madVR - high quality video renderer (GPU assisted)

[6233638]

Turns out when using NNEDI3, my video is delayed about 100ms, so I add that delay to my AVR for audio so everything is in sync. But this causes out-of-sync audio for content upscaled using Jinc3 (because for Jinc3, I don't actually need to add any delay to my AVR).
Any ideas, how to solve this so I don't have to keep changing lip sync?The more I try to use the processing of GPU for lower resolution, the worse it gets. It's getting very tricky to balance the number of buffers that helps with not dropping frames and getting the lip sync to align across the different profiles... SIGH.....

Do you have "delay playback start until render queue is full" and "delay playback start after seeking, too" enabled?

I get very variable audio sync without having them selected.

[huhn]

Do you have "delay playback start until render queue is full" and "delay playback start after seeking, too" enabled?

I get very variable audio sync without having them selected.

i hear about sync issue with madVR quite a lot. if you are able to reproduce these issue think about a bug report.
http://bugs.madshi.net/view_all_bug_page.php?project_id=4&dummy=bla/

i like the dummy=bla part X-)

[bcec]

Do you have "delay playback start until render queue is full" and "delay playback start after seeking, too" enabled?

I get very variable audio sync without having them selected.

Thanks for the reply.. Yeah I tried that, but doing that actually moves the lip sync difference between NNEDI3 and non-NNEDI3 further away (0ms for Jinc and 150ms for NNEDI3)..

[THX-UltraII]

Guys,

I own a passive 3D setup by Infitec. The Infitec system uses 2 1080p projectors, a splitter and filters in the projector (and of course glasses that correspond with the filters in the projector). It is undoable to get a good calibration from the system so what I want is to use the CMS 3DLUT solution. I also use this for 2D. With this 2D setup I can load a pre-calibrated .3dlut file in madVR and use it in MPC-HC. This way I have a 100% perfect calibration for 2D. '

Enter 3D........ what I want is to play 3D movies with madVR as renderer and be able to use a .3dlut file. The problem is that MPC-HC (or any other non-commercial software besides PDVD and TMT) does not support frame-packed 3D .iso. So I was wondering if there is a way to rip my 3D .iso's to a format that is supported by MPC-HC wihout losing ANY resolution/quality. Am I correct that this can be done by playing a full top-bottom? Or maybe there is another way to play a 3D iso + madVR.

[bcec]

i hear about sync issue with madVR quite a lot. if you are able to reproduce these issue think about a bug report.
http://bugs.madshi.net/view_all_bug_page.php?project_id=4&dummy=bla/

i like the dummy=bla part X-)

Thanks for the link. Done, http://bugs.madshi.net/view.php?id=252

[braddock]

Is anyone using a GTX 960 with madvr yet? I am currently using a 750ti in my htpc and am wondering if the 960 is worth the upgrade.

[bcec]

Is anyone using a GTX 960 with madvr yet? I am currently using a 750ti in my htpc and am wondering if the 960 is worth the upgrade.

I upgraded from a gtx 650 to 960. Happy so far.

[shaolin95]

I upgraded from a gtx 650 to 960. Happy so far.

But he has the 750ti which has VP6..isnt that worth something for madvr or not really used?

[a8213711]

I tested it with MPC-BE and these display modes: 1080p23, 1080p24, 1080p50, 1080p59, 1080p60

the file switches to 50 hz with madVR IVTC the same with deinterlacing.

the file is PsF 2:2 so madVR IVTC is the better choice.

Thank you.
I tried "disable automatic source type detection" on film/video/off, and always get 59.94Hz, but if I use ctrl+alt+shift+t it loops between 50 (Video) and 25 (Film). Also I have "automatically activate deinterlacing when needed" checked with "if in doubt, deactivate".
Really weird.

And where do you get that PsF 2:2?

[huhn]

Thank you.
I tried "disable automatic source type detection" on film/video/off, and always get 59.94Hz, but if I use ctrl+alt+shift+t it loops between 50 (Video) and 25 (Film). Also I have "automatically activate deinterlacing when needed" checked with "if in doubt, deactivate".
Really weird.

you could try to completely reset madVR.
do you maybe use profiles i don't get why madVR does what it does in your case.
And where do you get that PsF 2:2?

by simply looking at the frames.

[braddock]

But he has the 750ti which has VP6..isnt that worth something for madvr or not really used?

I guess I should've said I was wondering how much more NNEDI3 performance I could get by upgrading to a gtx 960.

[jkauff]

I guess I should've said I was wondering how much more NNEDI3 performance I could get by upgrading to a gtx 960.
I have an Asus gtx 770 that I overclocked from Day 1 using the Asus OC utility. Trying to cut my power consumption, I got rid of the OC.

I was shocked to find that I now have much more NNEDI3 headroom. I can use much higher neuron settings than before without dropping frames.

I don't know if you overclock, but I thought I'd mention it.

[Asmodian]

But he has the 750ti which has VP6..isnt that worth something for madvr or not really used?

VP6 is for hardware decoding not rendering so if you are using hardware decoding and VP5 was too slow for real-time playback VP6 could help, but not otherwise.

[braddock]

I have an Asus gtx 770 that I overclocked from Day 1 using the Asus OC utility. Trying to cut my power consumption, I got rid of the OC.

I was shocked to find that I now have much more NNEDI3 headroom. I can use much higher neuron settings than before without dropping frames.

I don't know if you overclock, but I thought I'd mention it.

I do not overclock the 750ti. Are you saying you get better NNEDI3 performance with stock clocks? How is that possible? Was the overclock unstable perhaps? Interesting.

[cyberbeing]

How is that possible? Was the overclock unstable perhaps? Interesting.

Well there are only few possibilities:

*Overclock involved raising the voltage above stock, resulting in the GPU using lower clocks under heavy load because temperature or TDP limits were being reached quicker. On recent NVIDIA GPUs, raising the voltage above stock can often have a negative effect on achievable Boost clock under heavy loads, resulting in reduced performance.

*Overclock was unstable. Either that was the cause alone, or he experience a TDR without noticing which would lock his GPU at idle clocks until a reboot.

*GPU BIOS has a bug. I remember Galaxy released a BIOS update for their GTX 770, for a "less than expected performance" issue. While you should only use BIOS updates for your specific vendor GPU model, you can check places like TechPowerUp's VGA BIOS Database (http://www.techpowerup.com/vgabios/) to see if there is something more recent. GPU-Z can be used to see your current BIOS version number. Just remember it's in hexadecimal, so the range of values is Older|0123456789ABCDEF|Newer.

*The Asus OC utility could be bugged and causing issues. This has been known to happen with some vendor NVIDIA overclocking applications. If this was the issue, uninstalling it and using a known reliable application such as MSI Afterburner or simply just nvidiaInspector may resolve it.

[TheRyuu]

Believe I found a bug in madvr. It does not appear to be releasing its system timer (tick rate) hold on unload. So for example:

In mpc-hc open a video file which uses the madvr renderer.
After it starts playing go to file -> close.
Keep mpc-hc open and measure system timer rate with your favorite measuring program such as clockres from sysinternals.

Expected result is that it should not still be increasing the tick rate (i.e. 15.625ms).
Actual result is that it is still increasing the tick rate (i.e. 0.500ms).

Repeat the same test with EVR-CP and the expected result will match the actual case (15.625ms). I can only guess but there's probably a missing call to NtSetTimerResolution with the SetResolution flag set to FALSE on unload.

[shaolin95]

VP6 is for hardware decoding not rendering so if you are using hardware decoding and VP5 was too slow for real-time playback VP6 could help, but not otherwise.

mmm So I guess I need to understand better how they differ. In the case of madvr does VP6 help at all or how do I make it count since I won a 750ti myself. Or is there a tradeoff also by using VP6 in IQ?
Thanks!

[huhn]

mmm So I guess I need to understand better how they differ. In the case of madvr does VP6 help at all or how do I make it count since I won a 750ti myself. Or is there a tradeoff also by using VP6 in IQ?
Thanks!

it's just a faster decoder that's it.
the result of each decoder is bit perfect so there is no difference between lav software or DXVA.

[Asmodian]

mmm So I guess I need to understand better how they differ. In the case of madvr does VP6 help at all or how do I make it count since I won a 750ti myself. Or is there a tradeoff also by using VP6 in IQ?
Thanks!

VP6 is a little hardware decoding chip; it cannot do anything but decode the video. The other option is to use the CPU to do the decoding. The only resources shared by VP6 and the GPU core is bandwidth.

madVR will not use the VP6 chip itself at all, that is a job for LAV. Using VP6 (CUDA or DXVA2 hardware decoding) will not affect madVR unless you are bandwidth limited in which case it will slow madVR down. An exception to this is deinterlacing and DXVA2 resizing in madVR as both of these will involve the VP6 chip.

Optimal performance for madVR would be to use software decoding in LAV, this will leave the VP6 chip idle and allow madVR to use all the GPU resources without sharing. The benefit of VP6 for playing video is lower power usage because it can decode the video using less power than the CPU. As huhn said, the IQ is the same either way.

[kalston]

Believe I found a bug in madvr. It does not appear to be releasing its system timer (tick rate) hold on unload. So for example:

In mpc-hc open a video file which uses the madvr renderer.
After it starts playing go to file -> close.
Keep mpc-hc open and measure system timer rate with your favorite measuring program such as clockres from sysinternals.

Expected result is that it should not still be increasing the tick rate (i.e. 15.625ms).
Actual result is that it is still increasing the tick rate (i.e. 0.500ms).

Repeat the same test with EVR-CP and the expected result will match the actual case (15.625ms). I can only guess but there's probably a missing call to NtSetTimerResolution with the SetResolution flag set to FALSE on unload.

What if you close down mpc-hc? Doesn't that reset it? I use a constant 0.5ms timer on my machine but last I checked after shutting down the playback software the timer did reset to whatever it was before playback started. Wouldn't say it's a big deal unless you're on a laptop and for some reason you insist on keeping the app open even when not playing anything.

[TheRyuu]

What if you close down mpc-hc? Doesn't that reset it? I use a constant 0.5ms timer on my machine but last I checked after shutting down the playback software the timer did reset to whatever it was before playback started. Wouldn't say it's a big deal unless you're on a laptop and for some reason you insist on keeping the app open even when not playing anything.

Yes closing the application will release its "hold" on whatever timer it requested.

However, it should be a trivial fix.

[THX-UltraII]

Can madVR handle MVC MKV 3D files that are made with Makemkv?

(I know that madVR is a renderer and not a video decoder but IF the internal MPC-HC filters and/or LAV filters support MVC MKV 3D files of course madVR have to be able to handle it too)

[Eluent]

Enabling the nnedi3 in image doubling doesn't do anything anymore.

(I know that cause if I use 256 neurons on doubling and quadrupling my GTX 460 should choke and no difference on GPU-Z, no dropped frames, I don't use DXVA2, I know it used to invalidate nnedi3 doubling).
It used to be different.
In the meantime I used DXVA2 upscaling (I found it to be of excellent quality), updated many pretty much everything, so unfortunately I can't say it happened after this or that update.

I'm on latest mpc-hc, latest madvr, geforce 344.80 (that's 344.75 with correction for GTX 460 and similar cards maybe).

Anybody has an idea ?

[huhn]

Enabling the nnedi3 in image doubling doesn't do anything anymore.

(I know that cause if I use 256 neurons on doubling and quadrupling my GTX 460 should choke and no difference on GPU-Z, no dropped frames, I don't use DXVA2, I know it used to invalidate nnedi3 doubling).
It used to be different.
In the meantime I used DXVA2 upscaling (I found it to be of excellent quality), updated many pretty much everything, so unfortunately I can't say it happened after this or that update.

I'm on latest mpc-hc, latest madvr, geforce 344.80 (that's 344.75 with correction for GTX 460 and similar cards maybe).

Anybody has an idea ?
are you sure the video you tested needs scaling?
like playing a 1080p file on a 1080p screen can't trigger image doubling.

[EmuAGR]

For anybody wondering, if you want subtitles from XySubFilter to show up in your madVR screenshots with MPC-HC there are a couple ways to do so:

A) Save the screenshot when the video is viewed at Original Video Size prior to resizing (100% Zoom, except on anamorphic video).

B) Save the screenshot at any Zoom level or fullscreen, when the "Render to Original Video Size" option is enabled in XySubFilter (note that this is a lower-quality option similar in appearance to VSFilter.dll RGB output)

This limitation exists because madVR outputs screenshots at original video size, prior to resizing the video frame. Without doing A) or B) the subtitle will only be alpha-blended into the video frame after resizing. Removing this limitation is something on madshi's to-do list, but no timeline that I'm aware of. For the time being you'll need to make do with the above, if you want subtitles in your screenshots with XySubFilter.

Was this implemented/fixed already? I'm having a lot of issues with this lately, and not only with screenshots.

[cyberbeing]

Was this implemented/fixed already?
There is one additional step missing regarding subtitles in screenshots with XySubFilter, that you also need to have the "optimize subtitle quality for performance instead of quality" option disabled. At the time of that post, that option didn't exist and the disabled behavior was default. I've now edited it for clarity.

madshi has already promised that in a future version of madVR he'll allow for screenshot to be taken later in the render chain (potentially post-scale), which would resolve this issue without the need for any workarounds. Keep watching madVR changelogs for future versions, and I'll sure you'll notice it when it's finally implemented.

I'm having a lot of issues with this lately, and not only with screenshots.
You would need to explain what type of other issues you are having, especially if they don't involve subtitles in screenshots which was the focus of the post you quoted.

[Eluent]

are you sure the video you tested needs scaling?
like playing a 1080p file on a 1080p screen can't trigger image doubling.

Of course. I'm not an expert but not that noobish either :)
Anyways, fixed it by driver rollback.

[ZipDaemon]

Hi all, on Windows 7 systems when the desktop depth is 16bpp I received the dreaded 8876086c error code, does anyone know a possible work around? I tried many settings but maybe have missed something :)

http://i.cubeupload.com/t/jTLJ9T.png (http://i.cubeupload.com/jTLJ9T.png)

[vivan]

Hi all, on Windows 7 systems when the desktop depth is 16bpp I received the dreaded 8876086c error code, does anyone know a possible work around? I tried many settings but maybe have missed something :)Switch back to 32bpp? It even disables desktop composition, so I'm not suprised that it breaks other stuff.

[ZipDaemon]

Switch back to 32bpp? It even disables desktop composition, so I'm not suprised that it breaks other stuff.

Hmmm, in this case the monitor expects 16bpp and madvr is the best tool I found to display video without color banding and other serious artifacts in the picture.

Sometimes very infrequently it works but it's impossible to predict when it will play nicely and I had zero success rate on three different systems with different owners so this error is really above my skill level :)

[videonerd]

Hi guys, I have the following 2 different scenarios where I need your guidance on where the bottleneck is and what can I do to improve the situation (based on madvr's stats):

scenario 1:
decoder queue 1-6/48
upload queue 1-2/24
render queue 1-2/24
present queue 0-1/8

scenario 2:
decoder queue 44-48/48
upload queue 21-24/24
render queue 0-2/24
present queue 0-1/8

Thank you!

[cyberbeing]

Hmmm, in this case the monitor expects 16bpp

What kind of monitor is this? I think you are mistaking something, unless your monitor is over 20 years old. Modern display and GPUs from at least the past couple decades have supported 8bit color (32bpp) for the Windows desktop.

[huhn]

Hi guys, I have the following 2 different scenarios where I need your guidance on where the bottleneck is and what can I do to improve the situation (based on madvr's stats):

scenario 1:
decoder queue 1-6/48
upload queue 1-2/24
render queue 1-2/24
present queue 0-1/8

scenario 2:
decoder queue 44-48/48
upload queue 21-24/24
render queue 0-2/24
present queue 0-1/8

Thank you!
in first decoding and in second the GPU.
not a lot to say without knowing your hardware/type of file in question. i could just guess.

[videonerd]

in first decoding and in second the GPU.
not a lot to say without knowing your hardware/type of file in question. i could just guess.

Thanks Huhn. This is a 4770/gtx980 playing at 16x playrate for (1) 4k and (2) 1080p60 content.

Since I'm using software decoding, would it be correct to assume (1) cpu not powerful enough (2) gpu not powerful enough?

Thank you.

[huhn]

Thanks Huhn. This is a 4770/gtx980 playing at 16x playrate for (1) 4k and (2) 1080p60 content.

Since I'm using software decoding, would it be correct to assume (1) cpu not powerful enough (2) gpu not powerful enough?

Thank you.

yes and there is a very high chance the GPU can't handle (1) too.

[THX-UltraII]

madshi,
Is it possible to use madVR as a video processing filter? I'm asking this because when using stereoscopic player (for 3D) madVR is listed as a video processing filter, but when I selected it it didn't work... I have the idea that it would not work, but since it was listed... I wanted to use it for YCbCr->RGB conversion and 3DLUT correction, because the native one crops BTB and WTW, so, for now, I have to use ffdshow for that.

By the way, the first post seems to not be correct... it looks like some of the items are outside the [ code ] area...Did you ever managed to get madVR as video processor work with Stereoscopic Player? I m also looking for a way to use a .3DLUT with Stereoscopic Player (or ideally PowerDVD but if Stereoscopic Player can do it it will be fine by me)

EDIT: I indeed just found out that madVR is available to use in the decoder options of Stereoscopic Player for:

- MPEG-2 Video Decoder
- MPEG-4 AVC/H.264 Video Decoder
- MPEG-4 MVC Video Decoder
- VC-1 Video Decoder
- But most important maybe: you can select an external ''Video Processor'' in Stereoscopic Player where madVR is listed as compatible. The only thing to find out now is if 1. it really works and 2. the loaded .3dlut file in madVR is part of the 'video processor'.

EDIT2: Just downloaded Stereoscopic Player and selected madVR as Video Processor (it IS indeed listed in the list when I select ''show only compatible filters'' in the stereoscopic player) but when playing a movie madVR is not listed in the filters that are beeing used.

Madshi, could you chime in on this maybe?

[huhn]

see here

http://forum.doom9.org/showpost.php?p=1704663&postcount=27954
Not at this time.

it's not rare that a device changes it behavior in 3D mode and a completely new calibration is needed.

the current version of madVr doesn't support frame packed 3d anyway. you can try to create a side by side version this may or may not work.

there is simple little to no love for 3d.

[resides]

Been using madVR + PotPlayer + SVP on my HTPC and loving it... so a huge thanks for that.

Now for the (noob) question... is there an "updated" guide that goes through all the available options and settings? The few I've read seem to be old (my apologies if I'm wrong)... but case in point, I'm wondering what to choose in chroma upscaling? Nearest Neighbor seems to be the best setting, but all the guides I've come across use Jinc with 3 or 4 taps. I'm wondering if there is a guide that explains the differences between Nearest Neighbor and Jinc.

I know quality or "best" is subjective at best... so I'm not going to be asking what are the best settings... but I'd love a guide that explains all the terminology and options so I can have a better understanding of what I'm choosing and why.

PS - I've been using Nearest Neighbor for everything. Regarding h/w I'm having an NVIDIA gpu, running on an i7 with 16GB RAM. Usage is mainly as a HTPC, but specifically running PotPlayer + madVR + SVP for the soap opera effect.

[Asmodian]

Nearest Neighbor is the worst scaler, it simply uses the value of the closest pixel without any interpolation. Jinc is a newish high quality method similar to lanczos or spline, if you want to understand how it works I believe it is the EWA Lanczos Sharp (http://www.imagemagick.org/discourse-server/viewtopic.php?t=21695) method in Image Magic. Nicolas Robidoux has been helpful to madshi on resize methods (from what I can see :o) and active in this thread.

I have attempted to write an option guide, a draft is somewhere in my documents folder, but I never finished it. MadVR's options seems stable now so maybe I will try to finish it.

[hannes69]

I have a question concerning the Display Switcher: I have made four custom resolutions by using CRU Utility to play videos with fps of 23.976, 24.000, 25.000 and 29.970. Windows is silly so there is no possibility to distinguish between the 23.976 and 24.000 preset (all of the four presets are doubled corresponding to 47.952, 48.000, 50.000 and 59.940 Hz refresh rate). The both presets in the 24p region (47.952 and 48.000) lead to ONE refresh rate named 48 in Windows when creating these two in CRU Utility. So my workaround was to create a 72.000 Hz preset which replaces the 48.000 setting. So i have refresh rates of 47.952, 50.000, 59.940 and 72.000 Hz. I don´t want to use Smoothmotion and it would be nice if the Displayswitcher could handle these settings because there are videos with framerates of 24.000 and 23.976. If entering all 4 presets in the Displayswitcher (in my case 720p48, 720p50, 720p60, 720p72) the 720p72 setting is ignored. Of course it is, because 720p48 and 720p72 both lead to a framerate of 24 (or 23.976).
What would help me is the implementation of this logic (C style):

if (framerate == 23.976fps) display mode = 720p48;
if (framerate == 24.000fps) display mode = 720p72;

Of course there are more movies with a framerate of 23.976 and it doesn´t hurt me that much to use only the 720p48, 720p50 and 720p60 presets in the display switcher and in the rather rare case of a 24.000fps movie entering the 720p72 setting and removing the 720p48 setting (or disable the display switcher and choose the refresh rate within Windows), but from the perspective of using the PERFECT renderer (yes, madvr is ;) ) this definitely should be possible.
Or is there a possible workaround I missed by using profiles?

Little offtopic question but is has to do with the problem mentioned:
My projector can handle e.g 25Hz, 50Hz and 75Hz. It doesn´t convert to a fixed refresh rate internally, I can hear the color wheel spinning faster the higher the refresh rate is. What would be the "best" setting from a theoretical point of view (not the "choose what looks best to you - view") concerning e.g. motion blur and the rainbow effect (it is a DLP projector)?

BTW time to convince madshi to accept donations so he can focus more on madvr, earning the money with this product - no more need to focus on other things - what could be more important like the perfect video renderer :cool:
Hannes.

[Andy o]

Just a heads up for those using AMD cards. I'd been getting hard stutters for the past few days, the decoder queue dropped and all the queues down the lane followed, then both a few frame drops and repeats. Banging my head, changing madVR, ReClock and LAV settings, until yesterday I decided to time the interval of the stutters. Turned out to be every 15 minutes to the second. Apparently the current version of raptr (AMD Gaming Evolved) is doing something that causes that stutter every 15 minutes.

I got a 7790 with the latest WHQL 14.12 drivers on Win 8.1 64.

[baii]

The default setting usually is good "enough". SVP make simple video much more demanding as most video fps is doubled and gpu acceleration can take away part of the gpu power.

Sent from my 306SH

[resides]

@Asmodian - Thanks I'll read up on it and also play around and see what looks best.

Also, wondering if using the GPU (DXVA2) is a good option or not? I have an NVIDIA GPU and have the option of selecting processing to be done by DXVA2 in image upscaling and image downscaling, but the option is greyed out in chroma upscaling so I went with Jinc, 8 taps, activate anti-ringing filter checked.

I'm wondering if DXVA2 is a better option or should I test more with Jinc for image upscaling/downscaling?

Thanks for answering my noob questions... and I look forward to the guide.

[Asmodian]

DXVA2 is probably similar to lanczos in terms of quality but is done on the GPU's hardware, it is different on AMD, Nvidia, or Intel and maybe even between different GPUs. I prefer Jinc3 (or NNEDI3) with my Titans.

8-Taps is too many! It might be a tiny bit sharper but if you want sharpness Lanczos 4 taps would be a better option. Stick to 3 or 4 taps (if your eyes really tell you more is better of course you can use them but understand (IMO) the maximum quality with fewest artifacts option is Jinc3). Of course this is debatable and NNEDI3 is an entirely different creature.

[vivan]

The problem with DXVA-scaling is that it drops precision to 8-bit, which results in banding and/or high size of dithering grain (it forces madVR to dither before scaling). IMO, it's useful only on low-power laptops.

[StinDaWg]

DXVA2 scaling on AMD is terrible. The quality is similar to bilinear.

[BetA13]

did/can someone with madvr please test if there are still artefacts when using NNEDI3 in chroma upscaling..

that is with the new nVIDIA GeForce 347.52 WHQL Driver..

if someone could test it, that would be very nice...

greetz BetA

[sheppaul]

DXVA2 scaling on AMD is terrible. The quality is similar to bilinear.

Nope, it depends on your graphic card. DxVA2 scaling is much better than bilinear in my case (HD7850). It looks similar to bicubic 1.0 in my eyes.

[Asmodian]

did/can someone with madvr please test if there are still artefacts when using NNEDI3 in chroma upscaling..

if someone could test it, that would be very nice...

greetz BetA

What artifacts were present? There was a bug with NNEDI3 that madshi has fixed but it wasn't specific to NNEDI3 chroma upscaling as far as I am aware.

[BetA13]

What artifacts were present? There was a bug with NNEDI3 that madshi has fixed but it wasn't specific to NNEDI3 chroma upscaling as far as I am aware.

hy there,

i meant this one:
http://forum.doom9.org/showpost.php?p=1703366&postcount=27880

http://forum.doom9.org/showpost.php?p=1704001&postcount=27912

http://forum.doom9.org/showpost.php?p=1707933&postcount=28155

it was fixed when moving to older driver, wich i use now..
But id like to upgrade and well, maybe someone tried it...