NVIDIA GPU - Unlimited streams RTX Models

[pir8radio 1292]

You want the NVDEC chip and you need just a little bit of CUDA :-)

 

 

Come on sounds like you are making me a mixed drink lol..      So are you saying more chips the better,   CUDA doesn't matter so much?

[softworkz 3326]

Come on sounds like you are making me a mixed drink lol..      

 

 

I actually could - I like drinks done well...

 

You are you saying more chips the better,   CUDA doesn't matter so much?

 

Not necessarily, only when comparing among the same generation. CUDA comes into play for scaling, deinterlacing and color conversion only. Those are not overly challenging tasks for these kinds of GPUs.

[pir8radio 1292]

I actually could - I like drinks done well...

 

 

Not necessarily, only when comparing among the same generation. CUDA comes into play for scaling, deinterlacing and color conversion only. Those are not overly challenging tasks for these kinds of GPUs.

 

Good info,  I dont know much about video...   Gotta get you to write something up about hardware and video as it pertains to emby...    

[Guest asrequested]

Yeah, decoding/encoding and image processing/changing, are different animals. That's why burning in subs is so crappy. That's a very high degree of image processing. A straight decode to encode is much less work. Mr. Softworkz is a bit of a perfectionist (a very good thing), the 10bit to 8bit conversion that has just been added to the beta cycle is a very big deal. 

 

So the question of which is better, is not straight forward. Use case is always a major roll.

[softworkz 3326]

 A straight decode to encode is much less work. 

 

Slight correction: It's just the opposite: Rendering text onto a video frame or overlaying an image (graphical subs) is a much easier task than encoding or decoding.

Encoding for example, needs to perform motion analysis of video content and identify regions that are being moved uniformly from one frame to another so that they can be represented by motion vectors referencing an earlier frame instead of providing new image pixels for each frame (and that was just a very simplistic - one-out-of-dozens of things that a sophisticated video encoder does. A decoder needs to translate back all that information into a sequence of images.

 

Overlaying and blending one image with another image is not much more that a simple formula that needs to be calculated once for each pixel.

 

 

The reason why we are often experiencing subtitle-burn-in to significantly slow down transcoding is that we can't do it using hardware processing yet.

So, we need to copy all video data from hw memory back to system memory, perform the overlay using CPU processing and transfer it back to the gpu later for encoding.

That's what causes the extreme slowdown.

 

Doing subtitle overlay and burn-in using hardware processing is one of the next goals.

[Guest asrequested]

Slight correction: It's just the opposite: Rendering text onto a video frame or overlaying an image (graphical subs) is a much easier task than encoding or decoding.

 

The reason why we are often experiencing subtitle-burn-in to significantly slow down transcoding is that we can't do it using hardware processing yet.

So, we need to copy all video data from hw memory back to system memory, perform the overlay using CPU processing and transfer it back to the gpu later for encoding.

That's what causes the extreme slowdown.

 

That's not a slight correction lol.... I'd forgotten about the requirement for software rendering for that (alcohol is in effect) I use my CPU for all transcoding, so things like that fall off my radar