Should a new TV do all this fps: 23.976p, 24p, 25p, 29.97p, 30p, 48p, 50i, 60i, 50p, 60p, 72p, 120p ?

Recommended refresh rate depends on the content:

24p and telecined 60i: 24Hz (then the display processes the incoming video and could display it at 48Hz, 72Hz, 96Hz, 120Hz etc.)
25p and telecined 50i: 25Hz or 50Hz
30p: 60Hz
50p and Interlaced 50i: 50Hz
60p and interlaced 60i: 60Hz.

Is it possible to watch 4096 x 2304 x264@80Mbps downscaled to 1080p today?
What about HVEC?
What about xvYCC (4K mastered Blu-Ray's)?
Is NNEDI3 doable at this time?

Depends on the downscaling algorithm and your graphics card. If you select a cheap algorithm (bilear, bicubic, CR without anti-ringing, without linear light), downscaling is usually no problem. '80Mbps' is irrelevant here, it's relevant to decoding process.

Decoding 4K HEVC is difficult for the current crop of hardware. There is no HEVC hardware decoder in GPU yet. Even Haswell Core i7 can't decode some of them by brute force. Some software decoder (Strongene PC OpenCL HEVC/H.265 Decoder) offloads CPU by OpenCL (so that Core i7 + Radeon card is enough), but it's not mature yet (it crashes with some HEVC files).

As for deep color, madVR output only 24bit.

If you use NNEDI3 implemented in madVR, it's easy with a proper graphics card.

Core i7-4790K + R9 270X (MSI) is a good choice. You can upgrade GPU when 20nm graphics cards are available (hopefully that support HDMI 2.0 and HDCP 2.2).

 

As for deep color, madVR output only 24bit.

So xvYCC is reladet to color deep? I cant find something about bit on xvYCC.

If you use NNEDI3 implemented in madVR, it's easy with a proper graphics card.

Form the articles i saw it seems that NNEDI3/image doubling/error diffusion kills the biggest graphics card?

Core i7-4790K + R9 270X (MSI) is a good choice.

Is there a difference when it comes to 2GB vs 4GB RAM graphic cards on HTPCs?
But that CPU is really an expensive one?

madVR settings:
I changed the profile to just

if ((srcWidth >= 1921) or (srcHeight >= 1081)) "UHD"
else if ((srcWidth = 1920) or ((srcHeight >= 721) and (srcHeight = 1280) and (srcWidth

 

So xvYCC is reladet to color deep? I cant find something about bit on xvYCC.

Form the articles i saw it seems that NNEDI3/image doubling/error diffusion kills the biggest graphics card?

Is there a difference when it comes to 2GB vs 4GB RAM graphic cards on HTPCs?
But that CPU is really an expensive one?

Yup, I am confused. Look at these posts in doom9's forum.

Error Diffusion loads GPU a lot. I wouldn't use it with NNEDI3. You'd better stick to Ordered Diffusion (madVR's default), the difference is very little in most cases anyway.

2GB (or even 1GB) video RAM is enough.

I suggested Core i7 just in case you are interested in HEVC decoding at this time. Another usage of Core i7 is SVP (Smooth Video Project). MadVR itself uses only GPU, CPU is mostly irrelevant.

 

OP: Are you using a script to change the settings in MadVR based on source input resolution? I would like more information about that if you are. Thanks..

 

@Sammy2,
here you can see how to make profiles (middle of the page): https://imouto.my/madvr/
And here some scripting stuff: http://forum.doom9.org/showthread.php?p=1271417#post1271417

I made a profile group in "scaling algorithms" and one in "processing" for "artifact removal".

 

Links bookmarked. VERY USEFUL information!! Thanks for sharing..

 

I see. I didn't know that this was an option now; especially setting up madVR outside a playback scenario.

I will need to investigate this further but most, not all, of my content is 1080p; about 99.5% is 720p/1080i/1080p so will this help me much?

 

Should I consider upgrading my 7850 to a 270x?

 

270X = HD 7870 (with minor tweaks), that's roughly 1.5 times faster than HD 7850. But you'd better continue to use 7850 and save your money for a Rx 3xx card (Pirate Islands series, manufactured in 20nm process), unless you have difficulty to do some algorithm (that you think a must). A rumor says R9 370X comes first.

 

OK. I'll wait as I'm not really having issues with this card anyhow..

 

Hi10p is done by the CPU and also 4K-H.264. What about 4K-Hi10p - what CPU does handle that?

Decoding 4K AVC by CPU is generally a lot harder. You'd better get Core i5 / i7.

The other question is FPS. Is this just a GPU thing or is it linked to what processor does the decoding? For example i have a 4K-H.264-AVC-119.880fps.mkv. Is the load on the GPU/CPU related to the frames the TV can do native?

FPS is the number of frames per second stored in the video file. Once decoded (by CPU or ASIC in GPU), it's an uncompressed video stream with that number of frames per second. Your desktop refresh rate (may not be necessarily supported by GPU / display device, of course) has to be the same as that (or an integer multiple of that) for smooth playback. This applies to progressive contents. For interlaced contents, see below.

Don't confuse fps in video with fps in gaming. In gaming, a higher GPU can produce higher fps. In video fps is fixed. (You can increase frame rate of the video for smoother motion by interpolation e.g. using SVP, of course. That's another story.)

Profiles

Basically you will need the following profiles, apart from UHD:

- SD 24fps (NTSC DVD/broadcast movies)
- SD 60fps (NTSC DVD/broadcast videos)
- HD 24fps (broadcast/downloaded movies; a broadcast 60fps movie is so by repeating frames in 3:2 pattern, so called '3:2 pulldown': 60 = (3+2)/2 * 24)
- HD 60fps (broadcast/camcorder videos)
- FHD 24fps (BD/broadcast movies)
- FHD 60fps (BD/broadcast/camcorder videos)

60fps is 2.5 times harder to process by madVR than 24fps (60/24=2.5). That's why 60fps is separated from 24fps.

Interlaced contents

60i (=30fps with interlace flag; some people also use 30i) content (NTSC DVD/broadcast, FHD broadcast, camcorders) will become either 24fps (for film-based contents) or 60fps (for video-based contents) by detelecine (by madVR or GPU)/deinterlacing (by GPU), then processed by madVR. That's why there is no need to create profiles for interlaced contents. MadVR can't detect film/video, so you have to add the tag "deint=film" or "deint=ivtc" in the file name of a film-based content manually. If there is no tag and madVR thinks it is interlaced (the information is obtained from the source filter), it leaves detelecine/deinterlacing to GPU (unless you change madVR settings). The current GPU is intelligent enough to detect film/video and detelecine/deinterlace correctly. Unfortunately it can't communicate well with madVR so that there is no way for madVR to change the fresh rate to 24Hz for films, resulting in pulldown judder. So leave detelecine to madVR and deinterlacing to GPU.

'Detelecine' is used by HandBrake, it is usually called IVTC (inverse telecine).

UHD

Most of the current UHD contents (encoded in AVC) are either 24fps or 30fps. So a single profile should be enough. In future UHD Blu-ray will be HEVC 24fps (and higher) and broadcast UHD will be HEVC 50fps/60fps.

 

A note: you can't overclock AMD GPU if you use UVD, you have to use software decoder or QS.

 

Just saw that the Tonga R9 285 is coming.
Maybe its an option if the price is right or maybe the price of the 270 will go down a bit.

And i just asked about QuickSync at Doom9:
http://forum.doom9.org/showthread.php?p=1691611#post1691611

 

Any 'dedicated' GPU also has ASIC for video decoding. You can use either that or QS. MadVR's video processing uses GPU's shader units, that's independent of the decoding ASIC. Reasons to use QS instead of AMD's UVD (Universal Video Decoder) are:

- 4K H.264 decode is supported by QS but not by UVD
- Overclocking AMD GPU is possible only when you don't use UVD. Use QS instead.

 

Ok, something new learned. Letting the GPU or CPU decode does not affect madVR performance...
I thought if i use the CPU for decoding then i free up resources for madVR.

I'm looking now at i3-4360 and i5-4570S - big price difference.
CPU Benchmark says 5,579 vs 6,860.
I wonder of how much difference the 1281 points make when it comes to HTPC.

 

I'm looking now at i3-4360 and i5-4570S - big price difference.
CPU Benchmark says 5,579 vs 6,860.
I wonder of how much difference the 1281 points make when it comes to HTPC.

Depends on your usage. For example, i3 can't transcode two FHD videos simultaneously, while i5 can. "S" or "T" is almost pointless, it does not save power at all at idle, it consumes slightly lower power because of lower clock/voltage. Just go for a normal version for better performance.

 

Sure the T version does not really make sense because it is allot slower then the new S and the price on all (T, S, non T/S) is the same now.

In idle you don't save power. But thats a HTPC and when is it idle?
I mean you have it on to watch movies and then you trough your Hi10p on the CPU.
The S version has 20 watt less then the normal version and here the CPU Bench is just 208 points apart.
The price is the same so i think the S version is the right one.

 

You can turn a normal model into S/T easily by lowering the max clock multiplier in BIOS setup. You can't do the reverse because S/T's max clock is capped. Just buy a normal version.