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. |
19th July 2014, 16:38 | #26942 | Link |
Troubleshooter
Join Date: Feb 2014
Posts: 339
|
The answer to that question is yes: you should use 32-bit mpc-hc. MadVR 32-bit does not work with MPC-HC 64-bit.
__________________
System specs: Sager NP9150 SE with i7-3630QM 2.40GHz, 16 GB RAM, 64-bit Windows 10 Pro, NVidia GTX 680M/Intel 4000 HD optimus dual GPU system. Video viewed on LG notebook screen and LG 3D passive TV. Last edited by Anime Viewer; 19th July 2014 at 16:45. |
19th July 2014, 17:35 | #26943 | Link | |
Registered User
Join Date: Oct 2012
Posts: 7,925
|
Quote:
there is simply no benefited from sending a YCbCr 4:2:0 source as RGB from Lavfilter. of couse you can try things like this but in general it shouldn't be used the default setting from Lavfilter are choice like this for a good reason. the normal user should never change them if they don't know what they exactly do. |
|
19th July 2014, 17:45 | #26944 | Link | ||
Registered User
Join Date: Aug 2009
Posts: 463
|
Quote:
BTW, I advocate that people should not complicate things because making something complicated most of the time does not lead to better quality. And a lot of online tutorials are too much complicated without any solid reason. When it comes to madVR+LAV Filters combo, rule is that default settings are very good. If you want better, just use better scaling algorithm in madVR (requires high end graphic card). Also test if you can actually see improvement over default settings. Quote:
While in LAV properties dialog he must uncheck all other output formats and leave RGB checked if he want LAV decoder to do conversion. Else, LAV will pass unchaged stream, as I already mentioned. Almost everyone here loves LAV Filters but almost everyone recommends madVR for chroma (and luma) up/downscaling because it has better choice of scaling algorithms and better dithering. Last edited by detmek; 19th July 2014 at 17:53. |
||
20th July 2014, 08:40 | #26945 | Link |
Nicolas Robidoux
Join Date: Mar 2011
Location: Montreal Canada
Posts: 269
|
Mathias: Sigmoidization was definitely pointing you in the wrong direction. It's dead. (Apologies.)
On the other hand, I'm pretty excited about this: http://www.imagemagick.org/discourse...113483#p113483 |
20th July 2014, 12:50 | #26947 | Link | |
Registered User
Join Date: Oct 2012
Posts: 7,925
|
Quote:
|
|
20th July 2014, 15:35 | #26948 | Link | ||
Registered User
Join Date: Sep 2010
Posts: 321
|
Quote:
Quote:
__________________
MPC-HC/MPC-BE, Lav Filters, MadVR CPU: AMD Ryzen 5 1600, Video: AMD Radeon RX Vega 56 -> TCL S405 55", Audio: Audio-Technica M50S |
||
20th July 2014, 15:45 | #26949 | Link | |
Registered User
Join Date: Nov 2012
Posts: 167
|
Quote:
SD (srcHeight<=576) - luma image doubling with as many neurons as possible, Jinc3 for image upscaling HD (eveyrhing else/default profile) - Jinc3 for image upscaling, image doubling off You could probably just set image doubling to "if scaling factor 2.0x or higher", but I myself have some 576p videos I want doubled with NNEDI and downscaled to 1080p, hence I use profiles. Some folks here (say that they) can see a difference between using image doubling and simple Jinc3 scaling when going 720p -> 1080p, so you need to check it yourself It's also kind of hard to recommend anything without knowing what your GPU can handle, but you can't go wrong with Jinc3 for HD and image doubling for SD (if your GPU can handle it). Contrary to fairchild's recommendation I would leave "trade quality for performance options" at defaults or perhaps go there after configuring other settings. |
|
20th July 2014, 17:51 | #26951 | Link | |
Registered User
Join Date: May 2012
Posts: 447
|
Quote:
1) Convert input (a) from sRGB to linear RGB 2a) Copy (a) to (b) 2b) Apply a gamma of 3 to (b) 2c) Resize (b) using whatever method 2d) Apply a gamma of 1/3 to (b) 3a) Copy (a) to (c) 3b) Negate each channel of (c) 3b) Apply a gamma of 3 to (c) 3c) Resize (c) using whatever method 3d) Apply a gamma of 1/3 to (c) 3e) Negate each channel of (c) 4a) Set output (d) to the mean of each channel of (b) and (c) 4b) Convert output (d) from linear RGB to sRGB This then produces a result with similarly reduced ringing as if you had converted to LAB before resizing? Edit: I guess if this thread is solely for madVR support, this might need its own topic.
__________________
Test patterns: Grayscale yuv444p16le perceptually spaced gradient v2.1 (8-bit version), Multicolor yuv444p16le perceptually spaced gradient v2.1 (8-bit version) |
|
21st July 2014, 13:41 | #26952 | Link | |
Registered User
Join Date: Dec 2002
Location: Region 0
Posts: 1,436
|
Quote:
Also, I know you're officially not taking feature requests, but have you given any thought to adding a masking step in the processing pipeline. Something where a custom number of pixels on each edge could be masked/replaced with black? Something like the equivalent of crop + addborders in Avisynth. This would allow people with fixed pixel displays who are running with 1:1 pixel mapping to hide unwanted artifacts from the edges of content. Thanks! Last edited by Stereodude; 21st July 2014 at 13:47. |
|
21st July 2014, 19:27 | #26953 | Link |
Registered User
Join Date: Nov 2004
Posts: 62
|
I know that you can uncheck various parts of MadVR if you don't have enough horsepower in your CPU/GPU, but I'm building a 4k box and want to know if my i7 3820 CPU and GTX 650Ti GPU would be strong enough to scale to 4k using Jinc3 + AR filter (My Sony projector upscales well, but has too much ringing.) If this is not quite powerful enough to do this, I am open to suggestions a better GPU.
Thanks! |
21st July 2014, 21:49 | #26954 | Link | |
Registered User
Join Date: Jan 2009
Posts: 1,210
|
Quote:
|
|
21st July 2014, 22:11 | #26955 | Link |
Registered User
Join Date: Feb 2002
Location: San Jose, California
Posts: 4,407
|
My Titan can do 1080p30 -> 4K with NNEDI3 32 neuron luma doubling and Jinc3+AR for both chroma and image scaling with smooth motion on, but it is pretty busy at that point. What this says about a 650Ti doing Jinc3, I am not sure.
If you are willing to use something like bicubic75+AR for chroma and Jinc3+AR for image the 650Ti might do it but it would be close at best. |
21st July 2014, 22:32 | #26956 | Link |
Registered User
Join Date: Oct 2012
Posts: 7,925
|
a r9 280 + should be able to do it my r9 270 gets 44 ms with image only and 50ms+ with chroma and image at jinc 3 ar on a 1080p23 source. spline 3 ar is a joke for it it's about 12 ms i'm not even sure if it is in the highest powerstate.
BTW you can easily test this bypressing numblock 9 and force a fake UHD output on mpc HC/BE. for some reasons image doubling is faster with 16 neurons... |
22nd July 2014, 02:38 | #26957 | Link |
Registered User
Join Date: Feb 2002
Location: San Jose, California
Posts: 4,407
|
I did a madVR performance test on my Titan with Jinc3+AR/Jinc3+AR, smooth motion on, debanding low/low, ordered dithering. 4 frame rendering buffer, 1 present queue, windowed mode (new path), Windows 8.1. SLI Off.
1080p24 -> 3840x2160@60Hz, 17.5ms While playing Nvidia Inspector says: GPU 41-52% MCU 10-14% GPU CLK 549-679 MHz RAM 1429MB Assuming performance is directly proportional to shaders*MHz a stock GTX650Ti has 39% of the GPU performance of a Titan @ 679 MHz. It also has 28% of the memory bandwidth of my Titan, well above the 14% I used. This works out to at worst a 25.9ms rendering time on a GTX650Ti at 90% GPU usage; fine for 30fps or lower but not 60fps (16.6ms). If I use Bicubic75+AR for chroma scaling and Jinc3 for image it changes to a 15.6 ms rendering time with at most 44% GPU usage. Scaling to a 19.5ms rendering time on a GTX650Ti @ 928MHz and 90% usage. Still not enough for 60 fps but plenty of headroom for 30 fps (33.3ms). These are worst case estimates though, I wouldn't be surprised if the GTX650Ti had the GPU power for Jinc3 Image and Bicubic75 chroma for 1080p60. Sadly the memory bandwidth is probably insufficient. Memory usage is a major issue though; if I turn the buffers down to 4 rending and 1 present I still hit 1429MB of GPU memory used and 4/1 is the minimum. Edit: I was able to get memory usage down to 958MB with Jinc3+AR chroma and Jinc3+AR image by turning on most of the trade quality for performance options (10 bit instead of 16 bit) and turning off debanding and smooth motion. Last edited by Asmodian; 22nd July 2014 at 02:58. |
22nd July 2014, 04:58 | #26958 | Link |
Registered User
Join Date: Nov 2004
Posts: 62
|
Thanks, guys! You've given me some great information! I'm not looking to game in 4k at this point, just watch movies, so I could probably handle staying at 30fps for the time being. I'm much more familiar with using ATI/AMD cards for my HTPCs but I had read where the Kepler cards had beta drivers allowing them to expose 4k all the way up to 60hz over the same HDMI 1.4 card. My projector doesn't take display port, is there a good 4k option over HDMI for the R9 series from AMD?
|
22nd July 2014, 06:59 | #26959 | Link |
Registered User
Join Date: Feb 2002
Location: San Jose, California
Posts: 4,407
|
Wow, I had no idea. I happened to be running the newest Nvidia betas and just unplugged my DP 1.2 cable and plugged in an HDMI. It switched to 3840x2160@30 Hz and I have the options for 29, 25, 24, and 23 Hz in Windows.
These are great options for madVR but other then that we are going off topic (as happens all too often here). |
22nd July 2014, 18:48 | #26960 | Link | |
Registered Developer
Join Date: Mar 2014
Posts: 9
|
Quote:
Be aware that nVidia is actually using chroma downsampling (4:2:2 if I recall correctly) to squeeze 4K into the bandwidth constraints of HDMI 1.4. Which defeats the whole purpose of madVR's RGB processing. |
|
Tags |
direct compute, dithering, error diffusion, madvr, ngu, nnedi3, quality, renderer, scaling, uhd upscaling, upsampling |
Thread Tools | Search this Thread |
Display Modes | |
|
|