madVR - high quality video renderer (GPU assisted)

[ajp_anton]

About the (red) OSD, could you make it another color? I too am slightly color blind, and it's sometimes very difficult to see the text. It's not about size, just the color.
Maybe white text with a thin black border?

[sneaker_ger]

Madshi, what improvements/differences are we likely to see with the final FSE rendering path?

From all the hints he's been giving lately I bet we'll see DXVA support or some other kind of hardware acceleration. But since he didn't talk about the upcoming features specifically, I guess he wants to surprise us (or he doesn't want to disappoint users by promising features he might no be able to implement in the end) so we'll have to wait for the release.

[SamuriHL]

It's not DXVA support. Cause I had asked about that in another thread cause I thought he was working on that and he said no.

[nevcairiel]

I don't think there will be actual "features", it'll most likely just be smoother and overall better. :)
But lets wait a few hours!

[sneaker_ger]

You've said you're working on a version of madVR that can work with DXVA, right?
No, I didn't.

That means he didn't say that he was working on DXVA support, not that he didn't (plan to) work on DXVA.

But please don't hang me in case I'm wrong.

I feel like one of those Harry Potter fans standing in line for the midnight sale right now. Random fact: madVR turned 2 years yesterday.

[SamuriHL]

Well, I'm watching a blu-ray with 0.49 right now and man, literally THE best PQ I have on my PC. So, if there's an improvement to make things smoother, I'm all for it. :D

[SamuriHL]

That means he didn't say that he was working on DXVA support, not that he didn't (plan to) work on DXVA.

But please don't hang me in case I'm wrong.

ROFLMAO! That's, um, one way to look at it I suppose. LOL! :)

[robpdotcom]

Of course, he has to tease us with posts like this, from the CoreAVC thread:

madVR simply doesn't support DXVA (yet?).
:devil:

[SamuriHL]

And he wonders why I asked about it in the other thread? sigh. :D

[janos666]

I think he simply states that he does not entirely refuse to support DXVA ("may be in the future" - but it is funny because DXVA will be less and less important with time, as we will have faster and faster CPUs...) but I think he doesn't really care. (For example, I wouldn't use it anyway because I simply don't trust in the DXVA decoder as much as I trust in the software libavcodec.)

What about the 10 bit/color output and "VGA LUT reset/reload", by the way? :rolleyes: :)

[SamuriHL]

I'd be more interested in STREAM/CUDA decoding which is what I'm using now. Because then he has full control over the decoded frame for post processing. And none of the drawbacks of DXVA. That's what gives madVR its power. :)

[Ziddy76]

Is there any way to prevent MadVR from making my screen flash and other junk every time I go from full screen to restore? It's very annoying.

[robpdotcom]

I only want DXVA support for hardware deinterlacing. I use my HTPC as my DVR, and HW deinterlacing is the only method that works well no matter what I throw at it - which I think is due more to poor quality ATSC broadcasts than anything else.

Is there any way to prevent MadVR from making my screen flash and other junk every time I go from full screen to restore? It's very annoying.

You could try disabling the automatic fullscreen exclusive mode.

[LoRd_MuldeR]

Is there any way to prevent MadVR from making my screen flash and other junk every time I go from full screen to restore? It's very annoying.

Have you tried disabling the option to turn off Aero Glass in fullscreen mode?

[madshi]

I'm afraid I have no idea what you are talking about. I mean this OSD in the top left...

It is always the same size regardless of the window/video size (as it should be).

I want it to be exactly the same as the OSD that you would get with MPC-HC + EVR.
Oh well, personally, I prefer the madVR OSD over the MPC-HC OSD. I think mine looks better. But that's probably a matter of taste. Anyway, this is not the right time to talk about cosmetics. Let's delay this until madVR is feature complete, ok?

Well, there are two parts of firewall: HIPS and network filter. Netfilter is ok now (no packets sent by madVR). But it still triggers HIPS by trying to use some kind of network related functions. You can get trial version here (http://www.jetico.com/jpf2setup.exe). But it should be same with any other HIPS.
Thanks.

About the (red) OSD, could you make it another color? I too am slightly color blind, and it's sometimes very difficult to see the text. It's not about size, just the color.
Maybe white text with a thin black border?
The problem with white text (without a black border) is that the sky is often very bright in movies, so a white OSD would be practically invisible. A black OSD would also be practically invisible in dark scenes. There are rarely scenes which are all red, so the current OSD color is a good choice, IMHO. Of course adding a thin black border would solve things, but doing that properly is harder than it seems and I've simply more important things to do first.

From all the hints he's been giving lately I bet we'll see DXVA support or some other kind of hardware acceleration.
No, sorry. v0.50 is only about performance/smoothness in fullscreen exclusive mode.

It's not DXVA support. Cause I had asked about that in another thread cause I thought he was working on that and he said no.
That's not what I said... :p

That means he didn't say that he was working on DXVA support, not that he didn't (plan to) work on DXVA.
Correct! I'm simply refusing to comment on DXVA at the moment. DXVA support may come, or maybe not. If it comes, maybe next week, maybe in 3 years. I'm generally not commenting on future versions.

I don't think there will be actual "features", it'll most likely just be smoother and overall better. :)
Correct.

Of course, he has to tease us with posts like this, from the CoreAVC thread: [...]
:p

I think he simply states that he does not entirely refuse to support DXVA ("may be in the future" - but it is funny because DXVA will be less and less important with time, as we will have faster and faster CPUs...) but I think he doesn't really care. (For example, I wouldn't use it anyway because I simply don't trust in the DXVA decoder as much as I trust in the software libavcodec.)
I'd be more interested in STREAM/CUDA decoding which is what I'm using now. Because then he has full control over the decoded frame for post processing. And none of the drawbacks of DXVA. That's what gives madVR its power. :)
FWIW, *if* I ever implement DXVA support then it won't have any image quality drawbacks (unless the DXVA decoder hardware is broken/buggy).

Is there any way to prevent MadVR from making my screen flash and other junk every time I go from full screen to restore? It's very annoying.
The latest madVR version disables Aero by default when you switch to fullscreen, and reenables it by default when you "restore". You can modify this behaviour in the madVR settings.

FWIW, I'm considering leaving Aero on *by default* (but I'll keep the options as they currently are).

-------

Some bad news, unfortunately: On my HTPC (NVidia 9400 mainboard), either Direct3D or the NVidia drivers are not behaving as the Direct3D documentation says, resulting in problems with my final exclusive mode rendering path. Not sure how to handle this yet... :( I still hope to have a new version out later today. But I'm not sure how much of an improvement it will be over the madVR v0.49 (if any at all). Need to cross check how it behaves with ATI.

[ch1n4]

Thank you madschi,

Just take your time, there is no need to rush. It's done when you think everything behaves like you want it to behave.

Have a nice sunday.

[ikarad]

That was not my last answer. Anyway, I do plan to add a "repeated frame" counter, but I cannot technically do that with the current presentation logic. It will come in a future madVR version. Might take a couple of weeks.
.

Excuse me to disturb you but I would like to know if do you expect to add this counter soon because in 0.49 there is not still this counter?


Can I expect in one week, one month or more?

Thanks.

[sneaker_ger]

So much for that :/

[SamuriHL]

Lol! Thanks for the clarification. That's no fun about the nvidia side of the house. I hope you're able to find a clever solution.

[madshi]

Sorry guys, no new version today. Still fighting with exclusive mode on my stupid NVidia 9400. Hopefully tomorrow.

[SamuriHL]

Take your time. I know I'd rather wait and have it working well than rushed. Good luck!

[Thunderbolt8]

its alright, as long as ati is working fine :D

[pankov]

madshi,
I've just changed my video card from AMD/ATI 5750 to NVIDIA GTX 460 with the idea of using LAV CUVID decoder for interlaced content and madVR and after today's test I'm not impressed at all, so please don't make me regret my "upgrade" even more.
Just to share one more disappointment with you I'll say that the NVIDIA cards go to full 3D clocks when they are connected to more than one display :( which in my case is needed. Also the CUDA decoders (CoreAVC or LAV CUVID) again force full 3D clocks even with one display. This is so stupid from NVIDIA - how is it possible that this new "beast" needs all it's power to decode a video that my old NV9400 motherboard can decode without a problem using it's old VP2 engine.
I have some hope in MSI AfterBurner but I still haven't polished things yet.

[Qaq]

Take your time.
Agreed. Sometimes best tactic is just wait until idea will come by itself.

[nevcairiel]

Just to share one more disappointment with you I'll say that the NVIDIA cards go to full 3D clocks when they are connected to more than one display :(

Mine doesn't.

[Andy o]

Probably drivers. ATI used to do the same depending on drivers. At one point, only enabling overdrive was disabling idle clocks. I don't know how they behave now with multiple displays, but they also had this problem.

[pankov]

Mine doesn't.
Nev,
are you sure? I've been reading a lot last night and today on this topic and even NVidia says it does
http://uk.download.nvidia.com/Window...ease_Notes.pdf
see page 29
GPU Runs at a High Performance Level (full clock speeds) in
Multi-display Modes
This is a hardware limitation and not a software bug. Even when no 3D programs are
running, the driver will operate the GPU at a high performance level in order to
efficiently drive multiple displays. In the case of SLI or multi‐GPU PCs, the second GPU
will always operate with full clock speeds; again, in order to efficiently drive multiple
displays. Today, all hardware from all GPU vendors have this limitation.
What drivers do you use? (I'm on 270.51) What's the model of your card? What are you displays?
http://www.legitreviews.com/article/1461/19/
there they state that if the displays use the same resolution it will downclock but in my case with two 1920x1080 displays (23" LCD Monitor and FullHD Projector) it doesn't.

[nevcairiel]

I have a GTX570 (also 270.51), running two Dell U2410, on 1920x1200 each. I can see the GPU clock going down when just sitting on the desktop just fine (with GPU-Z). Both connected via DVI-D - maybe it helps having 2 identical screens? I don't know.

I'm planning to upgrade one of my screens to 27 or 30", that would make it break, eh. Oh well, not that i really care. Even on full clock, if there is no load on the card, it wont get noisy.

[madshi]

Ok, things are looking better now. v0.50 should be out in 1-3 hours.

[madshi]

Well, there are two parts of firewall: HIPS and network filter. Netfilter is ok now (no packets sent by madVR). But it still triggers HIPS by trying to use some kind of network related functions. You can get trial version here (http://www.jetico.com/jpf2setup.exe). But it should be same with any other HIPS.
Haven't had the time to really look into this, but I've now disabled the WinSock initialization if LAN is disabled. With a bit of luck that might solve the problem. Could you please double check with v0.50? Thanks.

[SamuriHL]

Thanks for the release, madshi! Looking forward to testing it later. :)

[DigitalLF]

will look forward to testing it :)

[sneaker_ger]

Thx for the release, but on a first test it's worse than the old mode for me. Quite a few dropped frames on 1080p60 and my HD 5850. Decoder and upload are stable but the other three drop to zero occasionaly.

[madshi]

madVR v0.50 released

http://madshi.net/madVR.zip

* added new exclusive mode rendering path
* added option to use the old exclusive mode rendering path
* fixed: settings window suddenly showed in taskbar, wasn't supposed to
* changed "disable desktop composition" default setting to "off"
* removed option "upload frames in render thread", forced to "off" now
* removed "presentation glitches" information from OSD (was always 0, anyway)
* madHcNet doesn't initialize WinSock, anymore, when LAN access is disabled
The new exclusive mode rendering path is supposed to:

(1) perform a bit better (e.g. allow more demanding scaling algorithms to be used)
(2) be more resistent against frame drops and delayed frames
(3) fix the garbled output a few people had with the old exclusive mode rendering path
(4) work without needing manual tweaking (e.g. flush options don't have any effect, anymore)

On my NVidia 9400 integrated HTPC I can now use Cubic for both chroma upsampling and luma up/downsampling without frame drops/repeats. Depending on the scaling factor I can even use Lanczos3 for luma scaling now. With madVR v0.49 I pretty much had everything set to Bilinear to get really smooth motion.

<developer rant on>
I have to say that I'm quite a bit disappointed in Direct3D. Or maybe it's the GPU drivers? Can't say for sure. According to the Direct3D documentation I *should* have been able to render and present 30 video frames in advance, without having to wait for the VSync interrupt - ever. Unfortunately in real life I can only present 3 video frames in advance. Trying to present the 4th frame blocks. With both NVidia and ATI. Sometimes presentation even blocks before the 4th frame. Sometimes presentation blocks for even more than 1 whole vsync duration (e.g. > 40ms for 25fps). Getting this all to work at all was very painful. And I'm still not sure it's all working as expected.
</developer rant off>

I've tweaked things to work reasonably well with my NVidia 9400. I think/hope that doing so should not harm more powerful GPUs. At least my ATI 6850 is happy with v0.50, too. It's possible, though, that a bit more tweaking might be necessary. In any case, please consider the new exclusive mode rendering path experimental. Positive feedback is welcome. Bug reports and other (reasonable) complaints are welcome, too.

@cyberbeing, please double check whether the 3dlut stuttering is still there.

[madshi]

Thx for the release, but on a first test it's worse than the old mode for me. Quite a few dropped frames on 1080p60 and my HD 5850. Decoder and upload are stable but the other three drop to zero occasionaly.
Too bad. Which OS? What about 1080p24 content? Does that play ok at least? With the old rendering path, 1080p60 was perfect?

[SamuriHL]

And you tested that subtitles still work? ;) I'm just teasing! :D

[sneaker_ger]

Too bad. Which OS? What about 1080p24 content? Does that play ok at least? With the old rendering path, 1080p60 was perfect?

Win 7 x64
Catalyst 11.4

I didn't get any drops on 1080p24, but backbuffer and present are running dangerously low.

With 1080p60 sample: with the old path I get about 2 dropped frames per minute, the new path about 15.
1080p60 sample:
http://rapidshare.com/files/456942508/WipEout_HD_English_1080p.mkv

[iSunrise]

Great! Thanks for v0.50 madshi, going to test it right away.

Just one question in advance:

According to the Direct3D documentation I *should* have been able to render and present 30 video frames in advance, without having to wait for the VSync interrupt - ever. Unfortunately in real life I can only present 3 video frames in advance. Trying to present the 4th frame blocks. With both NVidia and ATI. Sometimes presentation even blocks before the 4th frame. Sometimes presentation blocks for even more than 1 whole vsync duration (e.g. > 40ms for 25fps). Getting this all to work at all was very painful. And I'm still not sure it's all working as expected.
Is this somehow related to "Maximum pre-rendered frames", which can be selected (default is "3") in the NVIDIA drivers (under Manage 3D Settings)? If it is, did you try to play with those values a bit (0-8 can be selected)? Just wondering.

[Mr Alpha]

I can report that with the new exclusive rendering mode the CrossFire related issues are gone.

[Abradoks]

Haven't had the time to really look into this, but I've now disabled the WinSock initialization if LAN is disabled. With a bit of luck that might solve the problem. Could you please double check with v0.50? Thanks.
It's still the same. Quite strange it doesn't fix a problem if by "WinSock initialization" you mean WSAStartup().

What about VSFilter in MPC? It should take just a few seconds to fix.

And another two minor issues:
1) When right-clicking madVR icon in the tray first time (without preliminary left-clicking) both madVR and taskbar menus appear overlaying each other. (WinXP)
2) After closing file in MPC (but not MPC itself) madHcNet.dll remains locked by MPC.

[sneaker_ger]

Feedback on 9200M GS and 1080p60 sample:
While the decoder queue was very stable in the old path and drops very low in new path, the overall number of dropped frames has decreased significantly using the new path: from about 250 to about 100.

Same observation as on my desktop:
present queue looks dangerously low, often at 0-4/4 even for 1080p24. But no dropped frames on 1080p24 (same as old path).

[leeperry]

:thanks: for the new build!

why no more "upload frames in render thread"? I thought this option was useful? this said, I think seeking is faster now. I'll watch a bunch of movies and report back.

[nevcairiel]

Watched about half an hour of a BluRay (1080p/24) on my GTS 450, and everything is super smooth, no dropped frames at all.

Awesome work. :) I'm going to finish that disc now. :)

[SamuriHL]

I can't wait to try it after work today!

[CruNcher]

madshi could i ask which Nvidia drivers you use for developing currently ?

[cvrkuth]

No dropped frames at all, playback is smooth!
Thanks!

[Razoola]

For me the new rendering path is a bit of a disaster. The render queue is extreamly limited (1-2 or 2-3) and I get a black screen and D3D error after about 10 to 20 seconds (always). Going back to the old rendering path gives me a 7-8 render queue and no D3D error.

gtx295, 270.51 drivers win7 x64 SP1.

MadVR log at http://unibios.free.fr/madlog.rar

[CruNcher]

Dropped frames can also come from Kernel Drivers or Applications polling the Hardware,Software in to short intervals or are bad coded things like AIDA, or other sensor information software has a tendency to cause this so be sure nothing runs in the background first (lowest context switching as possible when running the Player + MadVR, or any Game for example, time critical applications in General on Windows).

[mindz]

madshi, thanks alot for all the work you have done. One question from me is, when are you planning on providing DVD support?

[bjd]

:thanks:

Appreciate all the time and effort you have put into this recently - gonna give the new build a whirl now.