Videotoolbox or CPU

[Pakorn]

I am using Handbrake to do some encoding of videos into h265. I have plenty of contradictory info on the internet : should I use CPU or Videotoolbox encoding ?
Thanks.

 

[bogdanw]

Videotoolbox, ser the video https://forums.macrumors.com/threads/mac-mini-m4.2456698/?post=33905395#post-33905395

 

[bzgnyc2]

I am using Handbrake to do some encoding of videos into h265. I have plenty of contradictory info on the internet : should I use CPU or Videotoolbox encoding ?
Thanks.

It depends on your goals -- Videotoolbox (using Apple's hardware encoder) will encode by far the fastest. Software encoding on the CPU will generate smaller files for a given level of visual quality but will take multiple times longer.

At two extremes of use cases, people almost always use Videotoolbox when they want real-time encoding while people tend to use software/CPU encoding for archival purposes (or similarly creating files that will be saved for distribution to a lot of people).

Three additional notes:
-You didn't mention your hardware but Apple Silicon and the T2 chips share a similar media engine -- the ones in the Apple Silicon of course being a newer generation that are both faster and can handle newer media types (e.g. 10-bit HEVC for HDR)
-All else being equal you will probably want to use a relatively higher quality target setting when using hardware encoding (e.g. Videotoolbox) and will need to test different encoding parameters to find the quality/output size/encoding time tradeoff you want
-The relative tradeoffs vary between generations of hardware and so you need to check your assumptions/preferred parameters after major hardware upgrades

 

[Pakorn]

It depends on your goals -- Videotoolbox (using Apple's hardware encoder) will encode by far the fastest. Software encoding on the CPU will generate smaller files for a given level of visual quality but will take multiple times longer.

At two extremes of use cases, people almost always use Videotoolbox when they want real-time encoding while people tend to use software/CPU encoding for archival purposes (or similarly creating files that will be saved for distribution to a lot of people).

Three additional notes:
-You didn't mention your hardware but Apple Silicon and the T2 chips share a similar media engine -- the ones in the Apple Silicon of course being a newer generation that are both faster and can handle newer media types (e.g. 10-bit HEVC for HDR)
-All else being equal you will probably want to use a relatively higher quality target setting when using hardware encoding (e.g. Videotoolbox) and will need to test different encoding parameters to find the quality/output size/encoding time tradeoff you want
-The relative tradeoffs vary between generations of hardware and so you need to check your assumptions/preferred parameters after major hardware upgrades

I have the Mini M4, therefore thinking whether I should encode HW or CPU...

 

[bzgnyc2]

I have the Mini M4, therefore thinking whether I should encode HW or CPU...

So great hardware for this and final choice depends on your goals and use case.

Probably best at this point is to compress/encode/etc a typical video file using software encoding with lowest quality setting you find desirable (for most people that would a setting with results that are visually indistinguishable from the originally). Record how long that takes and note the resulting file size. Then do the same using Videotoolbox (i.e. hardware encoding). How much faster was it and how much larger was the resulting file? If hardware encoded files come out 10x faster but are 1.5x larger is the size savings worth the extra time?

Also note that software encoding tends to make the rest of the system feel less responsive so that doing it during the day just may be impractical for you. Whereas if you planned to kick off the encoding to run overnight the extra speed from hardware encoding may not matter to you.

 

[bogdanw]

Why would anyone use software encoding when Apple made dedicated hardware specifically for encoding? What is better than dedicated hardware?

 

[galad]

Software encoders can create a smaller file at the same quality of an hardware encoder, or a better quality file at the same size.

Hardware encoders are quick and dirty. It all depends on what you want to get.

 

[TzunamiOSX]

CPU encoding gives a better quality at the same size.

This is H265, but H264 is nearly the same. I light up the Picture with gamma correction to better show the blocks.

 

[bogdanw]

VideoToolbox does NOT use the GPU on Apple silicon. It uses the Media Engine, dedicated cores created specifically for decoding/encoding.

 

[bzgnyc2]

VideoToolbox does NOT use the GPU on Apple silicon. It uses the Media Engine, dedicated cores created specifically for decoding/encoding.

Yes that's true but it doesn't change the statements made above. Hardware encoding is faster but has worse compression-quality trade-off than good software encoders all else being equal. The T2 chip used in Intel Macs was actually a pretty good encoder and the ones used in Apple Silicon Macs even better. Nonetheless, software-based encoders have better results with the drawback of being much slower (generating more heat, etc).

Also note AMD, Nvidia, and Intel also all have dedicated engines for decoding/encoding -- Nvidia and AMD's cores are of course on their GPU cards but they aren't using the GPU functions to do the encoding as that approach has not proven itself as effective as media-specific hardware.

P.S.galad is one of the Handbrake developers and specifically for Mac so he knows this stuff pretty well...

 

[baconcat_8964]

Why would anyone use software encoding when Apple made dedicated hardware specifically for encoding? What is better than dedicated hardware?

It depends on the demand.

Software encoding takes much longer, but generates smaller files and relatively better quality. It’s recommended for archiving things like videos, movies.

Hardware encoding uses dedicated hardware. It’s much faster, suitable for real-time demands like live streaming and video editing.

 

[bogdanw]

Apple puts dedicated hardware for video decoding/encoding in every Mac. Professionals take advantage of it in Final Cut Pro/Compressor.
Users “know better” and keep using VirtualDub and Xvid, like it’s 2005. 🤷‍♂️

 

[bzgnyc2]

Apple puts dedicated hardware for video decoding/encoding in every Mac. Professionals take advantage of it in Final Cut Pro/Compressor.
Users “know better” and keep using VirtualDub and Xvid, like it’s 2005. 🤷‍♂️

Not sure why those two programs came up when the discussion started with Handbrake which has an easy option to choose software-based or hardware encoding (the latter via Apple's Videotoolbox).

In any case, are you disagreeing that in the context of video formats like H.264/H.265 software encoding typically produces smaller files for a given level of visual quality? If so do you have data to refute others' data and expertise showing the consumer hardware encoders (e.g. Apple's, Nvidia's, etc) for these formats doesn't generate as small files as software-based ones?

 

[mjbsonic]

I am using Handbrake to do some encoding of videos into h265. I have plenty of contradictory info on the internet : should I use CPU or Videotoolbox encoding ?
Thanks.

Hardware is a more efficient way to encode on the newer M series Macs (I do encoding on both Mac and AMD/Nvidia hardware). I have noticed in the thread many have claimed poorer quality, this is rectified by going into the video tab and pushing up the constant quality slider (I typically use CQ 80). Unless your doing really high end this is the perfect balance. The file size is large, but certainly nothing compared to H264.

 

[TzunamiOSX]

Hardware is a more efficient way to encode on the newer M series Macs (I do encoding on both Mac and AMD/Nvidia hardware). I have noticed in the thread many have claimed poorer quality, this is rectified by going into the video tab and pushing up the constant quality slider (I typically use CQ 80). Unless your doing really high end this is the perfect balance. The file size is large, but certainly nothing compared to H264.

Over the time of 14 years, I've encoded around 5000 videos. And during that time, I've learned that no AMD, NVIDIA, or M-Mac hardware encoder has the same output quality as the CPU encoder and get the same size.

You can have acceptable quality but with bigger size.
For example:
I have encoded a Clip in H264 on CPU in a very good quality and get a size of 5 MB. With Videotoolbox (M1 Ultra) I didn't even get the same quality when the file was over 20 MB. with Videotoolbox.

So I can choose whether to use CPU H264 or Videotoolbox H264/H265 and maybe get a file that is approximately the same size, but still not the same quality as the CPU encoded H264.

 

[TzunamiOSX]

Here:
Original Source, 1080p, 800% zoom

H264 CPU, M1 Ultra, 10000 kbps, very slow (6,2 MB)
level=4.1:ref=4:b-adapt=2:direct=auto:analyse=all:me=umh:subq=9:trellis=2:vbv-bufsize=78125:vbv-maxrate=62500:rc-lookahead=60:aq-strength=1.4:bframes=4:deblock=-2,-2:weightp=0

H264 Videotoolbox, M1 Ultra, 10000 kbps, quality (6,7 MB)

H264 Videotoolbox, M1 Ultra, 20000 kbps, quality (13,3 MB)

H265 Videotoolbox, M1 Ultra, 10000 kbps, quality (6,8 MB)

 

[mjbsonic]

Over the time of 14 years, I've encoded around 5000 videos. And during that time, I've learned that no AMD, NVIDIA, or M-Mac hardware encoder has the same output quality as the CPU encoder and get the same size.

You can have acceptable quality but with bigger size.
For example:
I have encoded a Clip in H264 on CPU in a very good quality and get a size of 5 MB. With Videotoolbox (M1 Ultra) I didn't even get the same quality when the file was over 20 MB. with Videotoolbox.

So I can choose whether to use CPU H264 or Videotoolbox H264/H265 and maybe get a file that is approximately the same size, but still not the same quality as the CPU encoded H264.

Hmmm thanks for the feedback, TBH I have not used just Handbrake for all of my encoding (use CLI FFMpeg too). All a bit of a mystery as we are comparing Hardware encoding vs Virtual Hardware encoding. So as per your results, the instructions passed on the same video must be very different in some way, perhaps software has access to different routines?. All very interesting, thanks.

 

[TzunamiOSX]

Hmmm thanks for the feedback, TBH I have not used just Handbrake for all of my encoding (use CLI FFMpeg too). All a bit of a mystery as we are comparing Hardware encoding vs Virtual Hardware encoding. So as per your results, the instructions passed on the same video must be very different in some way, perhaps software has access to different routines?. All very interesting, thanks.

If you give me the right command, I can also test the quality of ffmpeg.
I need the command for render a video with 2 pass at 10000 kbps

 

[TzunamiOSX]

Original again

ffmpeg -i Star.Wars.IV -c:v h264_videotoolbox -profile:v high -movflags "+faststart" -b:v 10000k -c:a aac -b:a 160k "video_10MB.mp4" (size 6,6 MB)

 

[bzgnyc2]

Hmmm thanks for the feedback, TBH I have not used just Handbrake for all of my encoding (use CLI FFMpeg too). All a bit of a mystery as we are comparing Hardware encoding vs Virtual Hardware encoding.

I am not sure what you mean by hardware encoding vs virtual hardware encoding?

These encoding formats are all defined by standard with known algorithms on how to do the encoding. If anything it is the software that defines the standard and represent the best known tradeoffs for encoding. The hardware encoders are an attempt to approximate the software encoders.

So as per your results, the instructions passed on the same video must be very different in some way, perhaps software has access to different routines?. All very interesting, thanks.

It's not a matter of different routines. Rather by definition software has the most flexibility to being optimized. Software encoders can pursue the optimal encoding of a batch of frames taking all the time they need. Consumer hardware encoders (i.e. not the kind that say broadcasters, etc use) find a good encoding quickly.