MP 1.3.x dshowhelper development (3 Viewers)

[mhoogenbosch]

I'm seeing lots more dropped packages,

Sorry - I don't understand 'packages' - do you mean frames or samples ?

DXVA2 native uses less GPU resources/performance than CUDA - also CUDA runs the GPU at maximum power state (AFAIK)

Yes i mean frames i guess I've pressed ! and it shows the three lines. I'ts running for a while now, i've got 505.000 frames of which are 247 drops and 4 rep. Any idea about these values? I'm getting the impression the dropped frames mostly were at the beginning of the channel change. Now the three lines are pretty stable.

So if native is less GPU intensive, why whould one choose CUDA, or does CUDA do a better job? Because i don't see any difference.

 

[Owlsroost]

I don't see any real difference between using CUDA and DXVA2 Native either (other than DXVA2 native uses less GPU).

Some dropped frames after a channel change is not unusual, so if TV generally looks OK (no stuttering etc) then don't worry about it.

 

[mhoogenbosch]

I don't see any real difference between using CUDA and DXVA2 Native either (other than DXVA2 native uses less GPU).

Some dropped frames after a channel change is not unusual, so if TV generally looks OK (no stuttering etc) then don't worry about it.

At the LAV settings there is another setting which can't be set when using DXVA2 Native. I'ts the advanced deinterlancing, but when i've set it to enable the codec within MePo falls back to Microsoft DTV.. so didn't use it

I'm sticking with DXVA2 Native for a while and see if i hear any more complains tnx for support once again!

 

[DragonQ]

At the LAV settings there is another setting which can't be set when using DXVA2 Native. I'ts the advanced deinterlancing, but when i've set it to enable the codec within MePo falls back to Microsoft DTV.. so didn't use it

I'm sticking with DXVA2 Native for a while and see if i hear any more complains tnx for support once again!

When you use DXVA2 Native, the best deinterlacing algorithm is chosen automatically when using an nVidia GPU. So you'll get the same results when using DXVA2 Native as you would when using CUVID with "Enable Adaptive HW Deinterlacing", "50p/60p (Video)" and "High-Quality Processing" all ticked in LAV Video Configuration.

So if native is less GPU intensive, why whould one choose CUDA, or does CUDA do a better job? Because i don't see any difference.

The result is the same, except that DXVA2 Native uses less power and is a bit less flexible when it comes to what content it can handle.

For MediaPortal, I can only think of one reason to use CUVID, which is the reason I use it: when doing post-processing. Deinterlacing should always be done before resizing or applying post-processing effects, and the only way to do hardware deinterlacing before post-processing is to use CUVID with the options listed above enabled. This means that the image is decoded and deinterlaced using CUVID before ffdshow gets its hand on it for post-processing.

Currently MediaPortal's scaling algorithm in EVR is poor, so ffdshow resizing is necessary for good quality SD video (IMO).

 

[tourettes]

reason to use CUVID, which is the reason I use it: when doing post-processing. Deinterlacing should always be done before resizing or applying post-processing effects, and the only way to do hardware deinterlacing before post-processing is to use CUVID with the options listed above enabled. This means that the image is decoded and deinterlaced using CUVID before ffdshow gets its hand on it for post-processing.

Blu-ray's in menu mode have smoother clip boundaries with CUVID and DXVA2 copy back than DXVA2. This is a limitation from the codec side.

 

[mhoogenbosch]

reason to use CUVID, which is the reason I use it: when doing post-processing. Deinterlacing should always be done before resizing or applying post-processing effects, and the only way to do hardware deinterlacing before post-processing is to use CUVID with the options listed above enabled. This means that the image is decoded and deinterlaced using CUVID before ffdshow gets its hand on it for post-processing.

Blu-ray's in menu mode have smoother clip boundaries with CUVID and DXVA2 copy back than DXVA2. This is a limitation from the codec side.

Is there a difference between DXVA2 Copy Back, and DXVA2 Native?

 

[legnod]

As far as i know you the copy-back mode copies the the decompressed frames back to main memory after they have been decompressed by the graphic card's hardware decoder. Native mode presents the frames right away without copying them back. I think copy back should only be used if you have enabled post-processing!?

 

[azzuro]

Yes copy Back is for use Post-processing compared to Native

 

[mhoogenbosch]

Yes copy Back is for use Post-processing compared to Native

so just to be clear. I'm not doing any post processing. Therefor it is recommanded to use DXVA2 Native. So No CUDA and No DXVA2 Copy Back?

 

[tourettes]

Yes copy Back is for use Post-processing compared to Native

so just to be clear. I'm not doing any post processing. Therefor it is recommanded to use DXVA2 Native. So No CUDA and No DXVA2 Copy Back?

Unles you watch Blu-ray's in menu mode, then copy back or CUDA is recommended to get smooth clip transforms between titles (commercial changes, trailer changes, menu changes etc.).