Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion. Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules. |
1st May 2017, 21:59 | #43562 | Link |
Registered User
Join Date: Mar 2009
Posts: 3,646
|
Can you guys stop quoting entire posts and writing one line replies to it, it's annoying as hell.
All you have to do is delete everything but what you're replying to within the quote /quote statements.. Pretty simple stuff. Last edited by ryrynz; 1st May 2017 at 22:01. |
2nd May 2017, 10:07 | #43563 | Link |
Registered User
Join Date: Apr 2017
Location: Germany
Posts: 31
|
Need help with Nvidia settings
Hi,
I have a few questions about the madVR settings with Nvidia etc and maybe someone can help me with it, because I always had a AMD card in the past and its very hard to understand all off it now, even with the wikis etc. Nvidia have several VSYNC settings, I try fast (which I use for gaming too) but get video/picture disturbances at the playback; VSYNC off make nothing no speed benefit to VSYNC on... what are you use? I'm trying a lot of settings to get a good rendering time for my 4K display with my 1060 card, what are you think about it? Config1: 1080p Chroma NNEDI3-64 / doubling xbr 100 / scaling Lanczos3 AR = ~20.45ms Config2: 1080p Chroma Lanczos4 AR / doubling xbr 100 / scaling Lanczos3 AR = ~10ms Config3: 1080p Chroma Lanczos4 AR / upscaling Jinc = ~14.80ms |
2nd May 2017, 11:53 | #43565 | Link |
Registered User
Join Date: Apr 2017
Location: Germany
Posts: 31
|
Thank you ryrynz for your help:
NGU AA low + xbr 100 + Lanczos3 AR = ~11ms NGU AA med + xbr 100 + Lanczos3 AR = ~12.80ms NGU AA high + xbr 100 + Lanczos3 AR = ~16.00ms XBR runs best as doubling on my system, I can use NGU's at low or the xbr 100... What brings more benefit the upscaling or the doubling? If I use NGU AA for the chroma upscaling which NGU I should use for the doubling? What should be the target ms at the end? Last edited by Tobiwan; 2nd May 2017 at 12:35. |
2nd May 2017, 12:42 | #43566 | Link |
Registered User
Join Date: May 2012
Posts: 447
|
I'd prioritize it like this: luma doubling > other enhancements > further upscaling/downscaling > chroma upscaling > chroma doubling. There are some situations where the chroma upscaling algorithm can make a noticeable difference, but they're few and far between, and the difference between chroma doubling algorithms will likely be invisible. The target ms depends on the frame rate of the content you watch, with 24fps content it's 41.67ms and with 60fps content it's 16.67ms. However you should probably leave 1-1.5ms of headroom. Ultimately what matters is that your buffers stay full, so just crank up your settings until your card can't keep up anymore and dial them back from there.
__________________
Test patterns: Grayscale yuv444p16le perceptually spaced gradient v2.1 (8-bit version), Multicolor yuv444p16le perceptually spaced gradient v2.1 (8-bit version) Last edited by Ver Greeneyes; 2nd May 2017 at 12:45. |
2nd May 2017, 13:46 | #43568 | Link |
Registered User
Join Date: May 2012
Posts: 447
|
Well, like I said it depends on the content. If anything over ~30ms lags then you're probably watching 30fps content (a frame every 33.33ms). You can create profiles with different settings for different kinds of content.
__________________
Test patterns: Grayscale yuv444p16le perceptually spaced gradient v2.1 (8-bit version), Multicolor yuv444p16le perceptually spaced gradient v2.1 (8-bit version) |
2nd May 2017, 15:14 | #43569 | Link |
Registered User
Join Date: Apr 2017
Location: Germany
Posts: 31
|
I'm not really sure what are you mean Ver Greeneyes. The rendering time in ms only means how fast it card works with the settings what does this have to do with the fps? Smooth motion is always on for me because my PC display doesn't support the original timings, so why should I care about the fps? Smooth motion "fix" the issue, I only need a fast and good rendering setting.
|
2nd May 2017, 16:22 | #43570 | Link |
Registered User
Join Date: Oct 2012
Posts: 7,903
|
the rendertime give on how long it takes to render a frame.
with a 24 fps source you have ~42 ms to render this frame with a 30 FPS source you have ~33 ms to render a frame and with a 60 fps source you have ~16 ms. so stable rendertime of 38 ms for a 24 FPS source should be totally stable. |
2nd May 2017, 17:06 | #43571 | Link |
Registered User
Join Date: Apr 2017
Location: Germany
Posts: 31
|
Ok thank you, I think I understand it now. If I can run a 23fps without problems I can run every fps, so I tweak here. Good.
Did I get glitches and dropped frames because the settings are to high? I already go from NGU AA high to med but glitches too. Last edited by Tobiwan; 3rd May 2017 at 13:22. |
2nd May 2017, 17:39 | #43572 | Link | |
Registered User
Join Date: May 2012
Posts: 447
|
Quote:
But the frame rate of the content you're watching determines the budget your card has to actually finish what it's doing. So with the previous example you'd be able to watch 30fps content (33.33ms budget) at full speed in both cases, but for 60fps content (16.67ms budget) the 1080p video would stutter whereas the 720p video would still be fine. So you have two options: 1) select settings that work for the highest resolution, highest frame rate content you have (e.g. 1080p at 60fps). Those settings should then work for everything else too*. 2) create settings for each kind of content you care about - one set of settings for 720p at 30fps, one set for 1080p at 60fps and so on. This is more work, but it lets you use more demanding options for lower quality content. * There is a caveat to this - if you're displaying 1080p content on a 1080p monitor it doesn't have to do any upscaling, so half the settings will go unused. In that case upscaling 720p content might be more expensive.
__________________
Test patterns: Grayscale yuv444p16le perceptually spaced gradient v2.1 (8-bit version), Multicolor yuv444p16le perceptually spaced gradient v2.1 (8-bit version) Last edited by Ver Greeneyes; 2nd May 2017 at 17:48. |
|
2nd May 2017, 18:01 | #43573 | Link |
Registered User
Join Date: Apr 2017
Location: Germany
Posts: 31
|
I try different profiles with my old AMD card but wasn't happy and give up, so I need to find anything for all. My display is 4K, everything is upscale from 1080p to 2160p. That's the point about my question "What brings more benefit the upscaling or the doubling?", because its already doubled on my display and if I understand it right its two times doubled 1x upscaling from 1080p to 2160p + 2160p doubled with XBR...
My old config runs not that bad: chroma Lanczos3 AR + xbr 100 doubled = ~12.30ms with peaks of 41ms max... but did I really need the two or quad doublings? I have no idea why the are presentation glitches 105 / 9min the are on every setting, I also check the general settings etc and try a bit. Last edited by Tobiwan; 2nd May 2017 at 19:00. |
2nd May 2017, 19:30 | #43574 | Link |
Registered User
Join Date: Feb 2002
Location: San Jose, California
Posts: 4,406
|
You don't want extra doubling, super sampling video has very limited benefits and some costs. Unless you are using the super sampling option in madVR's image doubling (not recommended) you don't get any extra doubling, the doubling algorithm is simply the algorithm used to do the 1080p to 2160p resize.
When using your old config and watching 1080p the video was simply doubled to 2160p using xbr 100, Lanczos3 did not to do anything after that because you wanted to display it at 2160p. If you needed to upscale it a little bit after doubling (e.g. if it was cropped 1080p) Lanczos3 would be the method used to upscale it further. edit: The pure doubling algorithms (NGU, NNEDI3, xBR) can offer better quality than the equation based methods like Jinc, Lanczos, Cubic, etc. because they can treat edges differently and other similar tricks involving more complex models of image content.
__________________
madVR options explained Last edited by Asmodian; 2nd May 2017 at 19:33. |
2nd May 2017, 20:16 | #43575 | Link |
Registered User
Join Date: Apr 2017
Location: Germany
Posts: 31
|
Very interesting Asmodian! Any idea why I have the presentation glitches and dropped frames? Its not any of the algorithms settings, not separate devices settings, vsync is on with present a frame for every VSync... Display: "vsync 16.67ms, frame 41.71ms" Nvidia driver is 381.89 latest
|
2nd May 2017, 22:00 | #43576 | Link |
Registered User
Join Date: Mar 2009
Posts: 3,646
|
Are we talking only a few glitches and dropped frames or all the time? Are you in Adaptive mode for your Nvidia graphics card?
Upload your files somewhere else, it takes days to get approved here. For starters reset madVR to defaults using the batch file inside the directory and test that, if that doesn't change anything uninstall and reinstall your drivers then report back. Last edited by ryrynz; 2nd May 2017 at 22:02. |
3rd May 2017, 00:54 | #43577 | Link |
Visual Novel Dev.
Join Date: Nov 2015
Location: Bucharest
Posts: 200
|
Seems that I've found a very happy place with madVR settings for my rather lower quality, not so well encoded content (anime music videos).
The settings are basically debanding to max, some edge thinning and everything running in soft cubic. For movies it looks a bit too soft, but for anime music videos and especially for those that weren't properly encoded, it looks awesome!
__________________
Visual Novels! https://www.facebook.com/seventh.heart.studios/ https://audiophile-heaven.blogspot.com/ https://www.youtube.com/watch?v=27VWU3ydRY8 |
3rd May 2017, 13:10 | #43578 | Link |
Registered User
Join Date: Jan 2017
Posts: 107
|
Hi all,
I have a question on HDR (maybe on metadata). I have a amd rx460 When i try to play HDR video with madvr seems that something is not going as I expected . Image looks washed out and color not correctly saturated. Even Tv switch correctly to HDR mode. I start to think metadata are not sent to TV, because if i check / uncheck "send metadata.." nothing changes. (obviously win10 api is selected) If if play that file using win10 tv and video player looks gorgeous: deep black , bright whites and colored. If turn off HDR in win10 slider i got totally wrong image in win10 tv and video player and in madvr i got a good sdr version of video. How can I debug this behavior ? |
3rd May 2017, 13:18 | #43579 | Link | |
Registered User
Join Date: Apr 2017
Location: Germany
Posts: 31
|
Quote:
I also find good stats for me: chroma NGU AA high + xbr 100 double = ~16-27ms / 23.976fps smooth motion on At the old configs there was always double+quad xbr active, result was 40+ms, now without quad its ok for me I think. Last edited by Tobiwan; 3rd May 2017 at 13:35. |
|
Tags |
direct compute, dithering, error diffusion, madvr, ngu, nnedi3, quality, renderer, scaling, uhd upscaling, upsampling |
Thread Tools | Search this Thread |
Display Modes | |
|
|