madVR - high quality video renderer (GPU assisted)

[leeperry]

I prefer DirectVobSub. I will demonstrate in pictures why.
yeah, ffdshow's subs are slightly blurred in YV12. Doesn't occur in RGB. Known problem.

They still can look pretty good IMO: http://www.pixelz.fr/5/0/8/335607c1883ee45ab40b6d200d30ett.jpg (http://www.pixelz.fr/5/0/8/335607c1883ee45ab40b6d200d30e.png)
That means you can get D3D exclusive playback without losing your manual calibrations.
I remember that the ATi drivers like to reset the CLUT in games(there are some workarounds, though), but my ARGYLLCMS CLUT is very much active in D3D exclusive mode mVR on an nvidia using XPSP3.

[bjd]

DirectX11 seems smoother in general, I don't know it might just be me.

OK, thanks for that. I am going to do some more tweaking/investigating on XP.

I have a 2gb DDR3 Geoforce 430 on order for the XP machine to improve things, but may end up resorting to a Windows 7 install as well.

The outlay cost for a new card and OS easily justifies what Madshi has developed

[namaiki]

yeah, ffdshow's subs are slightly blurred in YV12. Doesn't occur in RGB. Known problem.

They still can look pretty good IMO

Surprisingly lax response from you, considering your other ventures. :confused:

Anyway, I prefer to use DirectVobSub. Using FFDShow's subtitle renderer doesn't have any benefits for me.

[leeperry]

Surprisingly lax response from you, considering your other ventures.
I'm not sure what you're referring to, but subs sharpness is a complicated matter on CRT/DLP. Finding an acceptable in-between blurry and pixelated is very hard...this said, I do prefer your DirectVobSub screenshot over ffdshow's, but I like to change the ASS/SSA styles(DirectVobSub seems very limited for this?) and having them in ffdshow is pretty handy. I hardly ever need subs anyway.

[yesgrey]

Nvidia7600-256mb ...
Nvidia8400GS-512mb
You're making the comparison assuming that the first digit is the more significant one, and it isn't. The first is higher on 8400GS, but the second is lower, and that's what's making the difference. Compare the specs of both:
GeForce 7600 GT
Memory 256MB
Memory Interface 128-bit
Memory Bandwidth (GB/sec) 22.4
Fill Rate (Billion pixels/sec) 6.7
Vertices/Second (Millions) 700
Pixels per clock (peak) 12

GeForce 8400 GS
Memory Amount 512MB
Memory Interface 64-bit
Memory Bandwidth (GB/sec) 6.4
Texture Fill Rate (billion/sec) 3.6

The 7600GT has almost 4x the memory bandwidth of 8400GS, and that's the cause for the frame drops, not the DX11 vs DX9. There's more when choosing a graphics card than simply look for the model number... ;)

[Thunderbolt8]

so in brief, while we are already at it, which are the most decisive factors regarding speed for madvr to look out for when choosing a graphic card?

amount of memory?
type of memory?
memory interface?
memory bandwith?
texture fill rate?
...


anyone able to make a list from most important to least important?

[dansrfe]

memory bandwidth
texture fill rate
memory interface
amount of memory

roughly in the order that it matters from most important to least important. (I would think)

[jmone]

Thanks Madshi, Scoz, Mangix, Yesgrey - I've ordered a 1GB version of the HD5670 with a quite fan that fits into my Shuttle and should play nice with madVR (http://www.guru3d.com/article/his-radeon-hd-5670-iceq-review/1). Once installed I'll reconfig CCC for Full.
FYI - Installed and working well in the HTPC (quite, seems stable, and no probs with madVR).

HIS Radeon HD 5670 : according to gpu-z it has:
memory bandwidth: 64 GB/s
texture fill rate: 15.5 GT/s
memory interface: 128-bit
amount of memory: 1GB DDR5

[adam777]

I prefer DirectVobSub. I will demonstrate in pictures why.

FFDShow 3730
http://img155.imageshack.us/img155/50/ffdshow.th.png (http://img155.imageshack.us/img155/50/ffdshow.png)

VSFilter/DirectVobSub 2.40
http://img37.imageshack.us/img37/6136/vsfilter.th.png (http://img37.imageshack.us/img37/6136/vsfilter.png)

Thanks, namaiki and all that followed.
I was actually referring to the effect of having subtitles on the video itself, not the subtitles quality.
But, that being said, since this topic came up, should I decide to use DirectVobSub, where should I look for new versions and keep up with development?
Only source I could find is here (http://xhmikosr.1f0.de/index.php?folder=bXBjLWhjX2FwcHM=), provided by XhmikosR.
Is there any official site or something?
Thanks.

[jmone]

Thanks for your input to date, lets see if I have this correct:

System: Win7, HD5670 HDMI Cat 12.10 Set to RGB Full --> Yami Receiver --> Pio Plasma
File Playback: FFDSHOW --> madVR
Blu-ray Disc Playback: TMT 5.0

While I'm waiting for the tech to come and do his ISF Calibration stuff, I used the AVSHD ITU-R BT.709 Calibration Disc to adjust my Brigtness and Contrast in CCC
* first in TMT (from the ISO), then checked it in
* DirectShow / madVR (from the corresponding M2TS file)

I found that to keep the same Brightness and Contrast levels (when CCC is set to Full RGB) between TMT and madVR I need to also set madVR to 0-255 for the Pio Plasma. Is this correct...at no stage should I have anyting set to Limited RGB even though the end device is a TV?

Thanks
Nathan

[bjd]

The 7600GT has almost 4x the memory bandwidth of 8400GS, and that's the cause for the frame drops, not the DX11 vs DX9. There's more when choosing a graphics card than simply look for the model number

Thanks for this information which makes sense. At the time the 8400GS was bought as i was only looking for something cheap with HDCP and DXVA and assumed when comparing MadVR performance against an older 7 Series card that it would be more able which it clearly is not.

[yesgrey]

so in brief, while we are already at it, which are the most decisive factors regarding speed for madvr to look out for when choosing a graphic card?
What really counts is the memory bandwidth, but since it's calculated from the type of memory, the memory clock and the memory interface, we might need to look at these in case the memory bandwidth is not clearly specified.
The number of GPU cores is also important, as is their clock frequency, which will be used for calculating the shaders processing power. However, usually the manufacturers don't use more memory bandwidth than needed by the GPU, so if you have a good memory bandwith you should be fine with the shader power too.;)

Considering that there are already low cost cards with very good memory bandwidths, I would suggest to not go for less than 50GB/s, and with one of these you should be fine with GPU power too. For example, in my case, even though they were at the same price, I've decided for the GT 240 instead of the GT 430 because the latter due to only be available with DDR3 memory has half the memory bandwidth of the former. The GT 430 would be a good buy for madVR if/when they release a DDR5 memory version.
Note: I'm noy saying that the GT430 will not work with madVR, only that with a higher memory bandwidth you would be safer.

The size of the memory, for now, it's not critical, because the cheapest came with a minimum of 512MB, and that should be enough for current madVR releases.

At the time the 8400GS was bought as i was only looking for something cheap with HDCP and DXVA
Yes, the problem is that for the HTPC solutions still is believed that only DXVA support is enough, but madshi decided to change that with madVR. Unfortunately, some people bought their cards before knowing about it...

[Thunderbolt8]

What really counts is the memory bandwidth, but since it's calculated from the type of memory, the memory clock and the memory interface, we might need to look at these in case the memory bandwidth is not clearly specified.
is there a kind of formula for that in case the memory bandwith is not specified, like interface x clock x type etc. :D

if that list at wikipedia is correct, then I guess Im still fine with 51,2 GB/s of my radeon HD 4770

[rmp459]

bleh I think im done for now....

The audio accuracy in terms of it being synced with my video seems alot better with my current settings, but i can still tell that the video is a hair behind the audio.

I do love the quality of MadVR, but EVR Cus in MPC is dead nuts correct in the audio/video sync using the same reclock settings and all. I think im just going to have to wait a while... MadVR just seems to render slower and slower just by going from ~1280x720 windowed to full screen 1920x1080 @ 24hz. It thinks that it is in sync, but its not displaying fast enough to catch the audio being passed through the same hdmi cable. I have tried fsX and everything else ive read here and nothing helps. GPU isnt even running more than 10% usage.

On a side note, I reworked how I cabled things and set up my i/o assignments on my AVR and created a ' 'scene' on my Yamaha receiver for HDMI video & audio instead of using the coax from my sound card. As then I adjusted ffdshow and reclock to pass multich pcm instead of passthrough dts/ac3... so far im really like the difference, but for some reason I cannot find a way to resample it to 24bit... hmm i must be missing an option somewhere.

[bjd]

For example, in my case, even though they were at the same price, I've decided for the GT 240 instead of the GT 430 because the latter due to only be available with DDR3 memory has half the memory bandwidth of the former

Would have preferred GDDR5 memory, but limited to a single slot in my Shuttle and the single slot Radeon 5770 GDDR5's I have seen are DVI only not HDMI.

The Nvidia 7600 I have is actually a GS not a GT so memory bandwith is half of the GT, so I would expect a 430 with 25+ gb/s bandwith to be a big improvement

[leeperry]

I'm not sure you'll see such a big improvement from DDR3 to DDR5 w/ mVR...I dunno about AMD but what you should care about on the nvidia's is the number of SP(Cuda Cores), the more the merrier...even though anything >90 will do just fine w/ CoreAVC CUDA. My old 96SP G92 8800GS hardly tops >10% of GPU load(using spline resize and high bitrate 1080p CoreAVC CUDA). What matter more than DDR3/5 is the size of the memory interface, you want at least 192bit.

Until madshi finally implements those supposedly power hungry algorithms(I was hoping for frame interpolation but he refuted it), this is a waste of money IMHO.

Those top end graphic cards are meant for highly demanding PS computing(each pixel goes through +100 PS scripts in Crysis)...mVR is piece of cake in comparison, it doesn't quite require an expensive graphic card IME.

madshi also said that the AMD had a much higher unfiltered texture fill rate than the nvidia IIRC, but you don't get CoreAVC CUDA....supposedly CoreAVC 2.5 should be able to use DXVA/DXVA2 on the AMD but it's not available atm.

[pankov]

bjd,
there is no problem in using the DVI port of the Radeon cards. With the correct DVI-HDMI dongle you won't lose the audio.

[yesgrey]

is there a kind of formula for that in case the memory bandwith is not specified, like interface x clock x type etc. :D
Yes, but you have to be careful, because some sites include the type of memory on the clock frequency. So, for some cards which use DDR5 they state the clock as 3400MHz, when in reality it's 850MHz.

BW = clock (in MHz) x mem_type x interface (in bit) / 8
With DDR5 use 4 as mem_type, with DDR3 use 2 and with DDR/DDR2 use 1.
The Nvidia 7600 I have is actually a GS not a GT so memory bandwith is half of the GT, so I would expect a 430 with 25+ gb/s bandwith to be a big improvement
Yes, but considering that with the same price you can get a GT 240 that has the same GPU power with double the memory bandwidth... and in addition to that the memory can be easily overclocked at 1000MHz giving you 64GB/s.;)
I only see one reason to go with the GT 430: the fermi architecture, but due to the higher memory bandwidth I've preferred the GT 240.

[bjd]

there is no problem in using the DVI port of the Radeon cards. With the correct DVI-HDMI dongle you won't lose the audio.

I currently run DVI->HMDI. My concern was one review said there was no on board audio, but have re-researched, that looks incorrect, so it looks like a good choice for MadVR.

As a matter of interest the current Nvidia 8400GS has a DVI socket but it's EDID says it's HDMI, so if you don't connect a SPDIF signal to the card, you run into problems. When I first got the card, my LCD TV constantly clicked when you plugged in the HDMI cable as it was wanting to lock onto an audio signal and was not getting one until i connected the spdif.

[bjd]

I only see one reason to go with the GT 430: the fermi architecture, but due to the higher memory bandwidth I've preferred the GT 240.

Could not source a GT240/GDDR5 that will fit in a single slot. Anyway I appreciate all the help and insight on this matter - thanks again.

[cyberbeing]

If you want something powerful and single-slot, you could wait for the Galaxy GTX 460 Razor (http://www.galaxytech.com/en/productview.aspx?id=371) which should become available by the end of the month. It appears to be quite a bit longer than the standard GTX 460 though. Judging from pictures maybe 10.5 in (GTX 580 length) vs 8.25 in of the standard GTX 460. Haven't seen any real measurements though.

[bjd]

....nice :)

[Gser]

Everybody should just buy an 5870. Works like a charm.

[vlatko27]

Hi all, I just tested this renderer and in the short time I've seen it, it looks great. I just wanna make sure I've set everything up correctly.

HD 3870 card > ATI DVI/HDMI dongle> HDMI cable into Panasonic G10 plasma
Pixel format set to full RGB in CCC
Using Coreavc for decoding(input and output levels set to 16-235 and BT.709 colorspace selected as I only watch HD)
madVR selected as output in MPC-HC
madVR settings > TV levels selected

Is that OK?

[Plutotype]

Stay away from Nvidia until their win7 driver bug is fixed:
http://forums.nvidia.com/index.php?showtopic=153212&st=80

Get ATI 5670 Ultimate Sapphire or wait for the same in 6th series (passive version in summer 2011). And watch this thread:
http://www.avsforum.com/avs-vb/showthread.php?p=19553212#post19553212

Pluto

[6233638]

Stay away from Nvidia until their win7 driver bug is fixed:
http://forums.nvidia.com/index.php?showtopic=153212&st=80Output should be 24.000Hz and using ReClock on a HTPC anyway, and you can create custom timings if you want accurate 23.976. (even if it's wrong)

[pankov]

6233638,
what do you mean by "Output should be 24.000Hz"?
what if the content is 23.976 as is with most BD movies?

[Plutotype]

I prefer DirectVobSub. I will demonstrate in pictures why.

FFDShow 3730
http://img155.imageshack.us/img155/50/ffdshow.th.png (http://img155.imageshack.us/img155/50/ffdshow.png)

VSFilter/DirectVobSub 2.40
http://img37.imageshack.us/img37/6136/vsfilter.th.png (http://img37.imageshack.us/img37/6136/vsfilter.png)

Hi Naimaki,
I dont think ffdshow subtitles are worser than DirectVOBsub in Full HD. To test it right, put just simple Arial with black/grey outline 1 ( to have better visibility on whiter picture background ) with no other effects on both subtitle filters and compare. Again, compare in Full HD. I couldnt find a noticable difference.

I prefer ffdshow.:)

[6233638]

6233638,
what do you mean by "Output should be 24.000Hz"?
what if the content is 23.976 as is with most BD movies?It's my understanding that films are all shot at 24.000 fps and then slowed to 23.976Hz for consumer formats to maintain compatibility with 59.94Hz playback using 3:2.

Many films on Bluray are actually stored in the native 24.0fps, though I think its mostly releases in PAL regions where they do not need to maintain compatibility with 59.94Hz. (PAL uses 50Hz for legacy playback not 59.94Hz)

To watch/hear things as recorded you should be playing back at 24.000 and with reclock's upsampling (up to 32/192 if supported) there should be no loss in quality.

[pankov]

In general I agree with you about the original film frame rate and I hate the person who decided to use 23.976 (24/1001) but I'm not sure if there is actually a way to play a file at 24.000 if the source filter and video frame timestamps say 23.976. I think madVR and any other renderer will repeat a frame each X seconds (I can't calculate it right now). Reclock can't help here because all it can do is match the audio clock to the video one, but as I see it - it's the video one that's not correct to start with.
Do my words make any sense or am I totally out of line here?

[6233638]

In general I agree with you about the original film frame rate and I hate the person who decided to use 23.976 (24/1001) but I'm not sure if there is actually a way to play a file at 24.000 if the source filter and video frame timestamps say 23.976. I think madVR and any other renderer will repeat a frame each X seconds (I can't calculate it right now). Reclock can't help here because all it can do is match the audio clock to the video one, but as I see it - it's the video one that's not correct to start with.
Do my words make any sense or am I totally out of line here?ReClock was made to do this. Well, it was actually made to re-clock 25fps PAL films to 24fps, but is just as capable at re-clocking 23.976 to 24.000. (or whatever is required by your display)

It's trivial to change the playback framerate of video, you just display the images at a different speed. It's the audio that's tricky but ReClock takes care of it and resamples the audio to match the video speed. (and can upsample at the same time, to avoid a loss in quality)


It looks like madVR's dropped frame estimation is based on the refresh rate being output and the video file's framerate. ReClock ensures that there are no dropped frames at 24.000, though it is not reflected in the madVR estimation.

[pankov]

Thanks
Now I remembered that I reached to the same conclusion a while ago but obviously forgot it, because I don't use Reclock because I'm limited to SPDIF and I prefer to bitstream the DD/DTS track instead of decoding and then reencoding it to DD so it can pass through my optic cable.
I plan to upgrade my receiver soon so I'll be able to use reclock and send multichannel decoded audio over HDMI.
But what about DTS-HD MA and Dolby TrueHD? Are there any DirectShow decoders that can output the full audio resolution without downconverting it to 48kHz/16bit?

[madshi]

To watch/hear things as recorded you should be playing back at 24.000 and with reclock's upsampling (up to 32/192 if supported) there should be no loss in quality.
You're making things too simple here. There are 2 possible ways for a studio to slow an audio track down from 24.000 to 23.976:

(1) "simple" resampling
(2) pitch corrected time stretching

Option (1) will slightly lower the audio pitch, option (2) will maintain the audio pitch. Using Reclock's resampling will raise the audio pitch, which means that using Reclock for converting the 23.976 audio tracks to 24.000 restores the proper pitch for 23.976 audio tracks that were created by using method (1). But for audio tracks created by method (2) using Reclock will result in a too high pitch.

As far as I know, both options listed above are actually in use by studios, at least for PAL DVD audio track fps modification.

[jmone]

Pitch correction is certainly selectivly employed by Studios for some 24fps -> PAL conversions and while reclock can account for it it does bring it's own issues with audio artifacts. There is no automatic way I can work out of how to identify what has been pitch corrected and what has not... plus living in the PAL world we have heaps of true PAL content on top of that. So I just leave the pitch "as is" and have Reclock to do the frame rate detection and change the refresh rate of the TV to suit, then do it's audio resampling to keep from dropping frames.

While the 24fps --> PAL can lead to a 4% pitch change (pitch, length etc), in the Blu world the 23.976 to 24fps difference is only 0.1%. So while I have no reference to point to, I'd be surprised if the studios would bother with pitch correction OR that we humans would even notice such a small pitch change anyway. As a result I set CCC at 24hz for Blu with reclock keeping the 23.986 tagged material in sync and to me it looks are sounds great with madVR and reclock respectivly!

[6233638]

Thanks
Now I remembered that I reached to the same conclusion a while ago but obviously forgot it, because I don't use Reclock because I'm limited to SPDIF and I prefer to bitstream the DD/DTS track instead of decoding and then reencoding it to DD so it can pass through my optic cable.
I plan to upgrade my receiver soon so I'll be able to use reclock and send multichannel decoded audio over HDMI.
But what about DTS-HD MA and Dolby TrueHD? Are there any DirectShow decoders that can output the full audio resolution without downconverting it to 48kHz/16bit?I use the Arcsoft decoder which sends 24/48 to reclock.

I suppose it might make sense not to use reclock if you are outputting a lossy format from the PC.

You're making things too simple here. There are 2 possible ways for a studio to slow an audio track down from 24.000 to 23.976:

(1) "simple" resampling
(2) pitch corrected time stretching

Option (1) will slightly lower the audio pitch, option (2) will maintain the audio pitch. Using Reclock's resampling will raise the audio pitch, which means that using Reclock for converting the 23.976 audio tracks to 24.000 restores the proper pitch for 23.976 audio tracks that were created by using method (1). But for audio tracks created by method (2) using Reclock will result in a too high pitch.

As far as I know, both options listed above are actually in use by studios, at least for PAL DVD audio track fps modification.Reclock can do both these things, though I hear it's lossy when timestretching.

I have yet to find any PAL content that has required the use of timestretching though. Perhaps I've just been lucky in my choice of content. The only time it's probably necessary is if the content is natively 25fps rather than a 24 to 25fps conversion.

For what it's worth, it's very obvious when timestretching is necessary with PAL stuff. (I have played about with the settings to hear what things sound like when played at a lower pitch than they should)


I have been told that pitch correction is not used at all when going from 24.0 to 23.976. The difference is so small that the studios don't feel it necessary to perform a lossy operation like that on the audio.


Hypothetically speaking, even if some titles encoded in 23.976 are pitch corrected, assuming that you settle on either 23.976 or 24.000 rather than switching resolution on a per-film basis, surely 24.0 is still the better compromise of the two as we cannot know for certain either way.


1. It means that the video portion is played back at the correct speed, this much is certain, as your only options there are either slowing the video down, or using interpolation/frame blending/frame dropping, and those are not acceptable.
2. If the title was not pitch corrected when slowed down, the audio is now at the correct pitch.
3. If the title was pitch corrected, then the difference must be so small as to be negligible. (most people don't even notice films being played back at 25fps which is a much larger difference)
4. Films which are 24.0 on the disc are played back perfectly. This is the most accurate representation of the original film that you can have at the home.


If you settle on 23.976 then a lot of (I suspect most) films are slightly off-pitch, the video is always at the wrong speed, and films which are released in the original 24.0 format are now degraded.

[namaiki]

Hi Naimaki,
I dont think ffdshow subtitles are worser than DirectVOBsub in Full HD. To test it right, put just simple Arial with black/grey outline 1 ( to have better visibility on whiter picture background ) with no other effects on both subtitle filters and compare. Again, compare in Full HD. I couldnt find a noticable difference.

I prefer ffdshow.:)

If you want to selectively mask how bad something looks, then that is your choice.

[dansrfe]

madVR screen switching still does not work.

[G_M_C]

Output should be 24.000Hz and using ReClock on a HTPC anyway, and you can create custom timings if you want accurate 23.976. (even if it's wrong)

That is not always true. I'd like my graphics card to output whatever stream, on the framerate the stream has. Simply because i have a a video-processor in my receiver that does the 23.976 -> 24.000. And i think it does a better job at it.

I concur with plutotype that its a bug that should be fixed. But there are more company's that make errors interpreting what "24p" actually means ;)

[jmone]

The whole 24 / 23.976 stuff is full of myth. Film is 24fps, but both the video and audio are slowed down slightly to 23.976 on Blu to be backwardly compatible with NTSC CE Equipment. Please keep in mind that there is no conversion between 23.976 and 24 itis just a reference to the speed the content is played at. I'm fine to play both the Video and Audio at either as they are so close it does not matter (but 24hz is the "real" frame rate). The trick is what track (Video or Audio) will be used for the master clock - and this is what Reclock does so well...it lets the video play at whatever speed you have set in your video card (24 or 23.976) and then will resample the audio to fit this so you don't drop video frames. Also keep in mind no video clock is perfectly accurate anyway, mine is actually something like 24.005but it does not matter with this aproach as reclock will just resample the audio to keep pace with they video anyway (of course this does not work as well for bitstreamers as you need to repeat or remove audio packets to do the same thing).

[G_M_C]

The whole 24 / 23.976 stuff is full of myth. Film is 24fps, but both the video and audio are slowed down slightly to 23.976 on Blu to be backwardly compatible with NTSC CE Equipment. Please keep in mind that there is no conversion between 23.976 and 24 itis just a reference to the speed the content is played at. I'm fine to play both the Video and Audio at either as they are so close it does not matter (but 24hz is the "real" frame rate). The trick is what track (Video or Audio) will be used for the master clock - and this is what Reclock does so well...it lets the video play at whatever speed you have set in your video card (24 or 23.976) and then will resample the audio to fit this so you don't drop video frames. Also keep in mind no video clock is perfectly accurate anyway, mine is actually something like 24.005but it does not matter with this aproach as reclock will just resample the audio to keep pace with they video anyway (of course this does not work as well for bitstreamers as you need to repeat or remove audio packets to do the same thing).

Reclock may be fine, but my receiver does the trick itself, and syncs the audio to go with it, including DTS-MA and such. So from my viewpoint its the easiest en most reliable way, and no need for reclock etc. :)
And as such, its my opinion that a video card must output what it gets delivered, without posing requirements to framerate or not (like the bug plutotype mentioned that the Nv cards that seem to require 24.000 fps).

I'd also like to make clear that i dont even have an Nv card. I just posted to suggest that there are people with other configurations than usual, or you have considered.
To suggest another point of view; Some say you almost have to use reclock, others say you dont want to have a Nv card, all because of framerate conversions or judder etc. Thats fine and all, but for people like me that have an dedicated (external) video processor, discussions like this are less important; its PQ whats important. Thats why i follow the MadVR tread, and the new 32bit floating point shader stuff that is developed by some of the MPC-HT team.

[Mark_A_W]

It's my understanding that films are all shot at 24.000 fps and then slowed to 23.976Hz for consumer formats to maintain compatibility with 59.94Hz playback using 3:2.

Many films on Bluray are actually stored in the native 24.0fps, though I think its mostly releases in PAL regions where they do not need to maintain compatibility with 59.94Hz. (PAL uses 50Hz for legacy playback not 59.94Hz)

To watch/hear things as recorded you should be playing back at 24.000 and with reclock's upsampling (up to 32/192 if supported) there should be no loss in quality.

I'm in PAL land and I have heaps of blurays and NOT ONE is 24.000fps.

There are a couple of 25fps/1080i50hz, but nearly all are 23.976fps.


The correct number to use is 23.976, for better or worse, that's the way they are encoded. With a multiple of 23.976 madVR says 1 frame drop every 6.07 days (or a strange garbage number 1.#J days - what's up with that madshi?).

IMO if you use 24fps refresh you are forcing the wrong number.

[Mark_A_W]

madshi, I've been watching BSG BDs from the discs (the m2ts files direct), using the latest MPC-HC splitter as the Haali Splitter doesn't "see" the DTS-MA stream. The results were basically perfect, no jumps/judders/repeats. Perfectly smooth.


But the last episode is seamless branching, so I had to convert it to MKV, which I then watched using the latest Haali Splitter. Not so good. A couple of jumps - and the feeling of rock solid smoothness was gone.


I've done some tests, and on my system at least, the MPC-HC splitter leads to a better result than the Haali Splitter, either M2TS or MKV, it doesn't matter. With the MPC-HC splitter a "0" drops, "0" repeats result is not unusual (with a reset straight after the start of playback, to clean up the start drops - if would be nice if you did that automatically).

I'm using madVr 0.36 and reclock, active but reclock v-sync disabled, in Zoom Player.

[6233638]

That is not always true. I'd like my graphics card to output whatever stream, on the framerate the stream has. Simply because i have a a video-processor in my receiver that does the 23.976 -> 24.000. And i think it does a better job at it.

I concur with plutotype that its a bug that should be fixed. But there are more company's that make errors interpreting what "24p" actually means ;)Certainly they should sort out their 23.976 output for users that want it (though you can just create a custom resolution in the control panel with the correct timings - I have posted them quite early on in this topic) but many people are under the wrong impression that 23.976 is viewing the film as shot.

I very much doubt that your video processor can do nearly as good a job with the framerate conversion as reclock does. Certainly there are no consumer processors capable of it. (Lumagen, DVDO etc.) The advantage ReClock has is that it has access to the data and can simply tell the computer the speed the video should be played back at. A video processor would need a massive buffer to perform the same task by properly re-clocking the audio & video without simply frame blending/dropping/interpolating.

The whole 24 / 23.976 stuff is full of myth. Film is 24fps, but both the video and audio are slowed down slightly to 23.976 on Blu to be backwardly compatible with NTSC CE Equipment. Please keep in mind that there is no conversion between 23.976 and 24 itis just a reference to the speed the content is played at. I'm fine to play both the Video and Audio at either as they are so close it does not matter (but 24hz is the "real" frame rate). The trick is what track (Video or Audio) will be used for the master clock - and this is what Reclock does so well...it lets the video play at whatever speed you have set in your video card (24 or 23.976) and then will resample the audio to fit this so you don't drop video frames. Also keep in mind no video clock is perfectly accurate anyway, mine is actually something like 24.005but it does not matter with this aproach as reclock will just resample the audio to keep pace with they video anyway (of course this does not work as well for bitstreamers as you need to repeat or remove audio packets to do the same thing).Thanks for confirming what I already thought, and had previously been told - there is no pitch correction done when slowing from 24.000 to 23.976, and so outputting 24.000 (or as close to it as possible from a PC, as you say) is the correct thing to do.

I have also noticed that the clock at 24.000 is almost completely stable from a PC whereas 23.976, even when using the correct timings tends to fluctuate more.

And yes, this is why I think it is funny that some people are so focused on bitstreaming just to have TrueHD/DTS HDMA light up on their amp. It is actually better to have reclock resample (upsample preferably) than it is to bitstream, at least from a PC.

I'm in PAL land and I have heaps of blurays and NOT ONE is 24.000fps.

There are a couple of 25fps/1080i50hz, but nearly all are 23.976fps.


The correct number to use is 23.976, for better or worse, that's the way they are encoded. With a multiple of 23.976 madVR says 1 frame drop every 6.07 days (or a strange garbage number 1.#J days - what's up with that madshi?).

IMO if you use 24fps refresh you are forcing the wrong number.I suppose it just depends on the films you watch. I am also in "PAL Land" and have notice that quite a lot of films I have bought on Bluray have been 24.0 on the disc. (though I couldn't name any off the top of my head) And oh how I hate those 1080i50 releases. Even though they can be deinterlaced and reclocked to 24p, it's such a stupid thing to have to do.

23.976 is not correct though. It is slower than the original film was shot and is only done for legacy compatibility with NTSC playback. To play back the video portion at the original 24.0 is relatively simply on a PC as you can just change the time that each frame is displayed (unlike an external video processor which has to mangle the video to do the same thing) the audio then has to be resampled to keep it in perfect sync with the video when you do this, but reclock does a very good job of it, and even if you choose to output "23.976" from your PC, chances are that it is not a stable/accurate 24/1.001 and audio will need resampling to perfectly match the video clock anyway. (something only reclock does, anything else will simply drop/repeat packets)

[jmone]

The 23.976 or 24 frame rate is just a number in the header of the file, it has nothing to do with the number of video frames in a file or how it was encoded. Each frame is a single complete frame and the frame rate is just the instruction on what speed to play them back at.... eg in this example the frame rate tells the renderer to display each frame for either 1/23.976 OR 1/24th of a second. For example, I could take these same stream and play it at 50fps but the result would be it would all be in fast motion, sound would be a higher pitch etc. The issue is not so much if you choose a refresh rate of 1/23.976 or 1/24 frames per second (and remember it was all shot in 24fps and they did not "remove" 24 frames per 24,000 to get to 23.976, they just slowed it down a tiny bit) but how to keep the audio and video in sync as on the PC they both use different clocks, so it is inevitable that they will drift apart over time. The default behaviour on a PC is to drop/add Video Frames at some point to keep the Video in sync with the Audio as the two drift apart (the default behavior on a PC) OR use reclock to resample the Audio to keep it in sync with the Video playback. The issue would go away if there was a unified clock that both the Audio and Video renderer would use....but that is not how it works.

[6233638]

The issue would go away if there was a unified clock that both the Audio and Video renderer would use....but that is not how it works.And it's a shame that it isn't how it works. I would have thought that, at least when outputting over HDMI (no sound card involved) that would now be possible.

[jmone]

I've attached a screen shot of a simple recording. Unlike the "old" days off film where the audio track was physically glued on the film so the two were always in sync, you can see they they are split into there two seperate streams and processed indepentantly. What the Video Renderer does is keep checking with the Audio Renderer (see the little clock) to see if it is keeping in sync with the rate the audio is being played at and it will drop/repeat a frame if it drifts too far apart. Reclock just reverses this by resampling the audio to keep in time with whatever speed the video is playback at (eg approx 24 ish frames per second for a Blu).

I have '00 of Blu and HD-DVD's and I must say that all the all the films based Blu are tagged at 23.976, some are 50 or 60i where the source was a concert or shot for TV, and some HD-DVD's were tagged as 24. At the end of they day you are going to want to have the frequence of the display device match (ish) that of the source it was originally recoded in and due to the multiple clocks in the PC decide if you prefer to drop/repeat video frames or resample the audio to keep the two tracks in sync.

[Plutotype]

If you want to selectively mask how bad something looks, then that is your choice.

Thanks:), you are right. I accept the fact, that ffdshow is adding some default blur, what in my subtititles settings helps to look them better on big screen. Maybe using colored or effected subtitles will uncover the ffdshow subtitles renderer artifacts more.

[jmone]

And to add I've spent countless hours on my original G33 playing with front and back porch adjustments to try to get everthing in sync and found out the hard way....some HW just sucks and you can not do it (the best it would do is drop a frame each 17 sec at best). That said, if your HW setup likes 23.976 or 24 or anything around there then set it to what works best and be happy. I'd still recomend decoding to LPCM on the PC and letting reclock resample, the only downside being there are no freeware DTS-HD decoders (so you have to use a payware app like Arcsoft, or hack their filters, or convert these tracks to FLAC etc if you rip to MKV and want to keep the HD sound).

[Plutotype]

Output should be 24.000Hz and using ReClock on a HTPC anyway, and you can create custom timings if you want accurate 23.976. (even if it's wrong)

That means, it makes no sense when using Reclock to pre-set 23hz ( 23.976Hz ) in the resolution settings at ATI/Nvidia when 23.976fps movie is played?

If I set 23.976 fps in Reclock media adaptation settings, the icon turns out yellow. If I select auto and let the CINEMA Mode do the work, it is speeding up the media to match the LCD TV refresh rate 24hz ( media will be played at 24fps and audio goes to 48048hz ). I think the LCD TV producers did implement only 24hz refresh rate, not 23.976hz ( correct me if Im wrong ). Does an ordinary blu-ray player have also 24fps speed-up? If the TV is 24hz only, the movie can not be synced to 23.976 blu-ray output.

What if I dont want that Reclock will speed up the movie to 24fps if the video stream is 23.976 and have perfect sync with the LCD TV? Does anybody has a LCD TV with 23.976Hz AND 24Hz refresh rate options? I have SONY Bravia EX500 and although I set 23.976 in CCC, Reclock still speeds up the 23.976 movie to 24. SONY does not have 23.976hz refresh rate in timings according to moninfo tool. I assume also if it would have, the refresh rate would be never stable at 23.976Hz I think..



Thanks for the response.
Peter

[Razoola]

ReClock was made to do this. Well, it was actually made to re-clock 25fps PAL films to 24fps, but is just as capable at re-clocking 23.976 to 24.000. (or whatever is required by your display).

I have to disagree to an extent, what you are saying is of course correct but reclock is older than that too. Originally it was designed to simply get smooth playback via TV out of the gfx card etc. Back in those days it was hard to even get a pal DVD to play right through a TV or monitor at 50hz without some kind of jerky / tearing issue. It of course quickly expanded to include the things you mention which is great.

Reclock was originally designed to solve graphical issues and not sound, this came later along with the benifits of pal speeddown etc. Many people forget this.

These days however reclock is more based on sound being accurate over video and because of this there are problems in reclock with bitstreaming because the current dev (James) is not so keen to fix video issues (or give options to solve video situations at the expence of audio) which I personally feel is a shame for those of us who would choose video over audio as a priority. I personally wish OGG was still the dev.

The main issue I have with reclock is it will not go green when bitstreaming if you have the refresh rate at a multiple of 24 and content at 23.976. Reclock will stay yellow unless you speed up the content to 24hz. James feels audio packet drops caused by 23.976 into 24hz is too much, this is something I disagree with, it could be easily be fixed by allowing a greater deviation in max slowdown and speedup speed (which is already available if you use resampling). IE the greater packet drop would not effect all users or reclock, only those that choose to use it.