Sony Vegas 13 CPU vs GPU Rendering

[Steemax]

Hello, 

 

I've used final cut pro for a couple years and only recently switched to Vegas Pro 13. I've always rendered the final clips using my CPU with final cut pro and even on Vegas for a bit until I discovered the option to render using "CUDA".

 

Holy smokes! Rendering using the GPU is at least 10x faster! The same 10min clips that previously took 12min to render out with a 1700x now renders in 2 min using my GPU. 

 

I was always under the impression we always use the CPU which is why it's always a benchmark standard. 

Just wanted to know if there is some kind of downside to rendering with the GPU that i'm missing? Such as quality loss etc. 

 

Thanks!

[themctipers]

iirc GPU is less accurate than CPU or something and has worse compression or something. 

 

doesnt really matter when it's a home situation, use GPU.

 

CPU rendering is a benchmark standard because rendering is quite parallel and so it scales well usually, and because it stresses out the CPU quite well too. 

[mariushm]

If you want QUALITY, you go with CPU.

If you want SPEED and just decent quality, you go with GPU.

 

With CPU encoding, the video codec can really work hard on analyzing each frame, detecting how things move around , detecting differences between series of frames, all in the effort to squeeze as much quality as possible within one second, and within the bitrate you give the codec to work with.

You can choose to let the codec do a lot of motion search, a lot of analysis, search how the current frame is different from the previous n frames in order to compress it better, and lots of other things.

 

A hardware encoder on the other hand is much more restrictive, it's made with a limited number of transistors inside the silicon chip on the video card, and the hardware encoder can only keep a certain number of frames in memory before having to produce some result, and there's a lot of other limitations. 

For example, while with the CPU the video codec could look back on let's say the previous 6 frames to determine differences between current frame and those six, to retain more detail in the same amount of data, the hardware encoder may only be able to keep in memory the last 3-4 frames it worked on, so it would only be able to look for differences in the previous 3 frames.

Or, let's say the video codec tries to determine how much a 8 by 8 pixel square of the image  moved and in which direction, compared to previous frames ... with the cpu, the video codec may look up to 32 pixels in either direction in order to detect the motion, but a hardware encoder, because it's limited by the number of transistors in it, may only look up to 8-16 pixels in either direction.  This could mean that when you have some big motion on the screen (camera panning quickly, or some explosion), the hardware video encoder may not be able to optimally compress that series of frames and you may get lower quality.

 

So basically, the hardware video encoder is much faster because it takes shortcuts, it uses more basic, more rough encoding techniques compared to the software video codec which uses the processor - for the same amount of disk space, the software video codec will produce much better results any time.

 

Sometimes though, you don't really care about the absolute best quality, or you're trying to render a video that doesn't really need that highest quality.