My collection of short anime reviews

Shortlink of this post – http://wp.me/PrgSo-b1 (use this link in your forum/BBS/IRC/twitter/facebook etc. messages) || Download the PDF version of this page here.

TABLE OF CONTENTS

INTRODUCTION
CHANGELOG
CHECKLIST
–> TUTORIAL <–
COREAVC CUDA IN ACTION

INTRODUCTION

Inside my video playback article here, you will be able to play high definition H.264 videos with negligible CPU utilization using Microsoft's DirectX Video Acceleration (DXVA) API. All that is needed is a compatible video card, Microsoft Windows XP/Vista (and even Windows 7 with my tests) and Media Player Classic – HomeCinema (MPC-HC). But there are limitations with the DXVA method, and they are (amongst others):-

• Only certain renderers can be used depending on operating system (overlay mixer and VMR9 for Windows XP, Enhanced Video Renderer for Windows Vista/7).
• Cannot decode videos that was encoded with numbers of reference frames that exceed certain levels.
• Cannot have any intermediate filters (such as ffdshow Video Decoder or DirectVobSub) between the video decoder and the video renderer.

For watching anime, the last 2 facts are very important. Fansub groups tends to use high numbers reference frames because animations benefits from it unlike live-action Hollywood movies that can get away with lower reference frame count. Plus, watching anime on certain type of LCD monitors, especially twisted nematic (TN) LCDs and you will see gradient banding which can only be corrected by running the deband post-processing filter in ffdshow Video Decoder.

Now, with the release of CoreAVC 2 with Compute Unified Device Architecture (CUDA) support (hey did this sound like an advertorial?), those limitations will go away. Of course, this comes with a couple of new limitations:-

• Only nVidia cards that supported CUDA and has PureVideo 2, 3 and 4 can use the new CoreAVC 2 with CUDA. ATI/Intel users are out of luck.
• CoreAVC 2 is not free (cost USD12.95) unlike the free and open-source MPC-HC internal decoder.

About the qualifying cards, all nVidia cards mentioned in my DXVA tutorial can use CoreAVC CUDA, so if you have been following that post and has a nVidia card that are usable with that tutorial, there is no need to buy a new card. All you have to do is to upgrade your driver to version 191.07 (for laptop users, get the modified drivers from http://www.laptopvideo2go.com/drivers) or newer and you should be set. Below is a tentative list of nVidia GPU that can use H.264 CUDA acceleration.

nVidia:

Source: http://en.wikipedia.org/wiki/NVIDIA_PureVideo – any of the VP2, VP3 and VP4 GPU can be used. When buying, make sure it has at least 256MB of video RAM (512MB if you use Windows XP). VP3 GPUs cannot decode videos with the following widths: 769-784, 849-864, 929-944, 1009-1024, 1793-1808, 1873-1888, 1953-1968, 2033-2048 pixels.

Recommended nVidia cards:-

If you do not plan to play video games:

Galaxy GeForce GT220 1GB DDR2 PCI-Express – A PureVideo4 GPU that should be powerful enough even for interlaced H.264 AVC videos.

If you plan to play video games too (and money is no object):

Galaxy GeForce GTX460 1GB GDDR5 PCI-Express – A PureVideo4 GPU that is not only capable of playing H.264 AVC videos (progressive and interlaced), but also playing H.264 MVC videos in 3D Blu-ray movies. In SLI mode, can play almost all games out there even in FullHD resolution. Oh BTW, unlike the earlier PureVideo4-powered GPUs such as nVidia G210/GT220/GT240/GTX480/GTX470/GTX465, nVidia GTX460 can also bitstream high-definition audio codecs such as Dolby TrueHD and DTS-HD MA/HR.

One important fact that you should take into consideration when buying suitable nVidia GPU for the purpose of hardware acceleration is the amount of video RAM it has. More is always better. A FullHD video with 16 reference frames played at ~24fps can easily take up 180MB of video RAM when using EVR custom presenter with 5 EVR buffers. The internal subtitle engine also use video memory, which quantity depends on the rendering resolution and amount of subpicture buffering. Both stack up on top of what have been taken by Windows Vista/7 Aero feature.  512MB is my minimum recommendation, while 1GB should suffice for even 1080p videos with 16 reference frames playing at 60fps. Do not go overboard with the EVR buffers/subtitle engine subpicture prebuffering though.

CHANGELOG

- 08 March 2010: A screw-up of the highest order in my part caused this post to be nuked to oblivion. Therefore I have to rewrite this thing from scratch.

THE GUIDE OF USING COREAVC 2.0 CUDA TO ACCELERATE H.264 VIDEOS

The checklist of what is needed:-

1. Media Player Classic – HomeCinema (MPC-HC) – download from here (mirror if first link fails). MediaInfo dll is now included.
2. CoreAVC 2 Professional Edition – available from my affiliate link here for USD12.95. A trial version is not available yet. You can use the trial version first to check whether it will work on your nVidia system or not. If this guide works as expected on your system, you can buy the full version at their website, preferably via the affiliate link above.
3. Haali Media Splitter – The version inside CoreAVC 2 installer is now outdated. Download the latest version from here. Original website.
4. nVidia 191.07 drivers or later – get it via Windows Update (like me) or get it from nVidia website here. Laptop users can get their drivers (if nVidia dragged their feet) here.

For testing and troubleshooting, you will need to download:-

1. DXVA Checker - Download here (original website).
2. GPU-Z – Download from this website.
3. DirectX End-user Runtime – Download from here. Download and install if you have d3dx9.dll error message when you start MPC-HC.
4. Microsoft .NET Framework 4 – Download from here. For Windows XP users who wants to use vanilla EVR/EVR custom presenter and doesn't have the framework installed yet. Also needed to run DXVA Checker above.

Test videos that you can use for acceleration tests.

1. The Melancholy of Haruhi Suzumiya ED theme – Hare Hare Yukai (H.264 SD video with karaoke subtitles) – Download here.
2. Clannad ED theme – Dango Daikazoku (H.264 720p video with karaoke subtitles) – Download here.
3. AIR OP – Tori no Uta (H.264 HD 1080p video with karaoke subtitles) – Download here.
4. Kanon OP – Last regrets (H.264 HD 1080p with soft-subtitles) – Download here.
5. Planet Earth: From Pole to Pole (H.264 HD 1080p with subtitles) – Download here.

Download all 5 test videos (and one more free) above in one single archived zip file here (806MB). No need to wait 10 minutes to download the separate files individually. The links above still can be used if you do not want all of them.
EDIT: IT COMES TO MY ATTENTION THAT YOU WILL NEED A PREMIUM ACCOUNT TO DOWNLOAD FILES OVER 400MB. IF YOU DO NOT HAVE ONE, YOU CAN STILL DOWNLOAD THE VIDEOS INDIVIDUALLY FREE OF CHARGE!!

Make a comment below if any of these files cannot be downloaded.

TUTORIAL (NOW MADE SIMPLER AND FAR LESS INTIMIDATING THAN EVER BEFORE)

JUMP TO: 

• STEP 1 – INSTALLING MPC-HC.
• STEP 2 – INSTALLING COREAVC 2.0 CUDA AND HAALI MEDIA SPLITTER.
• STEP 3 – OPENING MPC-HC OPTION WINDOW.
• STEP 4 – ENABLING MPC-HC INTERNAL SUBTITLE ENGINE.
• STEP 5 – CHOOSING YOUR VIDEO RENDERER.
• STEP 6 – CONFIGURING MPC-HC INTERNAL FILTERS.
• STEP 7 – ADDING COREAVC 2.0 CUDA INTO MPC-HC EXTERNAL FILTER LIST.
• STEP 8 – CONFIGURING MPC-HC INTERNAL SUBTITLE ENGINE.
• STEP 9 – CONFIGURING EVR CUSTOM PRESENTER/VMR 9 RENDERLESS VIDEO RENDERERS.
• STEP 10 – CONFIGURING COREAVC 2.0 CUDA.

Step 1:-
Download MPC-HC and extract it into a folder of your choosing. If you have an existing codec pack such as CCCP or K-Lite Codec Pack, go to the folder where their version of MPC-HC was installed, delete the mplayerc.exe if exists and then extract the contents of the MPC-HC archives you have just downloaded into it. MediaInfo.dll is now included within the MPC-HC package so no need for a separate download nowadays.

Step 2:-
Install  CoreAVC 2 without Haali Media Splitter as shown below.

Download the latest version of Haali Media Splitter from the link provided above. During the setup process, when you arrived at the stage depicted below, unselect the option ‘Associate .mkv and .mka files with Windows Media Player’ before proceeding with the setup process.

If you somewhat skipped the process above, go to Start Menu —> All Programs —> Haali Media Splitter —> Media Splitter Settings (Windows 7 default start path, YMMV), and set the option ‘Use custom media type for H.264′ to ‘No’, as shown below.

Step 3:-
Start MPC-HC that you have just installed in Step 1 above, then press the 'O' key to open up the Options window as show below. This is also accessible from View —> Options. Here, you should enable the 'Keep history of recently opened files' option.

Step 4:-
Go to the Playback section as shown below and enable the 'Auto-load Subtitles' so that videos with soft-subtitles like many anime fansubs out there can be watched with subtitles intact.

Step 5:-
Go to the Playback —> Output section as shown below to choose an appropriate video renderer to suit your need. Unlike the DXVA method, you have flexible options with CUDA method. It doesn't mean that you can just randomly pick a renderer just like that, because each has their own features and drawback.

System Default – The same as selecting WMR7 windowed.
Overlay mixer – The fastest video renderer available. If used with Windows Vista or 7, Aero desktop composition will be disabled. Cannot display soft-subtitles with MPC-HC internal subtitle engine therefore you need to use DirectVobSub or ffdshow subtitle engine to display them.
Old renderer – Don't use this one. Simple as that.
VMW7 windowed – Far slower than overlay mixer, but Aero will not be disabled during playback. Still need DirectVobSub or ffdshow subtitle engine for soft-subtitles. Don't use this one unless needed.
VMR9 windowed –  Far slower than overlay mixer, but Aero will not be disabled during playback. Still need DirectVobSub or ffdshow subtitle engine for soft-subtitles. Don't use this one unless needed.
VMR7 renderless – Can use MPC-HC internal subtitle engine for subtitles. Still because it is VMR, do not use it unless needed.
VMR9 renderless – Same as VMR7 renderless.
Haali Video Renderer – Even slower than VMR renderless, but produces high quality PQ (for nVidia cards). Have fine-grained control for bicubic-resizing. Can use MPC-HC internal subtitle engine for displaying subtitles. Cannot use hardware deinterlacing. Cannot use pixel shader post-processing. Still recommended though if you don't care about those options.
Null renderers – If you only want to hear audio, select this option. Else, select something else.
EVR – Very fast video renderer, not too far behind overlay mixer, with better PQ too. Aero will not be disabled.  No support for MPC-HC internal subtitles engine, therefore you need to use DirectVobSub or ffdshow subtitle engine. Can use hardware deinterlacing but no pixel shader post-processing. Recommended.
EVR custom presenter – Unlike the vanilla version of EVR above, the speed of this renderer has slowed down very considerable since the Beliyall merge way back. Has many features though, like brand new controls for v-sync correction, anti-tearing feature, BTB/WTW level control etc. MPC-HC internal subtitle engine is supported. Hardware deinterlacing is supported. Pixel-shader post-processing is also supported. Recommended.
madVR – High speed video renderer (with a mainstream or high-end GPU) with excellent PQ. Need to use DirectVobSub or ffdshow subtitle engine for subtitles. No hardware deinterlacing and no pixel shader post-processing.

Man, I really need to update that woefully outdated article some time in the future.

Step 6:-
Go to the Internal Filters section, and under Source Filters, untick the  'Matroska' and 'MP4/MOV' options. Under Transform Filters, untick the 'H264/AVC (DXVA)' and 'H264/AVC (ffmpeg)' options.

Step 7:-
Go to the External Filters section and you should be seeing the same thing as the screenshot below.

Click 'Add Filters' and the 'Select Filters' pop-up window will show up like below (your entries may differ than the ones in the screenshot).

Select 'CoreAVC Video Decoder' and then click OK, and you will see the External Filters window being populated just like below.

Set 'CoreAVC Video Decoder' to 'Prefer' so that MPC-HC will always use CoreAVC when playing H.264 videos.

Step 8:-
Go to the Subtitles section, and untick the 'Disable animation (for slow computers)' option. Enable it if you have slow CPUs, like Atom and Celeron. For 'subpicture to buffer option', it depends on your GPU RAM. 10 should be good enough for all situations. For 'Maximum texture resolution' option, set it to 'Desktop', and if you have CPU utilization problems, it should be lowered.

Go to Subtitles —> Default Style section, and ensure that the option 'Position subtitles relative to the video frame' is fully ticked.

Lastly, click 'Apply' and then the 'OK' button.

Step 9:-

THIS STEP ONLY APPLY TO YOU IF YOU CHOOSE EVR CUSTOM PRESENTER OR VMR9 RENDERLESS.

Right-click within MPC-HC video area, go to Renderer Settings —> GPU Control and make sure that all three options within it (Flush GPU before VSync, Flush GPU after Present and Wait for flushes) are ticked, just like the screenshot below.

If you use Windows XP or use Windows Vista/7 with Aero disabled, right-click within MPC-HC video area, go to Renderer Settings —> VSync and tick any of the options there to prevent tearing. Alternative VSync is recommended for ATI GPU users, the Accurate VSync is for everyone (a stricter version of Vsync) while VSync is similar to the ‘Lock Back-buffer option in the previous stable MPC-HC build (build 908).

I use Windows 7 with Aero enabled, thus those three Vsync settings above doesn’t apply to me.

If you use Windows Vista/7 with Aero enabled, go to Renderer Settings —>Presentation and ensure that the option Disable Desktop Composition is not ticked. If it somehow was ticked, untick it. If you are masochist enough to enable it, you have to select one of the VSync methods above or else you will get tearing. This option is useless in Windows XP (and you have to select a VSync method above).

Enable Frame Time Correction option should always be checked.

10-bit RGB option is optional. Disable if you want to take screenshots of the video being played. D3D Fullscreen Mode been made obsolete by the three VSync options above (and by Aero in Windows Vista/7), but if you use Windows XP and the VSync options and GPU Flushes cannot stop tearing when playing videos, this method is still the most foolproof way to stop tearing. If you have to resort to this, the option D3D Fullscreen GUI Support allows you to use context menu to operate MPC-HC without having to exit the application.

Again, right click within MPC-HC video area, go to Renderer Settings —> Output Range and select between 0 – 255 or 16 – 235, depending on what B&W level your display device is calibrated to. Read your display device manual for more information. Usually, computer monitors (LCD or CRT) are calibrated to 0-255 level, and TV is usually set to 16-235. But many high definition TVs nowadays are calibrated to 0-255, and some of them can have both, depending on what HDMI socket being used! Read the manual for more information.

Step 10:-
Go to Windows' Start Menu —> All Programs —> CoreCodec —> CoreAVC Professional Edition —> Configure CoreAVC and the window below should appear.

Configure it like above:-

Input formats: Just tick everything.
Output formats:  Only tick YV12, YUY2 and NV12 and arrange them like in the order shown above.
Input Levels:  Always use TV (16-235) because almost all materials out there use this level.
Output Level: Always use TV (16-235) because MPC-HC will be the one that will select whether TV or PC level will be sent to the display device (see Step 9 above).
Input Colorspace: Auto-detect should be option that should be chosen. Then hope CoreAVC will pick the right colorspace during playback. If you want to do it manually, take into consideration that most HD sources like Blu-ray and HD TV programs use BT.709 while SD sources like DVD use BT.601. 
Deblocking: Standard and nothing else. 
Deinterlacing: Set it to 'Hardware' if you want to use your GPU to deinterlace interlaced video. The caveat is that you need CoreAVC to connect directly to the video renderer (EVR and VMR9 only) for this to work. If you need to use the likes of ffdshow Video Decoder post-processing abilities, set this option to 'Bob' and tell ffdshow to pass NV12 colorspace to the renderer. Use this test file to do deinterlacing capabilities.
Aggresive deinterlacing: If you use hardware acceleration, this setting did not matter.
Crop 1088 to 1080: This setting will crop any videos 1088px in width to 1080px. Should always be turned on.
Force VMR AR detection: You don't use VMR right? Therefore let this one disabled. 
Preferred Decoder: Tick it if you want to use WMP.
Use Tray Icon: Must be turned on, because the color of the tray icon is the only indicator whether CUDA acceleration is used or not. Green mean that CUDA is used, Blue means that software decoding is used. 
Prefer CUDA acceleration: Obviously should be turned on, or else what are you doing here?

COREAVC CUDA IN ACTION.

With the latest version of CoreAVC 2, MPC-HC will not report that it uses hardware acceleration anymore. You will have to depend on the color of CoreAVC tray icon to see whether GPU acceleration is used or not. See image below for clarification.

Download the test videos above.

Haruhi Suzumiya ED theme (Hare Hare Yukai) – SD video.

Elfen Lied ED theme (Be Your Girl) – 720p video.

Kanon OP theme (Last Regrets) – 1080p video

Air OP theme (Tori no Uta) – multi-angle 1080p out-of-spec video

Planet Earth: From Pole to Pole clip – 1080p out-of-spec video

Planet Earth: From Pole to Pole clip – 1080p out-of-spec video (paying in Windows XP – therefore CoreAVC 2 is the only reliable way to play H.264 in Windows XP)