ffmpeg running with Nvidia HW Acceleration?

[lightsout 144]

Sorry for all the threads lately just trying to get things dialed in.

 

I am doing some testing with my desktop to see what the most worthwhile upgrade to my server will be.

Desktop is running an AMD 2600x and GTX 1060 6GB.

 

I have all the HW acceleration checked, so it uses the GPU as much as possible. I notice when I play a TV recording,

a 1080i MPEG2 in a .TS, that ffmpeg is running and there is about a 25% load on the CPU.

 

I was expecting a greater offload to the gpu without much going on with the CPU. Is that just the way it is with MPEG2?

Is it normal for ffmpeg to need to run in software while the nvidia decoder does its thing?

 

I guess if this is the best it gets I may need to consider re-encoding the stuff I plan to keep.

 

[Luke 37128]

Sometimes yes. We'd have to look at an example in order to provide more information. Thanks.

[lightsout 144]

My apologies.

Sometimes yes. We'd have to look at an example in order to provide more information. Thanks.

I will PM the main log.

Edited April 2, 2019 by lightsout

[Luke 37128]

@@softworkz

[softworkz 3341]

I have all the HW acceleration checked, so it uses the GPU as much as possible. I notice when I play a TV recording,

a 1080i MPEG2 in a .TS, that ffmpeg is running and there is about a 25% load on the CPU.

 

I was expecting a greater offload to the gpu without much going on with the CPU. Is that just the way it is with MPEG2?

Is it normal for ffmpeg to need to run in software while the nvidia decoder does its thing?

 

In your case it is required to deinterlace the video, and deinterlacing is done by the CPU. That's the load you are seeing.

 

But transcoding is still running 10x times faster than realtime.

You have throttling enabled, so after an initial amount is encoded, you will only see that CPU load in small intervals.

That looks all good to me.

 

BTW: We're working on getting the deinterlacing done by the GPU as well in the near future!

[lightsout 144]

In your case it is required to deinterlace the video, and deinterlacing is done by the CPU. That's the load you are seeing.

 

But transcoding is still running 10x times faster than realtime.

You have throttling enabled, so after an initial amount is encoded, you will only see that CPU load in small intervals.

That looks all good to me.

 

BTW: We're working on getting the deinterlacing done by the GPU as well in the near future!

Ah that makes sense, great to know things are working correctly, and that is exciting to hear that this may be offloaded to the GPU soon, another reason for me to add a GPU to my server.

 

A quick question since I have you here, any idea if you can use an nvidia gpu and quicksync? Was wondering if the gpu was loaded and it fell back on the cpu, would it be able to also use HW acceleration or would it have to be software?

 

This question will largely determine my choice of adding a 4770 to my server or going amd with an exisiting motherboard with a 1700. The AMD has 4 more cores but not HW acceleration. But if that is a moot point with an Nvidia GPU inthe mix then I will just go AMD.

[softworkz 3341]

I think you should be pretty fine with your Nvidia GPU!

[lightsout 144]

I think you should be pretty fine with your Nvidia GPU!

Alright thanks.

[lightsout 144]

I think you should be pretty fine with your Nvidia GPU!

Do you know if HW transcoding with a GPU and Intel quicksync is an option?

[Luke 37128]

We do support quicksync, if that's what you're asking. Thanks.

[lightsout 144]

We do support quicksync, if that's what you're asking. Thanks.

What I meant was can NVENC and Quicksync be used together? Say the GPU is fully loaded will it offload to quicksync or just to the cpu in general?

[Luke 37128]

That is possible but Emby does not support that yet. It's coming though. Right now it will just auto-switch to CPU when the GPU is full.

[lightsout 144]

That is possible but Emby does not support that yet. It's coming though. Right now it will just auto-switch to CPU when the GPU is full.

Perfect, thanks for the quick response.