madVR - high quality video renderer (GPU assisted)

[leeperry]

That's weird. Maybe you should run a scandisk, just to be sure? :)


Thanks, I'd appreciate that. Tests with the timing mode and numbers of backbuffers would be great also.
It's on a ramdisk, I like it fast...BTW, you said you'd allow gamut mapping through PS scripts at some point, that'd be really sweet as loading 100MB LUT's takes quite a while...even from a ramdisk :o

well, I'm on XP so I guess I'm better off using 3 backbuffers anyway?

I dunno, I think the .ini resets to default from time to time...read-only fixes the problem.

[madshi]

The best combination seems to be no check/Aero/8. Note that the only way to see what setting is better for me is to intentionally set something else to the "bad" one eg. to see that the "upload frame in render thread" give me bad result, I have to set the buffer to 3 to be able to see it goes non-smooth.
Hmmmm... So if I recap, you get non-smooth results only when setting number of backbuffers to 3 *and* at the same time "upload frames in render thread" to on? If you revert either of these changes, things are smooth again? That's really a problem, because e.g. for Mark_A_W playback is only smooth with backbuffers set to 3. Anything higher and he gets problems. And for my setup, "upload frames in render thread" is significantly better. Both with my NVidia 9400 and with my ATI 3850. Argh, I *really* hoped we could end up with a setup which works for everyone!

Same goes for Aero timing and 8 buffer.
What do you mean with that?

I can't comment on flushing options for this setup, as I need no flush at all for 5770.
What happens if you set backbuffers to 3, enable the option "upload frames in render thread" and enable "flush" (no wait) after last render step? Does that make things go smooth again?

Now for the other setup

The best combination so far is no check/normal/8 and loop/flush/no/no. I still need that OSD enable trick, but once that's done, it's fairly ok. Only skipped frames once in a while. In this setup, Aero timing is the worse option to set. It's keep stuttering all the time.
I think that the flushing is needed here to work around the fact that the Aero composition rate is not good. The flushes probably help keeping the problems to a minimum.

It's on a ramdisk, I like it fast...
Could the ramdisk be responsible for the sometimes resetting ini file? I don't think anyone else has this problem!

well, I'm on XP so I guess I'm better off using 3 backbuffers anyway?
If you set to higher than 3, madVR will internally limit it to 3 on XP. So you will see no difference, if you set it to 3 or 8 or anything in between. But you could test with 1 or 2. That should behave differently...

[leeperry]

Could the ramdisk be responsible for the sometimes resetting ini file? I don't think anyone else has this problem!

If you set to higher than 3, madVR will internally limit it to 3 on XP. So you will see no difference, if you set it to 3 or 8 or anything in between. But you could test with 1 or 2. That should behave differently...
Yes, nevermind if I'm the only one complaining...I've also had a complete graphic card hardcore freeze when feeding 720p@50fps h264 to CoreAVC CUDA, I guess I was pushing it a bit..a reset wasn't enough, my P5K Deluxe mobo had to force a cold reboot.

Oh, I didn't think going <3 would make any point? the more buffering, the better...or not?

The other problem w/ testing smoothness in mVR...is finding material that's smooth enough in the first place. I was just trying the JPN BD of "Monster Inc", I dunno wth they did but it's *really* not smooth...the smoothest I've seen is "Space Chimps", I just don't feel like watching it in loop when changing options in mVR :p

PS: Dragon Hunters is also amazingly smooth, but "a bug's life" hiccups constantly..

[6233638]

Can you please add some more comments about the new tweak options? Which needs to be set which way for you? Which options impact performance in which way? Please start testing with a setting that works well for you, and then modify the tweak options one by one, so you can comment on the separate options. Thanks!It's too late tonight but I will do lots of testing for you tomorrow - had hoped to do it sooner. There definitely did seem to be a change between normal and high timing before, with high being better. I will investigate this properly tomorrow, it might just be a numbers thing rather than actual performance.

I am assuming I can count on dropped frames as a somewhat reliable indicator though. It never falsely reports them and results are consistent. None of the more taxing upsampling options are close enough yet that tweaking might result in completely smooth playback (like bicubic was previously for me) and previously judging by eye just chose what dropped frames more evenly rather than what actually dropped the least number of frames.

--

Lee, I think you are getting a bit too concerned about some aspects of performance. RAM disks? Sticking to XP because of the number of services?
I hated Vista and stuck with XP but 7 is better on any modern computer. I would only use XP on a low-end machine that can't cope.

I'm also concerned about the comments like "I've never seen mVR this smooth tbh" that you post every version or so. Once you get to the point where your system can handle it, it shouldn't be possible for things to get any smoother. (and I don't believe you are watching with settings your GPU can't handle)

Performance improvements might let your system use higher quality upsampling though, like how I have gone from bilinear to bicubic when my system could not handle it in older versions.

[mark0077]

madshi, would it be difficult on your end to add some sort of benchmark that could help us gather data for you....., perhaps when activated, it could cycle through some permutations of different options, gather some stats into a big log for you.

Just an idea. Perhaps it could be used in a future release to help someone obtain "optimal" settings for their particular setup... ie the combination that yields the..... lowest frame drops / or whatever the benchmark may be..

[LloydA]

(1) Which backbuffer count works best for you? Which is maybe not best, but still acceptable? Which doesn't work at all for you?

I have noticed that with backbuffer > 3, slow camera pans appear smoother yet at the same time "drag". Hard to describe, it appears as if it runs smooth but too slow for about half a second, then catches up fast, and so on.

(2) Which timing mode works best for you?

High for me.

(3) How about "upload frames in render thread" and "copy 1 backbuffer at a time"?

Ticking any one of those produces the same effect as with backbuffer > 3.

Also, on my setup turning off all flush options produces the smoothest pans. Tested with "Dead Man's Chest" (1080p Blu-ray MKV), "A Bug's Life", and the synthetic judder test files from AVForum (23.976p).

OS: W7 x64
GPU: ATI 5450 512MB, Catalyst 10.6
Aero: off
Display: 1920x1080, 60Hz, over HDMI

[leeperry]

RAM disks? [..]
I'm also concerned about the comments like "I've never seen mVR this smooth tbh" that you post
because the world is not just black or white? madshi keeps improving the smoothness for high refresh rates(I run 96Hz these days), you should try it sometime.

yes sir, ramdisks! It all started coz KMP would take forever to open, but I like it like that now...all my media just opens right away. It's also nice to open LUT files as fast as possible.

and when using superspeed ramdisk, you can use the >3.2GB RAM on a 32bit OS, very cool feature.

[B.F.]

Maybe eventually 64bit could help for video decoders, I don't know. But from what I've read so far, e.g. CoreAVC 64bit is not faster than 32bit. madVR itself will most probably not benefit from 64bit at all, and GPU drivers probably not, either. The real work in done by the GPU, and that is using its own bitdepth, anyway, and it not affected by the bitdepth used by the OS/CPU. madVR and the GPU drivers are mostly just sending commands to the GPU, uploading data to the GPU and waiting for replies from the GPU. Such kind of code doesn't benefit from 64bit at all.
Thats not the point.
It's just the question of compatibility with 64bit players and codecs. Even if madVR itself don't have benefits from 64bit, if the decoder or something else will - the player itself will work faster. :)
And with madVR MPC-HC 64bit will work faster because madVR is the fastest render avaible.

[namaiki]

Thats not the point.
It's just the question of compatibility with 64bit players and codecs.
...
But, what component are you are using is only available in 64 bit? eg MPC-HC is available in 64 and 32 bit.

Even if madVR itself don't have benefits from 64bit, if the decoder or something else will - the player itself will work faster.
...
Ideally, it would, but it doesn't. At least, this shouldn't be a focus for now.


By the way, I did read all of the rest, but I still can't come to any real conclusions about flush settings, etc. D:



Also, Hypernova, please consider which settings you can use for normal watching, not only that seems to be working better, when it's just not working at all.

Also, other members, remember that High refresh rate only applies when the video to screen refresh rate is >2.3x for example ~24fps to 60Hz. So if your screen and content are matched, it won't effect.

Also, other members, please consider if your screen and content are perfectly matched or not, otherwise there may be drift or whatever between the two and it might not be smooth. To me, sounds like what Hypernova is experiencing if he's watching 30fps content.

Also, other members, please check playback with at least ~24 and ~30fps content on your screen. Look at pans in particular, or even scrolling credits.

Madshi or venturous other, when you are ready, please post up a complete list of settings that you think might work for most people that everyone can test.

[lych_necross]

I need more feedback on the new windowed mode tweak options!
Windows 7 64bit
Aero Enabled
1440x900 @ 60hz
Single monitor
Nvidia Geforce 9800 GTX+ (257.21)
CoreAVC w/CUDA
Reclock w/v-sync off


Backbuffer=8 seems to work the best for me. When set to 3 or less, the video doesn't play very smoothly.
All three modes seem to work (none crashed MPC); however, normal seemed to be the smoothest. Aero is a close second. The perceivable difference between Normal and Aero is very small (almost non observable for me).
I couldn't notice a difference either way by playing with the other two checkboxes.

The flush options, I leave at the defaults because they seem to work the best for me.

[madshi]

Oh, I didn't think going <3 would make any point? the more buffering, the better...or not?
I thought so, too, but then several people reported that 8 is not working well for them. Soooo, maybe 1 or 2 is even better than 3? Shouldn't be the case, but who knows?

The other problem w/ testing smoothness in mVR...is finding material that's smooth enough in the first place. I was just trying the JPN BD of "Monster Inc", I dunno wth they did but it's *really* not smooth...the smoothest I've seen is "Space Chimps", I just don't feel like watching it in loop when changing options in mVR :p

PS: Dragon Hunters is also amazingly smooth, but "a bug's life" hiccups constantly..
Do you get dropped/delayed frames with Monster Inc or Space Bug's Life? Is Bug's Life 1080p24? Or 1080i60?

I am assuming I can count on dropped frames as a somewhat reliable indicator though.
Yes. Dropped and delayed frames should be more or less reliable. However, if you dial in settings which are too hard for your GPU, so that dropped and/or delayed frames are constantly increasing, then these values do lose a bit of their meaning. I mean: It doesn't matter if you get 10 drops per second with one settings combination compared to 20 drops per second with another settings combination. It's possible that one settings combination simply handles dropped frames better than another. But what we really want is no dropped frames at all. So ideally you should try to find a scaling / zoom combination which is smooth with some settings combinations and non-smooth with others. If Bicubic is smooth for you, but Lanczos3 is not, then you could try lowering Chroma to Bilinear. Maybe that will allow you to use Lanczos3 on Luma with some settings combination. Also you can try Bicubic and just zoom the image a lot. That will also increase the stress on madVR.

madshi, would it be difficult on your end to add some sort of benchmark that could help us gather data for you....., perhaps when activated, it could cycle through some permutations of different options, gather some stats into a big log for you.
The problem is that stats alone don't tell the full story. E.g. Aero sometimes stutters without any stats showing that. Also some settings can't be changed without restarting video playback. All in all I don't think I can create a reliable benchmark, unfortunately. I depend on your eyes.

I have noticed that with backbuffer > 3, slow camera pans appear smoother yet at the same time "drag". Hard to describe, it appears as if it runs smooth but too slow for about half a second, then catches up fast, and so on.

[...]

Ticking any one of those produces the same effect as with backbuffer > 3.
Weird. Do you get dropped/delayed frames with the settings you normally use? And do you get any dropped/delayed frames when this "drag" effect occurs? Thanks.

It's just the question of compatibility with 64bit players and codecs. Even if madVR itself don't have benefits from 64bit, if the decoder or something else will - the player itself will work faster. :)
"If the decoder or something else will". Yes, but currently there is no decoder and nothing else which is faster in 64bit than in 32bit! Some things are even noticeably slower in 64bit. E.g. Deband is broken in 64bit, but works well in 32bit.

[madshi]

Windows 7 64bit
Aero Enabled
1440x900 @ 60hz
Single monitor
Nvidia Geforce 9800 GTX+ (257.21)
CoreAVC w/CUDA
Reclock w/v-sync off


Backbuffer=8 seems to work the best for me. When set to 3 or less, the video doesn't play very smoothly.
All three modes seem to work (none crashed MPC); however, normal seemed to be the smoothest. Aero is a close second. The perceivable difference between Normal and Aero is very small (almost non observable for me).
I couldn't notice a difference either way by playing with the other two checkboxes.

The flush options, I leave at the defaults because they seem to work the best for me.
Thanks!

Is there anybody with Aero turned on, who benefits from setting backbuffers to smaller than 8?

Is there anybody with Aero turned off, who benefits from setting backbuffers to higher than 3?

If that's a "no" to both questions, I could set backbuffers to 8 for Aero. And to 3, when not using Aero.

[leeperry]

Do you get dropped/delayed frames with Monster Inc or Bug's Life? Is Bug's Life 1080p24? Or 1080i60?
They're both 1080p24 from the genuine BD's, it's just that their telecine transfer is not smooth...it's clearly visible in the first seconds of Bug's Life(when it slowly pans over a hill), and at the very beginning of Monsters Inc.(when there's a slow pan going over a mixing board, or even worse, right after that scene when there's a car going up in the street..it's hiccuping a bit in the end). Why would they do telecine transfers on anime is beyond me?! Space Chimps is clearly a direct digital transfer and it looks 48fps'ish in mVR.

Anyway, I've found a movie w/ a perfect telecine transfer(RocknRolla) and yesterday w/ the options I mentioned I've had a clearly visible random delayed frame after 30 mins...so I'll play around w/ the options, first I'll try to disable "upload frames in render thread". Its lossless soundtrack is amazing, I don't mind watching it 10 times for science sake's :D

[Razoola]

I think the biggest issue here is what one person says is smooth another would say is stuttering so its really hard for madshi to really know what's what.

For me at least its always smooth no matter the buffer size I choose. Basically everything is 100% with the default settings here.

I think the best solution is to keep all the options for users to play with but have them in an advanced tab with either a good readme.txt or mouseover help popup for each option to explain what it does (its thats easily possible). Then on the advanced tab have a button to reset all advanced options to their defaults for those who really mess things up there.

[B.F.]

Thanks!

Is there anybody with Aero turned on, who benefits from setting backbuffers to smaller than 8?

Is there anybody with Aero turned off, who benefits from setting backbuffers to higher than 3?

If that's a "no" to both questions, I could set backbuffers to 8 for Aero. And to 3, when not using Aero.

I had a Win7 with Aero on.
But I have quite a slow card (gf8500gt) and fullhd monitor.
It work good only with bilinear resize and backbuffers to 3.
So my guess is the number of backbuffers more depends on hardware.

[Hypernova]

Hmmmm... So if I recap, you get non-smooth results only when setting number of backbuffers to 3 *and* at the same time "upload frames in render thread" to on? If you revert either of these changes, things are smooth again? That's really a problem, because e.g. for Mark_A_W playback is only smooth with backbuffers set to 3. Anything higher and he gets problems. And for my setup, "upload frames in render thread" is significantly better. Both with my NVidia 9400 and with my ATI 3850. Argh, I *really* hoped we could end up with a setup which works for everyone!


Exactly as you said.


What do you mean with that?


I mean to see that aero is better than normal, I have to also set the buffer to 3 (the "bad" one for me) to see that aero is better.


What happens if you set backbuffers to 3, enable the option "upload frames in render thread" and enable "flush" (no wait) after last render step? Does that make things go smooth again?


Yes, the flushing helps

I want to add that I'm actually not so sure on the aero vs. normal because I actually test with 24p on 60Hz display, so I might just seeing 3:2 judder for the case. I don't have a 30p material that can actually push the limit. It's always smooth on one 1080p/30 anime (Air) op that I got, no matter what setting except super low buffer like 1. I'm still pretty sure on the buffer and upload in render thread though. I can see the different and no check+8 buffers looks really better. Wish I can do a blind test, however.


I think that the flushing is needed here to work around the fact that the Aero composition rate is not good. The flushes probably help keeping the problems to a minimum.


Totally agree here. The last two options never do any good (or bad) for me though.

Also, Hypernova, please consider which settings you can use for normal watching, not only that seems to be working better, when it's just not working at all.

I'm sorry. I don't understand what you mean. I'm guessing that you are asking why I have two setups? I report two settings because I use both. Dual monitors is for "serious" watching. I still don't want to use single on my projector unless absolutely needed because it will messed up the windows position on my main display when resolution changed. For previewing and a short or non-serious watch, I go by my main monitor. For now my "absolutely serious" watch is with EVR Sync exclusive mode as that is guarunteed to have no skipping.

[Peuj]

Yes. Dropped and delayed frames should be more or less reliable. However, if you dial in settings which are too hard for your GPU, so that dropped and/or delayed frames are constantly increasing, then these values do lose a bit of their meaning. I mean: It doesn't matter if you get 10 drops per second with one settings combination compared to 20 drops per second with another settings combination. It's possible that one settings combination simply handles dropped frames better than another. But what we really want is no dropped frames at all. So ideally you should try to find a scaling / zoom combination which is smooth with some settings combinations and non-smooth with others. If Bicubic is smooth for you, but Lanczos3 is not, then you could try lowering Chroma to Bilinear. Maybe that will allow you to use Lanczos3 on Luma with some settings combination. Also you can try Bicubic and just zoom the image a lot. That will also increase the stress on madVR.


So this means to tests the "windowed mode tweaks" it's preferable to use first the lowest scaling settings before , no?


The problem is that stats alone don't tell the full story. E.g. Aero sometimes stutters without any stats showing that. Also some settings can't be changed without restarting video playback. All in all I don't think I can create a reliable benchmark, unfortunately. I depend on your eyes.


I've thought about something like this also.
Maybe not usable for a benchmark but as the tests needs a lot of time, this could really help.
If I've understood well the "dropped and delayed frames" are a good indicator.
So based on the dropped and delayed frames "cycle through some permutations of different options" like flush settings could help us eliminate not "supported" settings.

[makakam]

Windows 7 32bit
Aero ON
1920x1080 @24 (depending on the movie)
Reclock (vsync off)
Single display
CoreAVC
Ati 5770
All flush options at default

To me it seems there's no difference if I change the backbuffer to 3 or less or if I leave 8. Either way I get no delayed/dropped frames and the playback is smooth.

[madshi]

Anyway, I've found a movie w/ a perfect telecine transfer(RocknRolla) and yesterday w/ the options I mentioned I've had a clearly visible random delayed frame after 30 mins...so I'll play around w/ the options, first I'll try to disable "upload frames in render thread". Its lossless soundtrack is amazing, I don't mind watching it 10 times for science sake's :D
One delayed frame after 30 mins? Ouch, with problems so rare testing must be painful for you!

I think the best solution is to keep all the options for users to play with but have them in an advanced tab with either a good readme.txt or mouseover help popup for each option to explain what it does
I'd really like to avoid that, but if it's necessary, I'll have no other choice.

I had a Win7 with Aero on.
But I have quite a slow card (gf8500gt) and fullhd monitor.
It work good only with bilinear resize and backbuffers to 3.
So my guess is the number of backbuffers more depends on hardware.
Ah, thanks! So there goes my idea.

BUT I just had another idea. I'll change the backbuffer logic a bit in the next version. I hope that this will allow best performance for all people with the max number of backbuffers available by the OS. Then I could at least drop the backbuffer count option.

Yes, the flushing helps
So, with the configuration I suggested, playback is exactly as smooth as with your favourite configuration? Or is it still worse?

I'm still pretty sure on the buffer and upload in render thread though. I can see the different and no check+8 buffers looks really better. Wish I can do a blind test, however.
Your feedback on the "upload in render thread" bothers me. Most other people report improvements or no change... :(

So this means to tests the "windowed mode tweaks" it's preferable to use first the lowest scaling settings before , no?
No, try scaling settings which are as high as possible, but still smooth. Then try whether any of the tweak options make things worse. Also when you have smooth playback, try zooming in as much as necessary to make things go non-smooth. Then try tweaking the options to see if some of the options allow things to go smooth again.

Please note that some of the options (specifically number of backbuffers and "upload frames in render thread") require a restart of the media player or at least a reloading of the video file.

Ati 5770

To me it seems there's no difference if I change the backbuffer to 3 or less or if I leave 8. Either way I get no delayed/dropped frames and the playback is smooth.
Well, that's your own fault for buying such a fast GPU... :p

[nlnl]

Mpc HC (Haali mkv splitter, CoreAVC, WMVideo Decoder DMO) + ATI 5400 (10.6) + Vista 32 + Aero on + 0.21 + monitor input 1080p@23, 24, 50 (single)

A little issue with 0.21.

The file is 720p50 upscaled to 1080p50 (bicubic75/bicubic75/soft100, win mode tweaks: default). Very smooth playback, no dropped or delayed frames.
But if I press space bar 12-18 times (pause, 3 seconds, play, ...), madVR starts dropping frames.

Can not make log file. Using madVR debug I can not get smooth playpack (influence of disk writing?).
Does it keep dropping frames from that point on? Or is it just a phase it goes through and gets non-dropping again after a few seconds?

madshi
Yes it does keep constantly dropping frames from that point.
Can not stop it.
I can reproduce it very reliably.
But when I play 1080p23 at 1080p23 display (no scaling) can not reproduce that.
Thank you :)

Edit:
Set everything to flush and wait (loop) and I can not reproduce dropping :).

[LloydA]

Weird. Do you get dropped/delayed frames with the settings you normally use? And do you get any dropped/delayed frames when this "drag" effect occurs? Thanks.

I only get 1-3 dropped/delayed frames right at the beginning of playback, regardless of madVR settings (probably caused by Haali Splitter buffering segments), and then no further dropped/delayed frames.

The problem is that stats alone don't tell the full story. E.g. Aero sometimes stutters without any stats showing that.

That's what I was thinking here, too. Are you sure that the scanline detection is working properly? In my setup I route the video signal over an A/V receiver (I have no other option), and the video processing unit in there might be introducing latency (though I have turned off all options in the receiver's HDMI menu, but who knows).

Is there anybody with Aero turned off, who benefits from setting backbuffers to higher than 3?

Not me.

They're both 1080p24 from the genuine BD's, it's just that their telecine transfer is not smooth...it's clearly visible in the first seconds of Bug's Life(when it slowly pans over a hill)

That's what I'm using for my madVR tests as well. But there's no "telecine transfer" here, all BD releases of this title are straight from the same master encoded 100% in the digital domain. So the stuttering was rendered like that, I'm afraid.

Another good Pixar title to test is "Up". At around 00:34, where the ole man walks at the cliffs of the plateau, you get very smooth pans. Then suddenly the "camera" pans to the right to show the edge of the plateau, and this sequence also stutters like hell. Nothing can be done about it, it's how it was rendered.

Your feedback on the "upload in render thread" bothers me. Most other people report improvements or no change...

Ticking the option make things worse for me, as well.

[leeperry]

One delayed frame after 30 mins? Ouch, with problems so rare testing must be painful for you!
yeah, it takes forever...I just tried those settings but it was hiccuping some frames from time to time after 1H in RocknRolla at 23.976@96.003Hz(not dropped, and it also didn't increment in mVR's OSD): http://thumbnails32.imagebam.com/8589/b9d07385887696.jpg (http://www.imagebam.com/image/b9d07385887696)

I'll try disabling the single backbuffer copy thingie...but anyway, windowed mode is more or less ripe I think? If you wanna watch a full-length feature film worry-free, D3D FS is the uncompromising solution IMHO. Nothing forces the VSYNC like an hardware interrupt, I think we have yet to see a game randomly hiccuping? I feel like everything's fighting against us in windowed mode trying to sink the ship :mad:

I forgot to enable logging, I'll do that next time..

That's what I'm using for my madVR tests as well. But there's no "telecine transfer" here, all BD releases of this title are straight from the same master encoded 100% in the digital domain. So the stuttering was rendered like that, I'm afraid.
Yes, amazing! Do these ppl watch their final rendering in 60Hz? It sure feels like those 2 Pixar movise haven't been authored in 24p in the first place.

I think we'd need to make a list of butter smooth movies, so testers don't waste their time...IIRC Bee Movie was also amazingly smooth.

[Trigunflame]

Madshi, I'm noticing that a few of my older videos do not seem to display in the correct Aspect Ratio; they work fine in EVR.
From what I can gather, it looks as if the DAR is not being respected. To be more specific, it seems the "Keep Aspect Ratio" option for MPC-HC is no longer working when in conjunction with MadVR.
With EVR & the aforementioned MPC-HC option enabled, the appearance of certain video will be adjusted according to its DAR; for MadVR, it has no effect whatsoever and the video remains incorrectly sized.

I've included a log & mediainfo output for two such files.
http://www.multiupload.com/JKPZUP3QRK
http://www.multiupload.com/8UW57R41S2

OS: Win7 x64 ultimate
Aero: Enabled
CPU: i7 920
GPU: GTX 480
Drivers: NV 257.21
Resolution/Refresh Rate: 1680x1050 59/60hz
Player: MPC-HC svn (r2075)
Codec: FFDSHOW-Tryouts (r3483)
Splitters/Filters: AVI Splitter > FFDSHOW-Tryouts > MadVR

[djsolidsnake86]

0.21 with mph hc latest svn, hd4850, and hd files is a little slow in some cases

[Hypernova]

So, with the configuration I suggested, playback is exactly as smooth as with your favourite configuration? Or is it still worse?

As far as I can tell on 24p/60Hz, there is no difference. It's really hard to tell to begin with.


Your feedback on the "upload in render thread" bothers me. Most other people report improvements or no change... :(


I would say don't worry about it then. I don't know what will happen when I switch back to 3870 at the end of summer, but as far as 5770 goes, there is nothing to worry about. Maybe it's because I got CPU with so many threads (eight)?

[Peuj]

(1) Which backbuffer count works best for you? Which is maybe not best, but still acceptable? Which doesn't work at all for you?
8 seems better for me than 3. Even if I didn't notice a huge difference.
(2) Which timing mode works best for you? Which is maybe not best, but still acceptable? Which doesn't work at all for you?
Aero doesn't work well for me for sure. Normal seems the best even if I don't really see a difference with High.
(3) How about "upload frames in render thread" and "copy 1 backbuffer at a time"?
I don't see any difference.

My best settings are the default.

Note that I always try to match the video fps with the screen (TV) resolution.

GPU: nvidia 8600M GT with driver 257.15
OS: Vista
Aero: on
1920x1080 24Hz (TV)
Dual monitor PC/TV

Thanks

[leeperry]

been running quite a bit of testing in the past hours, and this seems like the winning combo for me:
upload frames in render thread=0
copy only 1 backbuffer at a time (instead of 2)=0
timing model=high refresh rate

so it's like 0.18 but w/ the new timing model for 96Hz...about time I try 48Hz on the projector anyway :p

[naomatrix]

Thanks madshi. MadVR good program, and for me the best renderer, but I have a small problem, I can not completely remove the dropped frames on my configuration in Win XP. Only on Win 7, with very weak preferences (upscaling, downscaling) for my configuration.I've tried many settings, but only if these no missing frames - http://i1.imageban.ru/out/2010/06/26/e36665e1e6a38645e24102ad5add1d12.png

Athlon 64 x2 5600+(2800Mhz)/2Gb
MB Palit-780G
1440x900 @ 75hz/Single monitor
Nvidia Geforce 9800 GT/512Mb
Windows 7 Ultimate 32bit[nvdrv-196.21]+Aero Enabled
( or Win XPprofSP3 32 bit[nvdrv-257.21])
DirectX june 2010
PotPlayer1.5.22627+built-in splitter/decoder(FFmpeg-MT)+madVR0.21( on 720p and 1080p)
No ReCLock
Strange but with CoreAVC 2.0 is constantly dropped frames.

[Mark_A_W]

Ok, how about 2 or 1 backbuffers? Better or worse?



madshi

1 or 2 backbuffers seem to work as well as 3.


Above 3 is where the trouble starts.


Also, I *think* that the video is smoother with Reclock fully enabled. Rather than loaded, but "disabled" (just a wasapi renderer).

[leeperry]

ok, not sure if that helps...but anyway, that's RocknRolla in 23.976@96.003Hz w/ Reclock. I ran it, went away for 45 mins, came back and it had one delayed frame(XPSP3/8800GS/latest Forceware/Reclock/PotPlayer): http://thumbnails30.imagebam.com/8595/081f3585947558.jpg (http://www.imagebam.com/image/081f3585947558)
these are the mVR settings, and I guess it's about time I flush something: http://thumbnails33.imagebam.com/8595/cfe3fc85947559.jpg (http://www.imagebam.com/image/cfe3fc85947559)

and that's the logfile: http://www.mediafire.com/?yx45ztrk2om

[ianken]

I've got some 1080p files that do not render correctly. Some weird effect that looks like chroma offset vertically in the image.

And I did search. :-)

No biggie. My HTPC runs on a 9300 integrated GPU which lacks the GPU beefcake to really make this renderer shine.

EDIT: Ahh. It's the internal decoder in MPC-HC. FFDShow Tryouts works like a charm.
EDIT2: CoreAVC2.0 work correctly as well. Whheeee!

So frustrated this is only performant on my gaming rig. Poor HTPC rig (evga 730i/9300,QX6700) is just not up to it. More research this evening.

Anyway, props to the developer. This is very cool.

[mark0077]

ok, not sure if that helps...but anyway, that's RocknRolla in 23.976@96.003Hz w/ Reclock. I ran it, went away for 45 mins, came back and it had one delayed frame(XPSP3/8800GS/latest Forceware/Reclock/PotPlayer): http://thumbnails30.imagebam.com/8595/081f3585947558.jpg (http://www.imagebam.com/image/081f3585947558)
these are the mVR settings, and I guess it's about time I flush something: http://thumbnails33.imagebam.com/8595/cfe3fc85947559.jpg (http://www.imagebam.com/image/cfe3fc85947559)

and that's the logfile: http://www.mediafire.com/?yx45ztrk2om

If your 23.976 isn't getting changed by reclock to a speed that fits more evenly into 96.003 then 1 delayed frame seems to be unavoidable IMO.

96.003 % 23.976 is 0.01, and that catches up to make 1 frame in 23.976 / 0.01 = 2397.6 seconds.... about 40 minutes... so 1 delayed frame every 40 minutes or so.

[lych_necross]

Is it possible that the use of coreavc (w/cuda) could be affecting the results? Should we be testing with different decoders in both DXVA and software mode just to be safe?

[leeperry]

If your 23.976 isn't getting changed by reclock to a speed that fits more evenly into 96.003 then 1 delayed frame seems to be unavoidable IMO.
of course Reclock runs in "auto" mode..no "slave" here.

[namaiki]

Is it possible that the use of coreavc (w/cuda) could be affecting the results? Should we be testing with different decoders in both DXVA and software mode just to be safe?

Certainly. Please stick with software decoders. They are a lot faster and have less latency, more time for the renderer. Put another way, DXVA can offload from the CPU, but it's not actually faster.

[XPC]

Hi Madshi,

I've been testing a patched MPC-HC version that avoids creating more than one madVR instance at the same time and I could get it working for both SD and HD. So this was the problem.

I need to do more testing to ensure it is stable, though.

Wow, that is weird. It seems that two madVR instances are created after each other, and then one of them is deleted again, while the other one stays active. madVR currently does not expect/support such a behaviour. As a result the surviving madVR instance refuses to display anything in your situation. I'll have to change some things to support having multiple madVR instances running at the same time. <Argh>

[nevcairiel]

I've been playing with all settings, and honestly, i cannot see any real difference in real 1080p samples or the judder test files from avsforums. The 24p file is obviously not so smooth at my 60hz (obvious 3:2 judder), but the 29.97fps file plays fine. Its not 100% perfect, but no setting that i adjust seems to improve it (or change it for the worse). I'm not even sure how to describe the effect i'm seeing, might just be "normal". On the judder samples, Smooth motion, but some kind of flickering on the border of the moving white block.

Maybe all those settings only make a real difference on slower systems?
I'll be building a HTPC in a while, and make sure to run the tests on there again ;)

Specs:
Win7 x64, Aero on, 1920x1200@60
NVIDIA Geforce GTX 260 - 257.21
Core i7 860

[yesgrey]

madshi,
I'll tell you a secret... my main reason for stop using W7 and roll back to XP was madVR. The previous version (before you started its development again) always had tearing, no matter the settings I chose, but strangely that only happened with my projector (1360x768 @48Hz). When using my crt monitor (1280x960@72Hz) I had not tearing at all. I thought that it might be some problem with the timings I was using for creating my projectors working resolution, but they were perfect with XP, so I simply decided to quit W7.

Yesterday I've made some testing with W7 x64 to see how the new versions are working with it. Sadly, with Aero disabled, I still get the tearing, no matter what combination of settings I use. Even with all settings at Flush and loop I have tearing. Then, I tried switching from the Classic GUI to the Aero GUI, and finally I got no tearing at all. The settings did not matter, any combination gave me no tearing. Unfortunately, this is still not good enough, because I don't like the idea of having to switch to an Aero GUI whenever I want to watch a movie... so for now I will still stick with XP. ;)

To be honest, that's not a big issue, because I always watch movies at fullscreen, so I think your new full screen exclusive mode would be the one!... :)

Just as a final note, I want to add that with XP x64 madVR is working great. Unfortunately my GPU is not powerful enough to use better scaling algorithms, but since I only need them for downscale, that's not a big issue, because bilinear works pretty good.

My setup:
- W7 x64 / XP SP2 x64
- GF 8600GT 256 MB (257.21 drivers)
- mpc-hc 1.3.2061 / ffdshow-mt
- Intel E2160 @ 2.7GHz 4GB RAM
- Critical resolution: 1360x768@48Hz

[iSunrise]

I need to know (and please include your OS, Aero, graphics mode / refresh rate, single/dual monitor, GPU information):

(1) Which backbuffer count works best for you? Which is maybe not best, but still acceptable? Which doesn't work at all for you?
(2) Which timing mode works best for you? Which is maybe not best, but still acceptable? Which doesn't work at all for you?
(3) How about "upload frames in render thread" and "copy 1 backbuffer at a time"?
Windowed mode is already working so well, I found it to be really hard to find a movie to actually see any differences. I finally found a blu-ray trailer, though, which has several very sensitive pans, so I will just go on and post you my settings and results.

I´ve just watched a dark knight blu-ray 1080p 23.976fps trailer (it´s included on the german batman begins blu-ray, not sure about other countries) for half an hour and also did several repeats (including restarting the player several times for all the options to take effect) of some of the included panning scenes, e.g. where joker is about to enter the vehicle they´re using to enter the bank or when the robbers are using the wire to get to the top of the bank. To my eyes and with my setup I can definately see an improvement when:

"No of Backbuffers" is set to "1" (lowest available) instead of "8", "upload frames in render thread" is off and "copy only 1 backbuffer at a time" is on. Everything is set to "don´t flush". Timing mode is set to "aero". The pans are extremely smooth, even when running my display at 59.950Hz with these settings. No delayed or dropped frames. Setting the timing mode to "high refresh rate" seems to work equally as well, at least I can´t make out any differences.

Also, when using the above settings, skipping is faster and switching from windowed playback to fullscreen is definately faster than the current defaults. It is practically instant.

It seems the setting that is responsible for the weird looking pans is "copy only 1 backbuffer at a time". This needs to be checked/on always or otherwise, pans won´t be perfectly smooth (at least on my system). "Upload frames in render thread" is was always off and setting it to on doesn´t do anything at all here.

Going with these results and reading some of the reports from others, it looks like people with very fast GPUs, very few background services, processes or tasks can set madVR to very "relaxed" settings and have the best experience, while people with a lot of background activity, slower desktop or integrated GPUs will need flushing and backbuffers to keep it smooth.

Specs:
(1) Geforce GTX 260-216 [Nvidia Quadro 258.49 WHQL]
(2) Windows Vista x64
(3) Aero on
(4) 1920x1200@59.940Hz (normal use)
(5) PotPlayer, KMPlayer or MPC-HC
(6) ffdshow video and audio decoder
(7) Haali Media splitter

madVR settings:
(1) Use 3DLUT
(2) Bicubic75, Spline64, Bicubic75
(3) don´t flush for everything

[djsolidsnake86]

i have e8400, hd4850, 4gb of ram and win7 x64 with mpc hc latest svn
what are the best settings on madvr 0.21 for my config? please

[fuzz!]

good to see madVR is being worked on again :)

don't suppose there'll ever be support for DVD playback?

[LoRd_MuldeR]

I came across some strange artifacts with MadVR (currently using v0.21). Seems to happen in "dark" areas.
Also I think it's not a decoder issue, because I used DivX H.264 Decoder with different renders, only MadVR showed these artifacts.
Furthermore there's no such problem in MPlayer (libavcodec), using the OpenGL renderer...

MadVR:
http://img36.imageshack.us/img36/6757/madvr021.png

EVR:
http://img822.imageshack.us/img822/6853/evr.png

Haali's Renderer:
http://img340.imageshack.us/img340/9471/haali.png


Here's another instance:
http://img690.imageshack.us/img690/8350/madvr021inst2.png


Any ideas :confused:

[nevcairiel]

Are you using 3DLUT?
If yes, make sure to update yCMS, and clear your 3dlut files so they get re-generated.

[LoRd_MuldeR]

Are you using 3DLUT?
If yes, make sure to update yCMS, and clear your 3dlut files so they get re-generated.

That was it! I checked all the options, except for that one. Sorry :o

Without "use 3dlut" all is fine again :thanks:

[djsolidsnake86]

i have e8400, hd4850, 4gb of ram and win7 x64 with mpc hc latest svn
what are the best settings on madvr 0.21 for my config? please

please i have seriously slowdowns with madvr and mpc!

[yesgrey]

That was it! I checked all the options, except for that one. Sorry :o

Without "use 3dlut" all is fine again :thanks:
It was a bug in yCMS. Get v1.4 and it should work fine.

[LoRd_MuldeR]

It was a bug in yCMS. Get v1.4 and it should work fine.

Confirmed :)

[Hypernova]

please i have seriously slowdowns with madvr and mpc!

What kind of slowdown? Your setup should be doing fine with default setting. If not, start playing around with the options. I suggest in this order: Flushings, timing, # backbuffer, and the two checkboxes.

[leeperry]

I've watched a bunch of +90 mins 23.976fps movies w/ the latest settings I posted, everything's pretty darn smooth in 96Hz...I guess I misbehaved in the last log I posted coz I went windowed>FS at the very beginning of the movie w/o seeking back, which I've been used to do since forever to avoid synchronized judder issues. I might as well set PotPlayer to go FS right away, but both PotP/KMP freeze to death if I go windowed>FS before the picture showed up...if there's a chance you could fix this problem, this would be much appreciated :)

[djsolidsnake86]

What kind of slowdown? Your setup should be doing fine with default setting. If not, start playing around with the options. I suggest in this order: Flushings, timing, # backbuffer, and the two checkboxes.

yes, it hangs with 90% of files!