/



It's more than enough to get madVR going and still be quiet. Just be sure to use the 13.12 drivers though, they are still king!



QB

Are these level cards along with the GTX 960/970 overkill if not gaming or are they precisely the way to get the best video quality with madVR? Does madVR get used for gaming? (I don't play games)

[ryrynz]

Originally Posted by madshi

I don't find any "Strg" in changelog.txt?

Ah, didn't check the actual changelog, assumed it was the same as the original 0.62 post which wasn't updated and still references Strg. Also the changelog on videohelp still references strg for 0.62 and 0.63, I'll get them to change it.

[huhn]

Originally Posted by bytheway

Are these level cards along with the GTX 960/970 overkill if not gaming or are they precisely the way to get the best video quality with madVR? Does madVR get used for gaming? (I don't play games)

this has nothing to do with gaming.

some scaler in madVR can be so demanding that even a GTX 980 can't handle them.

it is simply up to what you want to use and what YOU think looks best to your eyes using your TV. a r7 250 can already be the best choice for you.

[MistahBonzai]

Just something to think about here..we're at a turning point where 4K starts to go mainstream in video cards. If considering tying the video card/PC into a '4k HT setup' sooner or later don't overlook 4K compatibility issues. Current cards support HDMI 1.4 which will do 3820x2160/60 but at up to 4:2:0. HDMI 2.0 has the BW to support that at 4:4:4 10-bit.

The first cards with native HDMI 2.0 appear to be in the AMD R 300 series - due to ship within a few months - with Nvidia to follow latter on. The other 'features' to look for are hdcp 2.2 and native HEVC H.265. Of course this is all predicated using a model of the very latest generation of 4K TV panels.

As it stands today deciding later on that this would be doable by getting the appropriate 'adapter' will not be easy (if possible).

[MistahBonzai]

Originally Posted by MS-DOS

Wish I could add a few more bits there =) Currently, the lack of 64 bit madVR is the only reason why I'm still using the 32 bit versions of MPC, LAV and some other stuff.

I've been farting around with the 10bit h265 Astra.SES.Demo.HEVC.mp4 file. The 64-bit MPC-HC/LAV version plays it fine while it's a slide show with the latest 32-bit version. The difference on my i7-3770 3.4 GHz is like 50% CPU w/64-bit and 100%/slide show (dropping better than half the frames) with 32-bit (both using EVR(CP).

[ryrynz]

Originally Posted by madshi

x64 support has at least moved up a bit in my priority list.

I would hope it's near the top, for many users it most certainly is of the highest priority. It's usually good to plan forward a bit.

[leonccyiu]

Originally Posted by MistahBonzai

Just something to think about here..we're at a turning point where 4K starts to go mainstream in video cards. If considering tying the video card/PC into a '4k HT setup' sooner or later don't overlook 4K compatibility issues. Current cards support HDMI 1.4 which will do 3820x2160/60 but at up to 4:2:0. HDMI 2.0 has the BW to support that at 4:4:4 10-bit.

The first cards with native HDMI 2.0 appear to be in the AMD R 300 series - due to ship within a few months - with Nvidia to follow latter on. The other 'features' to look for are hdcp 2.2 and native HEVC H.265. Of course this is all predicated using a model of the very latest generation of 4K TV panels.

As it stands today deciding later on that this would be doable by getting the appropriate 'adapter' will not be easy (if possible).

Are you sure HDMI 2 has the Bandwidth to support 4:4:4 10-bit? I thought the maximum was 4:2:2 10/12bit or 4:4:4 8-bit.

Do the Nvidia GTX 9 series not have native HDMI 2 support?
I look forward to seeing if the AMD R 300 series also include VP9 decode support, if they move to a 20nm process and if they perform much better than the existing cards as there are more stream processors.

I'd be grateful if Nvidia GTX 960/970/980 users could post some of their nnedi3 benchmarks. bcec has already done so with his 960 on avsforum, he had a render time of 35ms with nnedi3@32 with lanczos3ar so 30p content is only possible with 16 neurons.

It was reported earlier that AMD cards perform better than their nvidia equivalents for nnedi3 die to their stronger compute and opencl performance, it seems while nvidia have closed the gap, AMD still have the lead.

[makz]

Originally Posted by madshi

Weird. Ok, one more thought: Is it possible that your media player has color or saturation controls set to non-neutral values?

In MPC-HC and MadVR neutral values.

[ryrynz]

Originally Posted by leonccyiu

I'd be grateful if Nvidia GTX 960/970/980 users could post some of their nnedi3 benchmarks. bcec has already done so with his 960 on avsforum, he had a render time of 35ms with nnedi3@32 with lanczos3ar so 30p content is only possible with 16 neurons.

The post leonccyiu refers to is here and is a 1080P to 2160P upscale @ 24FPS. More results here.

[a8213711]

Is there a way to enlarge the font of the output of ctrl+j?

[Stereodude]

Originally Posted by leonccyiu

I'd be grateful if Nvidia GTX 960/970/980 users could post some of their nnedi3 benchmarks. bcec has already done so with his 960 on avsforum, he had a render time of 35ms with nnedi3@32 with lanczos3ar so 30p content is only possible with 16 neurons.

It was reported earlier that AMD cards perform better than their nvidia equivalents for nnedi3 die to their stronger compute and opencl performance, it seems while nvidia have closed the gap, AMD still have the lead.

I'd say AMD is still way out front. Assuming bcec's numbers aren't messed up a GTX 960 is only marginally more capable at NNEDI3 than a R7 260x. 28ms vs. 31.1ms (Renethx) for the same settings scaling 1080p to UHD.

The GTX 960 is way more expensive than the 260x and has a higher TDP.