madVR - high quality video renderer (GPU assisted)

[DragonQ]

That post seemingly makes it more complicated than it is. I'm pretty sure I was able to do it just by using the nVidia Control Panel, where you can set which GPU a specific application uses.

[konakona]

What are exacly pc specs needed to playback 1080p hi10p videos?
ive tried running a few and after few seconds of no dropped frames it starts to lose 3/4 of them after-what is the reason?

my pc is hd4870 1gb,amd phenom II x2 550 (3.1ghz) ,4gb of ram (actually im on 32 bit system so its clipped to like 3,5gb),win xp

[noee]

What are exacly pc specs needed to playback 1080p hi10p videos?
ive tried running a few and after few seconds of no dropped frames it starts to lose 3/4 of them after-what is the reason?


Scaled to 1080 or source 1080? I have an old clunker running an HD2600XT with an overclocked X2 BE-2400 (2.9Ghz) and it actually works well with Lanczos3/AR scaling SD 24fps 10bit material (CPU works at around 50% IIRC). Haven't tried it with the new DXVA2 stuff (for 8-bit).

[nevcairiel]

1080p 10-bit material is rather CPU intensive, i'm not sure how a Phenom II would do there. Check your CPU usage to verify.

[konakona]

100%-i guess my cpu is too slow for full hd hi10p
LAV cant also decode 10bit material using gpu,but i guess then my gpu would be the bottleneck :<

[nevcairiel]

No GPU today supports 10-bit decoding, its a hardware limitation.

[pie1394]

What are exacly pc specs needed to playback 1080p hi10p videos?
ive tried running a few and after few seconds of no dropped frames it starts to lose 3/4 of them after-what is the reason?

my pc is hd4870 1gb,amd phenom II x2 550 (3.1ghz) ,4gb of ram (actually im on 32 bit system so its clipped to like 3,5gb),win xp

My MODT server's C2D P8600 2.4GHz + nVidia ION + Dual-ch DDR3-1333 is capable of handling such 24fps FILM type contents except to some few high-bitrate complex + quick motion scenes.

My HTPC machine's C2D E8400@3.6GHz + Intel P45 + Dual-ch DDR2-800 has no problem to deal with various FHD H.264 Hi10p video file decoding... maybe up to 60 fps source.

[cyberbeing]

When I had my AMD X2 4800+ (939) @2.64Ghz, it could do 1080p hi10p 24fps only as long as absolute peak bitrate stayed below 25-30Mbps.

As for new CPUs from the Intel side, something like a i3-3220 3.3Ghz dual-core I would consider the minimum recommended specs for 1080p hi10p 24fps, handling bitrate peaks up to roughly 50-60Mbps or so.

[chros]

Could you also test the overhead of only PS? (without upscaling or downscaling) Thanks!
I have checked with 720p hdtv content in windowed mode (without any resizing). And so the difference to be more visible I forced P8:400Mhz state of my GPU instead of P0:500MHz (previous test used this). I'm using P8 state for everyday use anyway.
- 0.85.1 , no PS: GPU: 26% , MCU: 29%
- 0.85.1 , with PS: GPU: 42% , MCU: 33%
- 0.85.3 , no PS: GPU: 26% , MCU: 27%
- 0.85.3 , with PS: GPU: 69% , MCU: 49%

So, the overhead is around 60% (GPU) , 67% (MCU), which is really much.

Not sure if I like the idea, I'll think about it.
Thank you for considering it. Otherwise I would have to stuck with the 0.85.1 until I change my laptop ...

[glc650]

this is an (black white) 0 - 255 PIC..on my CRT i see all color differnces..on my HD TC however i do not..on my hdtv i use 10 - 255 and its perfect now thanks to madshis custom color range option..

http://i155.photobucket.com/albums/s297/mikeyurkus/0-255BWDVD.png

16 - 255 gave me overgreyed blacks and artefacts in grey, black portions of teh screen..The only difference I really see on my TV is with Limited RGB selected in the CCC I see "DVD" in each of the 4 larger boxes in the middle (with the letters being a bit less visible in the right side boxes). With CCC set to Full, I don't see "DVD" in any of the boxes.

[toniash]

I have checked with 720p hdtv content in windowed mode (without any resizing). And so the difference to be more visible I forced P8:400Mhz state of my GPU instead of P0:500MHz (previous test used this). I'm using P8 state for everyday use anyway.
- 0.85.1 , no PS: GPU: 26% , MCU: 29%
- 0.85.1 , with PS: GPU: 42% , MCU: 33%
- 0.85.3 , no PS: GPU: 26% , MCU: 27%
- 0.85.3 , with PS: GPU: 69% , MCU: 49%

So, the overhead is around 60% (GPU) , 67% (MCU), which is really much.


Thank you for considering it. Otherwise I would have to stuck with the 0.85.1 until I change my laptop ...

@chros thank you!
@madshi please those with a weak GPU need some tweaking to lessen the overhead

[nevcairiel]

Not every feature is for everyone, sometimes the added complexity for such a tweak is not worth it, just so that extremely low GPUs can just barely use a specific feature. What you're asking sounds like quite a difference in processing.

You could test with the first promised tweak option once its in, maybe it already helps enough for your setups.

[BetA13]

The only difference I really see on my TV is with Limited RGB selected in the CCC I see "DVD" in each of the 4 larger boxes in the middle (with the letters being a bit less visible in the right side boxes). With CCC set to Full, I don't see "DVD" in any of the boxes.

well if u dont see some colors or the dvd thing, then ur TV or the color range u selected does get lost in blacks or whites..
I tested around till i found that 10 - 255 was perfect. i did see everything like i should. I have a CRT monitor also on wich i can see all colors for reference..So its easy for me to get it setup.

Play around and test the custom color range in madvr and see if u can find a setting in wich u see everything perfectly fine..then id say ur good to go cause no colors get lost in blacks or whites..

it really is a difference for me now, before i did that i only had perfect color output on my Flat CRT screen..Thanks to madshi i now have it on my HDTV too :)

Its better then playing with gamma..no artefacts etc..

[glc650]

well if u dont see some colors or the dvd thing, then ur TV or the color range u selected does get lost in blacks or whites..
I tested around till i found that 10 - 255 was perfect. i did see everything like i should. I have a CRT monitor also on wich i can see all colors for reference..So its easy for me to get it setup.

Play around and test the custom color range in madvr and see if u can find a setting in wich u see everything perfectly fine..then id say ur good to go cause no colors get lost in blacks or whites..

it really is a difference for me now, before i did that i only had perfect color output on my Flat CRT screen..Thanks to madshi i now have it on my HDTV too :)

Its better then playing with gamma..no artefacts etc..How do I test with that test pic in madVR though? I was just viewing it on my desktop like I would with any of my photos.

[BetA13]

well, i just open or drag the Picture into my MPC HC, and madvr etc displays it..
or alternative, u could use a movie and search for an dark scene, then test the settings and see when its going to bright (black aint black) or going too dark, then find teh perfect dark spot..so everything is displayed and nothing lost..

but normally u should be able to open the picture with the player..


_betA_

EDIT:

i have nvidia, and in the control panel settings i use this option:

Under Video color management i set:

- Use Video Player settings..(disabling the NVIDIA color settings internally i guess..so the player (in this case MadVR gets to choose what color space , enhancements you use..)

Maybe u have an option for that..so it only affects VIdeo and not desktop plus video et cetc..

--------------------------------

@ madshi..

i dont know if i found a bug or something..But sometimes after i restart my PC MADVR finds a new monitor. But i didnt change anything, what happens then is, i have 4 or 5 displays listed in madvr, but only have my 2 monitors connected like allways..

heres a screen: ( the AOC one allways stays teh same, so its not affected, but teh Philips one gets doublicated all teh time. (standart)(1) is now the philips, all above aint working anymore.. So i have no clue why that is.

http://www.abload.de/img/unbesnanntslp85.png


Its pretty annoiing at times..other day i start up the PC, wanted to watch a movie and teh picture was strange, till i saw, that madvr added a new HDTV, but its teh same..so i had to go into config again and set it up again..

Just thought i ask if you know about this and can help me. Or maybe i have something on my side wrong..who knows, hehe..

Greetz BetA

[nlnl]

@all having Nvidia card
I have installed new GTX 650ti in old motherboard (based on 9400 Nvidia chipset) and I have two issues:
1) Can not get exactly 23.976. Madvr's OSD tells me that display refresh rate is 23. 971.
And my old AMD 5750 card outputs exactly 23.976.
2) The card is connected to Onkyo 1007 receiver. Windows tells that 650ti does not support 88.2, 176,4 kHz audio using hdmi. And my old AMD 5750 card does output them.
Does anybody have the same problems with Nvidia cards? Or the problem is my old motherboard?

[nevcairiel]

1) Create a custom resolution for better timings
2) NVIDIA does not support output of those "odd" sample rates.

This is hardly a topic for the madVR thread though.

[nlnl]

1) Create a custom resolution for better timings
2) NVIDIA does not support output of those "odd" sample rates.

This is hardly a topic for the madVR thread though.
Thanks!
So AMD is my fortune (with it I can get 23.9760 and 176.4 kHz for SASD and CD upsampling out of box) ?

[nevcairiel]

Regarding 1), i rather have options to setup the timings myself then hope the out of the box timings work - because if they don't, i can't do shit =p

[Heuer]

I have a GTX 650Ti and I can get almost exactly 23.976 using custom resolution with vertical pixels set to 1128 and refresh rate set to 23.975.

[khanmein]

http://forum.doom9.org/showpost.php?p=1603351&postcount=15891

thanks. all i need to do is rename the mpc-hc to another different name.

@madshi i solved the BT.609 & BT.709. due to my BD-RMVB version.

[ThurstonX]

thanks. all i need to do is rename the mpc-hc to another different name.

@madshi i solved the BT.609 & BT.709. due to my BD-RMVB version.
Yep. Optimus was driving me nuts on my laptop with a GTX 560M and the 307.21 driver. Renaming mpc-hc.exe to mpc-hc2.exe allowed me to add it to the "Program Settings" in the nVidia control panel, and be able to choose the GPU rather than iGPU. Now I can use Jinc3 AR for everything. Well, still experimenting with downscaling, but don't need it too often.

HTH.

[madshi]

(1) Use software decoding and e.g. Bilinear scaling.
(2) Use native DXVA2 decoding and e.g. Bilinear scaling.
(3) Use software decoding and DXVA2 scaling.
(4) Use native DXVA2 decoding and DXVA2 scaling.

Redone the test with 0.85.3:
Same behaviour like 0.85.2.
(1), (3) and (4) are the same, (2) is different (oversaturated green). So using native decoding without dxva scaling is still broken (at least for my setup).
Madshi, levels are now incorrect using 85.3 with dxva scaling and mpc-hc internal decoder and splitter. Blacks are now grey. Other scaling methods provide proper levels. I'm using the old path exclusive without overlay on an Amd card.
crash report: radeon 7750 catalyst 12.10, mpc-hc 1.6.5.6261, lav filters 054.1 dxva-native, madvr 085.3, 2 monitors, 1x dvi / 1x hdmi

- 720p or smaller dxva: play video windowed, enter fse, return to window, move to other screen, crash (repeatable)
- 1080p dxva: play video on screen1 (windowed), move to screen2 (windowed), crash (repeatable)

always exception message 'access violation at address $4a4141a2 in module 'madvr.ax' read of address $0'
I found an annoying issue with madVR. On several .vob/.mkv files which are 30i, using LAV Video to decode it(software) results in deinterlacing being activated. The same is not true of madVR's built-in decoder. it says deintertlacing off (framerate).
These 4 problems should all be fixed in the next build.

If I open paint shop pro and test colors 000000 to 161616, it's perfectly black. I do start to see (extremely faint) grayish around 171717.
(video)
If I play videos at 0-255, I suffer majorly of burnt blacks.
Then your GPU is outputting untouched 0-255. Try using the brightness & contrast controls of your display to make photos look alright. Then set madVR to 0-255, too, and all your problems should go away. Both photos and madVR screenshots should then look perfectly fine.

But regardless of the issue i'm experiencing, I still disagree on the fact that the best solution to "unify" screenshot color levels is to use PC levels.
After all the sources aren't the same. It makes sense for vacation photos or anything coming out of the PC world to be in PC levels.
But again, in the video world as we both know, 99% is in 16-235 color range by definition so taking "screenshot" using this color range makes more sense to me
We're discussing three different things here:

(1) The way the movie is encoded. This is limited range YCbCr 4:2:0.
(2) The data format which is used to transport RGB to the display. This can be either Limited and Full Range RGB.
(3) The data format which screenshots are saved in. The PNG specification clearly supports only Full Range RGB. Limited Range RGB is *not* supported by PNG.

Now that we agree that all screenshots should have the same levels, I hope you can agree with me that we should create PNG files that follow the PNG spec. As a result the only valid choice is to store screenshots in full range.

The only reason why the screenshots look bad in your setup is that your setup is screwed up for desktop, photos and games etc. If I created PNG screenshots with video levels, the screenshots would look alright on your setup, but they would look washed out on every other setup. You can't seriously ask me to create screenshots that only look correct when photos look burnt out at the same time, can you? :scared:

... if only to allow comparision with "screenshot" from other renderers which seem to be outputting 16-235 by default.
On my PC the MS renderers default to creating PC levels screenshots. Also, if you look at screenshot comparisons on the internet, at least 95% of them are using PC levels.

I FINALLY figured out why I'm getting dropped frames in madvr. Its leafs. Scenes with lots of green leafs, trees, and that sorta thing make my GPU usage rocket up to 100% and madvr starts dropping frames like crazy. Is there anything that could be done to make madvr handle leafs better? If it matters, I'm using Jinc4, and my GPU Is a GTX460.
madVR's algorithms are currently not data dependent. Which means that it doesn't matter to madVR whether the video has leaves in it, or just a solid white screen. Maybe the GPU itself behaves differently, depending on e.g. how easy to compress the frames are, I don't know. In any case, I'd suggest that you try Jinc3 instead of Jinc4. The difference in image quality between Jinc3 and Jinc4 is very low.

My Plasma can do 60 hz and 48. It's a panasonic st30. When I send it 24p, I can switch to 48hz, but it has horrible flicker, so that's a no go.

So, I'm left with 60hz. My question is, I can have madvr switch the refresh rate to 24p, which means the tv does 3:2 pull down. The tv says the option is off, but since there's video playing on the screen, it must be doing 3:2. My other option is to not do refresh rate switching, leaving the video card at 60hz. Does that mean madvr is doing the 3:2 pulldown or is it my vid card? Which is preferable, letting the tv handle it, or sending it already switched?
At the moment I can't say for sure what is better. Just try it out with a video which has very smooth motion (e.g. a loooong slow camera pan). In the long run it will be better to let madVR do the 3:2, but that's future talk.

Have you tried whether your TV supports 72Hz?

I'm just curious if anyone else has experienced this bug while using cuvid and madvr on win7

http://support.microsoft.com/kb/983615
Not that I know of.

i dont know if i found a bug or something..But sometimes after i restart my PC MADVR finds a new monitor. But i didnt change anything, what happens then is, i have 4 or 5 displays listed in madvr, but only have my 2 monitors connected like allways..

heres a screen: ( the AOC one allways stays teh same, so its not affected, but teh Philips one gets doublicated all teh time. (standart)(1) is now the philips, all above aint working anymore.. So i have no clue why that is.

Its pretty annoiing at times..other day i start up the PC, wanted to watch a movie and teh picture was strange, till i saw, that madvr added a new HDTV, but its teh same..so i had to go into config again and set it up again..
madVR detects displays based on the EDID. It seems that for some weird reason the EDID block of your display changes sometimes after a reboot. I don't know if there's a random number that changes in the EDID block. If so, there could be a virtually unlimited number of new displays being in store for you... :( But more probably there will only be a specific number of different EDID blocks. The best way to handle this is to drag & drop the "identification" from the newly detected displays to your original display in the madVR device manager. This will tell madVR that all those EDID blocks identify the same display. After you've moved the identification, you can delete the "empty" display shells. Hopefully after you've done this with all newly detected displays, it might stop.

[madshi]

Same behaviour when paused. I tried disabling the madVR buffers [or at least change buffer to '1'] but that didn't change anything.

And it's the same behaviour whether I choose 'capture source' or 'capture real screen'

I don't know if it will tell you anything but I uploaded a log: http://www.mediafire.com/view/?t6bm44b4nj4a8xh

-Open video
-Seek ahead
-Take capture
-Exit
I'm having a hard time understanding why people don't zip text files before uploading them. Unzipped: 16MB. Zipped: 370KB. :confused:

Anyway, according to the log, PotPlayer does not use madVR's screenshot functionality at all. I suppose that PotPlayer grabs the video frames directly from the decoder. You'll have to ask the PotPlayer devs to use madVR's screenshot functionality instead. It's available through the official Microsoft interface (IBasicVideo::GetCurrentImage).

I have a problem with 85.x versions, when opening mpc-be\hc on extended display the first frame is displayed and then pauses. Tried resetting madvr and mpc to defaults with same results. 84.7 and EVR plays like it should with same madvr, mpc settings. Tested on 2 computers, nvidia and intel gpus with a few different videos.
I've tried to reproduce that on my AMD PC, but it works just fine here. However, I've used the very latest v0.85.x sources. Please try again with the next madVR build, when it is released. If it still doesn't work then, a debug log with the new v0.85.x build might be helpful.

Yes, Win 7 x64.

The question is, why am I the only user to report this? I can't be the only ATI user on Win 7. So chances are you won't be able to reproduce it, I guess.
Well, I tried, and I couldn't. In any case I've done a couple more DXVA related changes, so maybe there's a bit of hope that the issue might be gone for you with the next build.

[madshi]

I have an NTSC DVD that is telecined. madVR turns on deinterlacement and I change the source type to film but it seems like madVR is removing extra frames and the resultant motion isn't smooth.
This plays perfectly smooth for me, as long as I switch my display to 23.976Hz. It's quite unsmooth with 59.940Hz, though. Generally madVR does not handle 24p@60Hz well. This is on my to do list for some future version. At the moment for smooth playback you should use an exact movie framerate = display refresh rate match. Or use an exact multiply of the movie framerate for the refresh rate.

[BetA13]

madVR detects displays based on the EDID. It seems that for some weird reason the EDID block of your display changes sometimes after a reboot. I don't know if there's a random number that changes in the EDID block. If so, there could be a virtually unlimited number of new displays being in store for you... :( But more probably there will only be a specific number of different EDID blocks. The best way to handle this is to drag & drop the "identification" from the newly detected displays to your original display in the madVR device manager. This will tell madVR that all those EDID blocks identify the same display. After you've moved the identification, you can delete the "empty" display shells. Hopefully after you've done this with all newly detected displays, it might stop.

:goodpost:
Ok, i draged them into the main Display, Philips bla bla..
And deleted the others afterwards.. a quick test showed he didnt create a new one, so i guess thats solved :)
But i didnt reboot, so i will see tomorow what happens..

Thanks a lot again madshi..

Greetz BetA

[ajp_anton]

Generally madVR does not handle 24p@60Hz well. This is on my to do list for some future version.Will this be done with interpolation, blending, tearing (=)) or something else?

[madshi]

Blending.

[DragonQ]

Presumably at the moment it just displays frames 2,3,2,3...etc. times, just like EVR?

[madshi]

No, madVR calculates the best matching VSync interrupt to present each frame for. This can eventually result in uneven repeat patterns for 24p@60Hz. EVR might actually be smoother in such situations at the moment.

[ryrynz]

I'm having a hard time understanding why people don't zip text files before uploading them.

You could always add zip functionality as a step of creating the log files.

[Superb]

This plays perfectly smooth for me, as long as I switch my display to 23.976Hz. It's quite unsmooth with 59.940Hz, though. Generally madVR does not handle 24p@60Hz well. This is on my to do list for some future version. At the moment for smooth playback you should use an exact movie framerate = display refresh rate match. Or use an exact multiply of the movie framerate for the refresh rate.Oh, that explains why I have to set my TV (connected as secondary screen) to 23.976Hz for smooth playback w/ madVR!

[dansrfe]

This plays perfectly smooth for me, as long as I switch my display to 23.976Hz. It's quite unsmooth with 59.940Hz, though. Generally madVR does not handle 24p@60Hz well. This is on my to do list for some future version. At the moment for smooth playback you should use an exact movie framerate = display refresh rate match. Or use an exact multiply of the movie framerate for the refresh rate.

Is the sample telecined? Should the source type be "video" or "film"?

[secvensor]

Please, add ms in Time line!!
h:m:s.ms

[dankef]

I notice a few people have been talking about the 650ti and 660. I'm building a dedicated HTPC very soon (no gaming, no other tasks, just TV, video and music) and am trying to choose between the two for madvr. I will be watching a combination of (1) 1080p Blu-ray rips; (2) 720p internet video; (3) Australian broadcast TV recordings (usually 576i, 576p, 720p or 1080i); and (4) PAL DVDs (576p/i). I'll be watching mainly on a 1080p display and occasionally a 1440p monitor. I'm wanting the highest image quality within reason (Jinc 3/4 + AR) and preferably something I won't have to upgrade too soon and will hopefully cope with future upgrades madshi might implement. But I also want something pretty much silent and with fairly low power consumption. I'm guessing the Asus DC2 660 is probably worth the extra $80 over the MSI 650Ti Power Edition given these requirements?

Also, just to confirm, there's no real benefit in going from an i3-3220 on a H77 motherboard to something like an i5-3470 or 3570 on a Z77 motherboard for pure HTPC use is there? I understand the GPU is the most important component?

[MokrySedeS]

Blending.

How can I do this in avisynth?
Sorry if I'm gonig off topic but I'm curious how this would look like.

[sneaker_ger]

Well, I tried, and I couldn't. In any case I've done a couple more DXVA related changes, so maybe there's a bit of hope that the issue might be gone for you with the next build.

Actually, I didn't encounter this problem again with 0.85.3 yet. I was still waiting to report as it was an erratic problem and I wanted to give it some time before rushing to conclusions. These two (http://forum.doom9.org/showpost.php?p=1605263&postcount=16221) also seem to not have complained again and their reports seem to be very similar to mine.
Since I also had a seeking freeze sometime, maybe your changes "hopefully fixed: freezes with some DXVA decoders (MS, MPC-HC)" did the trick.

I still have to test out frame accurate seeking a bit, though.

The other thing about playback taking ~200ms to start I'm now suspecting it to not be madVR's fault. EVR seems to start with a black screen while madVR is transparent. I might have been misinterpreting EVR's black screen as the start of a video. I will also give this some more testing.

[ryrynz]

I'll be watching mainly on a 1080p display and occasionally a 1440p monitor. I'm wanting the highest image quality within reason (Jinc 3/4 + AR) and preferably something I won't have to upgrade too soon and will hopefully cope with future upgrades madshi might implement. But I also want something pretty much silent and with fairly low power consumption. I'm guessing the Asus DC2 660 is probably worth the extra $80 over the MSI 650Ti Power Edition given these requirements?

The 650Ti is fine, the 660 is definitely future proofing.


Also, just to confirm, there's no real benefit in going from an i3-3220 on a H77 motherboard to something like an i5-3470 or 3570 on a Z77 motherboard for pure HTPC use is there? I understand the GPU is the most important component?

That's right without any post processing your GPU one of your most important choices.

[madshi]

You could always add zip functionality as a step of creating the log files.
Writing the log file is not something that is clearly "done" at some point, so madVR wouldn't know when exactly to zip the file.

Is the sample telecined? Should the source type be "video" or "film"?
It is hard-telecined, so it's "film". If it were "video", playing it at 24Hz would produce millions of frame drops.

Please, add ms in Time line!!
h:m:s.ms
Why? I'm only updating the time line once per second, anyway, in order to save performance.

I'll be watching mainly on a 1080p display and occasionally a 1440p monitor. I'm wanting the highest image quality within reason (Jinc 3/4 + AR) and preferably something I won't have to upgrade too soon and will hopefully cope with future upgrades madshi might implement. But I also want something pretty much silent and with fairly low power consumption. I'm guessing the Asus DC2 660 is probably worth the extra $80 over the MSI 650Ti Power Edition given these requirements?
I would say yes. But I can't guarantee that even the 660 will be fast enough for future upgrades I might implement. Maybe it will be, maybe not. I don't know myself yet how power hungry future algorithms might be.

Also, just to confirm, there's no real benefit in going from an i3-3220 on a H77 motherboard to something like an i5-3470 or 3570 on a Z77 motherboard for pure HTPC use is there? I understand the GPU is the most important component?
I'd recommend choosing a CPU which can decode highest bitrate 10bit h264 in software, just to make sure you can play all sorts of contents that might show up in the near future. I don't know which CPU can do that and which can't. This was discussed just a couple of pages ago, maybe you can read back.

How can I do this in avisynth?
Sorry if I'm gonig off topic but I'm curious how this would look like.
The look should be very similar to playing back at 24Hz.

Actually, I didn't encounter this problem again with 0.85.3 yet. I was still waiting to report as it was an erratic problem and I wanted to give it some time before rushing to conclusions. These two (http://forum.doom9.org/showpost.php?p=1605263&postcount=16221) also seem to not have complained again and their reports seem to be very similar to mine.
Since I also had a seeking freeze sometime, maybe your changes "hopefully fixed: freezes with some DXVA decoders (MS, MPC-HC)" did the trick.

I still have to test out frame accurate seeking a bit, though.

The other thing about playback taking ~200ms to start I'm now suspecting it to not be madVR's fault. EVR seems to start with a black screen while madVR is transparent. I might have been misinterpreting EVR's black screen as the start of a video. I will also give this some more testing.
Good to hear.

[ryrynz]

Writing the log file is not something that is clearly "done" at some point, so madVR wouldn't know when exactly to zip the file.

What with freezes and crashes etc, yeah makes sense.

Madshi, can you foresee any higher requirements for MadVR scaling at least in the short term (up to v1.0) beyond using Jinc and anti-ringing?

[madshi]

Maybe.

[l0rdraiden]

I have read this guide http://imouto.my/madvr/ and there is an advice to use DXVA2 for intel GPU (i have HD4000)

note: If you have a Sandy Bridge/Ivy Bridge GPU, use DXVA2 for both image upscaling and image downscaling. nVidia and ATI users should not use the DXVA2 scaler algorithm at all.


Why is better DXVA2? It has better quality than the other filters? and what is your advice?

[ryrynz]

Not better quality, just less power consumption. Personally I'd choose pixel shader code resizers over DXVA2 resizers on Intel graphics any day of the week. The resizer used with DXVA2 is IMO too sharp.

[madshi]

DXVA2 *up*scaling on Intel GPUs uses a pretty good algorithm which is somewhat similar to madVR's Lanczos AR. However, it's not as good as madVR's Jinc AR. Also Intel's anti-ringing doesn't work for *down*scaling, as far as I can see. One further key limitation of Intel's DXVA2 scaling at the moment is that it only supports 8bit output, so it's not as banding-free as madVR's scaling is. The main advantage of using DXVA2 scaling with Intel GPUs is that (at least for upscaling) it has a good performance/quality ratio.

[dansrfe]

madshi,

Is there a way to solve the problem that occurs when a music video is played that was shot at 25 fps and one has checked the checkbox within madVR's settings that says to treat 25p as 24p for ReClock purposes? That changes the refresh rate to the closest multiple of 24p in the refresh rate list and obviously, in this case, it isn't appropriate to use ReClock to stretch the audio much less change the pitch of it to match 24 fps.

The easiest thing to do in my opinion would be to add a user interface option to enable/disable that checkbox. The reason why the display mode switcher user interface option isn't that helpful in this case is because I still need the refresh rate to be changed from 60Hz to 50Hz and back for true 25 fps content.

Thanks!

[secvensor]

Why? I'm only updating the time line once per second, anyway, in order to save performance.

For displaying the frame number.
To be able to pinpoint the frame exactly even in the player mode, without going to the editor.
To be sure which frame you're on.
0:06:04/0:06:04
>>>
8734 (0:06:04.281) [ ]/Frame 8743 (0:06:04.656) [K]

[mark0077]

Hi guys, have a really weird dropped frames issue, I really can't explain it. In Windows 8, 60hz fullscreen mode actually switches to 59.94 hz mode (not 100% sure if thats by design or a bug), but for me this isn't good as my TV goes to 59.90 or 58.89 hz at that rate, according to madVR's display Hz rate. So I have created custom resolutions that keep the display in 60.001~hz even when windows 8 tries to switch to 59.94. This works.

I'm interpolating everything to exactly the same rate in SVP, 60.001.

My issue is, madVR is dropping maybe 1 frame every 5-10 minutes. I have tried interpolating to other values above and below 60.001 but still get dropped frames. When I interpolate down to 59.94 hz, I get 0 frame drops.

Has anyone had any situation like this, its like as if nvidia is still doing something funny in 59.94hz rate, even though madVR shows 60.001Hz. madVR's composition rate shows 59.94 hz in windowed mode but that shouldn't effect anytying in FSE mode right?

I'm using a GTX 295 in Win 8 64bit.

[ajp_anton]

Writing the log file is not something that is clearly "done" at some point, so madVR wouldn't know when exactly to zip the file.Maybe you can launch a separate zipper process when madVR is closed?

How can I do this in avisynth?
Sorry if I'm gonig off topic but I'm curious how this would look like.convertfps(your_new_framerate)

[corporalgator]

At the moment I can't say for sure what is better. Just try it out with a video which has very smooth motion (e.g. a loooong slow camera pan). In the long run it will be better to let madVR do the 3:2, but that's future talk.

Have you tried whether your TV supports 72Hz?




It doesn't support it. I'll need a raise to get one of those. :(

[madshi]

Is there a way to solve the problem that occurs when a music video is played that was shot at 25 fps and one has checked the checkbox within madVR's settings that says to treat 25p as 24p for ReClock purposes? That changes the refresh rate to the closest multiple of 24p in the refresh rate list and obviously, in this case, it isn't appropriate to use ReClock to stretch the audio much less change the pitch of it to match 24 fps.

The easiest thing to do in my opinion would be to add a user interface option to enable/disable that checkbox. The reason why the display mode switcher user interface option isn't that helpful in this case is because I still need the refresh rate to be changed from 60Hz to 50Hz and back for true 25 fps content.
Looking at my code, madVR only seems to look for a 24p refresh rate multiple for 25p/50i content, if you have deinterlacing turned off. If you turn deinterlacing on, that shouldn't happen. Are you saying that madVR switches to a 24p multiple for 25p/50i content, even though you have deinterlacing turned on?

For displaying the frame number.
To be able to pinpoint the frame exactly even in the player mode, without going to the editor.
To be sure which frame you're on.
0:06:04/0:06:04
>>>
8734 (0:06:04.281) [ ]/Frame 8743 (0:06:04.656) [K]
I see. Could you please keep this wish in mind and repost it when madVR has reached v1.0?

I'm interpolating everything to exactly the same rate in SVP, 60.001.

My issue is, madVR is dropping maybe 1 frame every 5-10 minutes. I have tried interpolating to other values above and below 60.001 but still get dropped frames. When I interpolate down to 59.94 hz, I get 0 frame drops.
Audio clock deviation comes into play, too. I'd suggest to lower the interpolation framerate down until you get no more dropped frames. I guess it's probably somewhere between 60.001 and 59.940.