madVR - high quality video renderer (GPU assisted)

[Schwartz]

Try DXVA copy-back instead of CUVID. I had a Catalyst driver that auto-selected Jinc AR instead of NNEDI3 because I was using native instead of cb.

[x7007]

Originally Posted by huhn

use drivers with openCL.

Which drivers are those ?

Originally Posted by Schwartz

Try DXVA copy-back instead of CUVID. I had a Catalyst driver that auto-selected Jinc AR instead of NNEDI3 because I was using native instead of cb.

Still stays JINC even when DXVA - Copyback is selected

[aufkrawall]

Originally Posted by x7007

Which drivers are those ?

Drivers directly from Nvidia.
Use this:
http://nvidia.custhelp.com/app/answers/detail/a_id/3698
Or wait until a new driver is released directly from Nvidia next week for Windows 10 RTM.
But DON'T use anything from Windows Update and be sure that it doesn't automatically replace your driver. I posted instructions for this above.

[x7007]

Originally Posted by aufkrawall

Drivers directly from Nvidia.
Use this:
http://nvidia.custhelp.com/app/answers/detail/a_id/3698
Or wait until a new driver is released directly from Nvidia next week for Windows 10 RTM.
But DON'T use anything from Windows Update and be sure that it doesn't automatically replace your driver. I posted instructions for this above.

353.49 only works with DXVA- Copyback right ? not Cuvid.

What more issues does it has ?

[aufkrawall]

Cuvid works, just not together with NNEDI3 OpenCL.
Why would you want to use Cuvid? I don't think the linked driver has any important issues. I wouldn't have linked it if I'd known better.

[leeperry]

Originally Posted by aufkrawall

Any idea why? I had doubling activated for >= factor 1.5 (default) and it worked just normally.

No idea, that's what I get:

It works in FSE and/or with motion pictures, but pictures in FSW just don't upscale with NNEDI256 for chroma+luma for some reason. I'm on W7SP1 in DX9 and I got Aero disabled, I do prefer FSE of course but no screenie in this mode.

Anyway, 3@0.41LQ with sxbr75 in .15 is definitely a keeper to me. J3AR is nice for chroma but sxbr125 AR would provide a nice sharpness boost w/o looking artificial(great looking on ppl with freckles for instance), I can't add AS or anything else though. Would have to try SR for chroma but it's still experimental in .15 IIRC.

Originally Posted by Ver Greeneyes

I've had success loading images in MPC-HC, though it's not ideal for saving the result since I think madVR uses slightly different logic for that. I'm pretty sure the upscaling works just fine though.

I use the default queues and it's a black frames mess for me in PotP, I wish mVR could identify still frames source filters and run in a pictures-only mode without any queue. Fair enough, will try MPC-HC then

This obscure picture viewer would provide NNEDI3 scaling but it seems quite a PITA to setup, no automation, no AR filter bleh: http://www.mydailymanga.com/pin/7768/

OK, I removed 353.54 drivers using DDU, restarted, installed 353.49 Hotfix drivers for Windows 10. Now madVR does not work at all. I get audio and no video... If I use another renderer, then I do get proper playback. I guess madVR does not work with Windows 10 353.49 drivers...

Are you saying we cannot use OpenCL NNEDI3 at all with Windows 10?????

Ha, you were right -disabling NNEDI3 did fix madVR. THING IS - 353.54 drivers DID allow the use of NNEDI3 in madVR! 353.54 > 353.49 Hotfix. It appears both are missing CUDA and OpenCL is present in 353.54 drivers!

Wait! Now I selected DXVA Copy-back and NNEDI3 suddenly works. So 353.49 Hotfix only work with NNEDI3 when DXVA Copy-back is selected, while 353.54 WHQL work with NNEDI3 when CUVID is selected. What am I losing when CUVID cannot be selected? Does image quality or performance get worse?

Also, how can I check whether OpenCL is installed?

[x7007]

Originally Posted by XMonarchY

Ha, you were right -disabling NNEDI3 did fix madVR. THING IS - 353.54 drivers DID allow the use of NNEDI3 in madVR! 353.54 > 353.49 Hotfix. It appears both are missing CUDA and OpenCL is present in 353.54 drivers!

It allowed to use but it doesn't work because the OpenCL doesn't work.. lol, NVidia are PITA.

Can you explain why not to use the CUVID ? Hardware/GPU Deinterlacing (CUVID/QS only)

Enable Adaptive HW Deinterlacing

Output mode 50/60Fps (Video)

(V) High-Quality Processing

[Schwartz]

Read the last couple of pages about CUVID and the other modes. Basically you're not losing anything but a forced high power state on your GPU. Oh and LAV hardware deinterlacing. But as I understand it that's something you can leave up to madVR anyway?

It's not only NVidia making a mess, both manufacturers are pretty spotty in their OpenCL support and performance as well as availability of NNEDI3 is always driver-dependant.

[aufkrawall]

Originally Posted by XMonarchY

What am I losing when CUVID cannot be selected? Does image quality or performance get worse?

This has already been discussed on the previous pages of this thread...

to put it short:
using cuvid improves nothing
don't use WU drivers

I'm out now, it feels like talking to a wall.

[KoD]

Originally Posted by madshi

Is this a new problem with the newer madVR builds? Or what it "always" this way?

Hi madshi, since my post last week, I discovered that, in fact, pausing twice is often not enough to resume playback. And what's worse, is that if I keep doing switches between fullscreen and windowed modes, the player ends up in a state where it can't recover anymore even after pause/unpause, playback is impossible even in windowed mode. I have to close MPC and open it again, and then playback works fine again (until I start switching between full-screen and windowed modes a few times).

As for version 0.88.20, the problem is even worse than in 0.88.19.
For one, now I have the issue even with the default settings in madVR, which don't use D3D11 and FSE, so no 10 bit per color output. I did not even try to test the D3D11 renderer this time. And instead of dropped frames, now playback simply seems to hang for 20 seconds or so, showing the same frame, while MPC is reporting to be playing, the madVR OSD shows all queues full but the frame displayed on the screen does not change, and the "rendering" time in the OSD is reporting 0.00ms. If I wait long enough (tens of seconds), playback then suddenly starts to resume.

I agree with the reports from others, rendering times when it is playing are definitely not correct - I never saw 1-2ms before.

As a side-note, the player does not become unresponsive, it's just the playback that suddenly stops on one frame. If I press the ESC key, MPC closes immediately.

Previous to trying 0.88.12, I have been using 0.87.21 and there are no issues there. I just tested again that version, and indeed, no issues. 0.87.21 does not have the option for a D3D11 renderer, so if you can let me know which is the first version that has it, I can download it and test if the issue is present there or not.

I don't think I said this before, but I am using the 64 bit version of MPC with the 64 bit version of madVR.



Later edit:
Ok, please disregard my assumptions in the previous post. In fact, there are two issues:
- the issue in my original post, about the frame dropping and queue depletion happening only when 10bit output is sent to the TV is still there in 0.88.20, just like in 0.88.19 and 0.88.12. With 0.87.21 this issue is not present, but there was no way to output 10 bit color back then.

- the "new" issue I am speaking about above, where I had the frame freeze, despite using default settings, with no 10 bit output, and everything seemingly playing, seems to be caused by something else!

MPC-HC is set to send its audio output to the optical S/PDIF output, through the "DirectSound: Digital audio (S/PDIF) (High Definition Audio)" device. But foobar2000 is also configured to send the audio to the S/PDIF using the "WASAPI (event) : Digital Audio (S/PDIF) (High Definition Audio Device)", which looks like gives it exclusive access to the audio device. The issue I described above with the playback freezing for a while, and then suddenly resuming, happens only when foobar2000 is playing music and I pause the video in MPC -> when I unpause the video in MPC I get the sudden frame freeze for a couple of seconds. And this issue is present in 0.87.21 too, and even with EVR, so it's a MPC-HC issue. If I use the "Digital Audio (S/PDIF) (High Definition Audio Device)" device in MPC, or the new Internal Audio renderer which has exclusive mode capabilities, this issue then does not occur.

[Warner306]

Originally Posted by aufkrawall

This has already been discussed on the previous pages of this thread...

to put it short:
using cuvid improves nothing
don't use WU drivers

I'm out now, it feels like talking to a wall.

The discussion in this thread may be useful regarding the benefits of DXVA2 Copy-back vs Nvidia Cuvid: http://yabb.jriver.com/interact/index.php?topic=80258.0.

[huhn]

Originally Posted by Schwartz

Read the last couple of pages about CUVID and the other modes. Basically you're not losing anything but a forced high power state on your GPU. Oh and LAV hardware deinterlacing. But as I understand it that's something you can leave up to madVR anyway?

It's not only NVidia making a mess, both manufacturers are pretty spotty in their OpenCL support and performance as well as availability of NNEDI3 is always driver-dependant.

DXVA deinterlacing is the same deinterlacer as CUVID deinterlacer. HQ is default for DXVA deinterlacing.

and to be more precise CUVID deinterlacing doesn't allow madVR to use IVTC on the interlaced video stream. so there is nothing to win only things to loose.

and different decoder doesn't have better picture quality.

and about openCL looks like nvidia has removed cl_nv_d3d9_sharing in newer drivers. i'm not 100% sure if madVR uses that extensions. if it does everything make sense.

[mogli]

When displaying a movie one frame at a time in MPC-HC one sees a future frame flashing shortly before the correct one is shown. This is new since a few madVR versions ago.

Is this a known issue?

[DragonQ]

I upgraded to the latest MadVR on both my laptop and desktop and now have the same issue with both: deinterlacing no longer activates for 1080i/25 AVC videos in MKVs. I get a "DXVA processing failed" message. 576i/25 MPEG2 videos in MKVs seem to work fine.

The laptop has Intel and nVidia GPUs, the desktop has AMD. 0.88.19 also suffers this problem but 0.88.15 doesn't, so something happened somewhere between those two versions to break this.

[aufkrawall]

Originally Posted by huhn

and about openCL looks like nvidia has removed cl_nv_d3d9_sharing in newer drivers. i'm not 100% sure if madVR uses that extensions. if it does everything make sense.

Urks. If this proves to be true, we need a non-OpenCL implementation of NNEDI3 very badly.
Edit: Or can d3d11_sharing be used?

[chros]

Originally Posted by har3inger

I have found a bug with 30 fps content with the latest .20 build. OSD reported render times are well under 2 ms (which is impossible for my hardware specs: last version reports render times ~28ms for same settings), but cranking up settings will caused dropped frames while the OSD reports <20ms render times. This is a cosmetic bug with no functional problems that I can see. Problem didn't occur in .19 and earlier. No rendering time calculation errors with 24 fps content. Don't have access to any other framerate content to test.

Windows 8.1 x64, mpc-hc x64, madvr x64, LAV filters 65.0. Image upscale jinc ar, chroma bicubic 75 ar. No image doubling.

I can confirm the OSD rendering times issue with D3D9 Old Path as well (windowed and FS) even with 24fps content. (v0.88.20)

[huhn]

Originally Posted by aufkrawall

Urks. If this proves to be true, we need a non-OpenCL implementation of NNEDI3 very badly.
Edit: Or can d3d11_sharing be used?

should be possible with this or NV/KHL d3d10 but who said they will not remove this next version?
but only dx10 or newer card should support this. the old version should work on dx9 cards at least in theory.

just removing an extension... and no they didn't added cl_khl_dx9_media_sharing