MKV -> MP4 CPU Choice

[Mister Woof]

Regarding reduction in conversion time.

 

10900k or 5800x? Which is better?

 

How big of an uplift in time reduction over an i7-8700?

 

Edit: with compression. Presuming utilizing NVENC for video, and still CPU for audio.

[Guest]

You really are trying to find an excuse to upgrade her aren't you.

 

Just do it when she's not looking and see if she notices anything. Wifes always like surprises!  

[Kilrah]

What conversion exactly and with what tools? MKV->MP4 can be a simple remux that takes 0 CPU effort. Some tools have single-threaded bottlenecks, etc...

[mariushm]

You would only care about CPU type if you need to re-compress the video or audio.

 

Both MKV and MP4 are containers, they contain video, audio and subtitle tracks and potentially other things.

 

You can simply convert a MKV to a MP4 without recompressing the video or the audio, if the video format or audio format are supported by the MP4 file format.

If this is the case, the speed of your storage will be mostly what matters, not the processor.

 

If you need to convert the audio (for example from AAC to mp3 or some other format), most audio encoders are single threaded, so in theory the highest frequency / highest CPU IPC (performance per core) will win ... and between those processors, it's a wash, the difference of performance between them is extremely small.

 

If you need to convert the video, then it depends if you plan to use strictly software encoder, or if the video encoder supports Quicksync or if you're willing to use video card's encoder to speed up encoding.

 

You can use a tool like MeGUI to extract the video and audio tracks from a MKV file (see HD Streams Extractor in the menu) and then Mux the video and audio tracks into MP4 container from somewhere in the menu. 

You can use mkvextractGUI +MKVToolnix  to extract tracks from MKV if you don't want to use MeGUI  and you can use MP4Box or Handbrake to combine video and audio tracks into MP4 without recompression.

[Mister Woof]

It will involve compression - i actually don't really know how she does it that's her realm. Raw output files from the rips are ~30gb and apparently they get compressed down to 3-4gb.

 

My understanding is she can use NVENC for the video aspect, but will still rely on CPU for audio

30 minutes ago, mariushm said:

You would only care about CPU type if you need to re-compress the video or audio.

 

Both MKV and MP4 are containers, they contain video, audio and subtitle tracks and potentially other things.

 

You can simply convert a MKV to a MP4 without recompressing the video or the audio, if the video format or audio format are supported by the MP4 file format.

If this is the case, the speed of your storage will be mostly what matters, not the processor.

 

If you need to convert the audio (for example from AAC to mp3 or some other format), most audio encoders are single threaded, so in theory the highest frequency / highest CPU IPC (performance per core) will win ... and between those processors, it's a wash, the difference of performance between them is extremely small.

 

If you need to convert the video, then it depends if you plan to use strictly software encoder, or if the video encoder supports Quicksync or if you're willing to use video card's encoder to speed up encoding.

 

You can use a tool like MeGUI to extract the video and audio tracks from a MKV file (see HD Streams Extractor in the menu) and then Mux the video and audio tracks into MP4 container from somewhere in the menu. 

You can use mkvextractGUI +MKVToolnix  to extract tracks from MKV if you don't want to use MeGUI  and you can use MP4Box or Handbrake to combine video and audio tracks into MP4 without recompression.

 

[Guest]

35 minutes ago, Mister Woof said:

It will involve compression - i actually don't really know how she does it that's her realm. Raw output files from the rips are ~30gb and apparently they get compressed down to 3-4gb.

 

My understanding is she can use NVENC for the video aspect, but will still rely on CPU for audio

 

The 10900K is 20 threads though. It's really a no brainer. The IPC difference isn't enough to justify the 4 thread difference IMO. Unless you can sport her with a 5900x, go with the Intel then. 

[Mister Woof]

4 minutes ago, ShrimpBrime said:

The 10900K is 20 threads though. It's really a no brainer. The IPC difference isn't enough to justify the 4 thread difference IMO. Unless you can sport her with a 5900x, go with the Intel then. 

Originally planned to give the extra parts to my sister but then she said she didn't want it, lol.

 

My Craigslist, Facebook, and Nextdoor ads are not getting any traction despite very generous pricing.

 

Considering my options for using these things in the event I can't get rid of it.

[Guest]

3 hours ago, Mister Woof said:

Originally planned to give the extra parts to my sister but then she said she didn't want it, lol.

 

My Craigslist, Facebook, and Nextdoor ads are not getting any traction despite very generous pricing.

 

Considering my options for using these things in the event I can't get rid of it.

Those extra parts are AM4, might as well go for the 5800X. 

Our 2700X build over here is pretty solid. Haven't had any deal with software or being glitchy.

I'd say just build the wife the AMD system, it'll be an improvement on several different fronts.

[CerealExperimentsLain]

10 hours ago, Mister Woof said:

It will involve compression - i actually don't really know how she does it that's her realm. Raw output files from the rips are ~30gb and apparently they get compressed down to 3-4gb.

 

My understanding is she can use NVENC for the video aspect, but will still rely on CPU for audio

As someone who compressed 30 *days* of anime for her GF so she could fit it all on a 512GB MicroSD card before she deployed with the military and used NVENC on a 3080 to do it.... It sucks for compression.  NVENC is *fast* but with HEVC NVENC vs software encoding at 'Slower' the software encoding gets you half the file size at the same quality setting.

Granted, I had to squish 30 days of anime in less than 2 weeks, NVENC was the answer there, if file size is your concern and you have time to wait on encodes, don't use NVENC.

[Mister Woof]

7 hours ago, CerealExperimentsLain said:

As someone who compressed 30 *days* of anime for her GF so she could fit it all on a 512GB MicroSD card before she deployed with the military and used NVENC on a 3080 to do it.... It sucks for compression.  NVENC is *fast* but with HEVC NVENC vs software encoding at 'Slower' the software encoding gets you half the file size at the same quality setting.

Granted, I had to squish 30 days of anime in less than 2 weeks, NVENC was the answer there, if file size is your concern and you have time to wait on encodes, don't use NVENC.

Thank you for the input. We are making our DVD/Bluray backups so we can use them on our home server. Does CPU video encode in handbrake favor cores or frequency? 

[CerealExperimentsLain]

2 hours ago, Mister Woof said:

Thank you for the input. We are making our DVD/Bluray backups so we can use them on our home server. Does CPU video encode in handbrake favor cores or frequency? 

At least for HEVC it leans towards core speed over count.  Since my partner deployed I've been encoding more stuff for her, current airing shows, which she then downloads on the ships sat internet, not great bandwidth so files have to be small but I have the time to do slower encodes.  Going from a 3900X to a 3950X, encoding speed in handbrake only went up 15%.  Though I have more cores so I can do more concurrent encodes to speed it up.  That is to say, if it's barely using more than half the CPU, why not do two encoding jobs at once?

[Mister Woof]

7 minutes ago, CerealExperimentsLain said:

At least for HEVC it leans towards core speed over count.  Since my partner deployed I've been encoding more stuff for her, current airing shows, which she then downloads on the ships sat internet, not great bandwidth so files have to be small but I have the time to do slower encodes.  Going from a 3900X to a 3950X, encoding speed in handbrake only went up 15%.  Though I have more cores so I can do more concurrent encodes to speed it up.  That is to say, if it's barely using more than half the CPU, why not do two encoding jobs at once?

Much appreciate the feedback. I'm going to think on this a bit more.

 

[Craftyawesome]

18 hours ago, CerealExperimentsLain said:

As someone who compressed 30 *days* of anime for her GF so she could fit it all on a 512GB MicroSD card before she deployed with the military and used NVENC on a 3080 to do it.... It sucks for compression.  NVENC is *fast* but with HEVC NVENC vs software encoding at 'Slower' the software encoding gets you half the file size at the same quality setting.

How did you compare quality? That seems abnormally bad. NVENC (turing or better) HEVC somewhat significantly beats x264. And HEVC is only supposed to be 25-50% better than h264.

[CerealExperimentsLain]

8 minutes ago, Craftyawesome said:

How did you compare quality? That seems abnormally bad. NVENC (turing or better) HEVC somewhat significantly beats x264. And HEVC is only supposed to be 25-50% better than h264.

To be clear, I'm comparing two HEVC encodes from the same source, one done with NVENC one done with software.  None of this is about h.264 vs HEVC.  At the same Constant Fate Factor NVENC yields a file about 2x larger than software encoding at 'Slower' setting.  NVENC is known for not being the most efficient, in terms of bitrate, encoder.  It comes from it being a fixed ASIC that's fast as all heck.  Software encoders are constantly improved and tweaked but that comes at a big cost of processing time using a general purpose CPU.

But NVENC is not really made for file efficiency, it's about speed, doing jobs faster than real time generally for the purposes of streaming.

When encoding a file to be downloaded from a military satellite to a ship in the ocean, my objective isn't 'Make an HEVC encode half the size of an h.264 encode with the same quality' it's 'Cram this 25mins of anime into about 65 megabytes as efficiently as possible'.

[Craftyawesome]

17 minutes ago, CerealExperimentsLain said:

To be clear, I'm comparing two HEVC encodes from the same source, one done with NVENC one done with software.  None of this is about h.264 vs HEVC.

I know, I was just saying that because I couldn't find a direct comparison.

HEVC > H264 by 25-50% (presumably software)

NVENC HEVC > x264

So software HEVC should be somewhat close to NVENC, not twice as good. Maybe anime could be a weak point?

28 minutes ago, CerealExperimentsLain said:

At the same Constant Fate Factor NVENC yields a file about 2x larger than software encoding at 'Slower' setting.

NVENC doesn't support CRF to my understanding. CQ is less efficient if you used that. https://slhck.info/video/2017/02/24/crf-guide.html "A constant QP encode at QP=18 will stay at QP=18 regardless of the frame (there is some small offset for different frame types, but it is negligible here). Constant Rate Factor at CRF=18 will increase the QP to, say, 20, for high motion frames (compressing them more) and lower it down to 16 for low motion parts of the sequence."

 

Not to mention CQ and CRF aren't comparable between encoders.

[mariushm]

HEVC is "optimized" for 1080p and higher, the quality increases will be less noticeable with less than 1080p and almost nonexistent with SD content (dvds, less than 720p)

 

Most modern codecs are designed with a sort of maximum threads per frame ... if you use more threads, those threads won't be used as effective as possible and the quality may also decrease... so often it's better to use less threads, but encode several videos in parallel instead.

 

I may be wrong, but if I remember correctly, x264 software encoder suggested one thread per 64 vertical pixels ... so once you're past 16 threads on 1080p content you're not benefiting that much from threads. You can however render two videos at the same time, and configure each encoder to use something like 12 threads, for a total of 24 threads 

 

Like everyone above me said, NvEnc is less configurable than a software encoder, and it's designed for speed and for reduced hardware complexity (use as little silicon as possible, to make the chips cheaper) so it will never be able to do as deep of an analysis on the video it gets to encode as a software encoder would be able to do. 

 

So a software encoder has the potential (if you configure it correctly) to retain more quality in the same amount of disk space compared to NvEnc, but it will use a lot more CPU time to achieve that. 

 

Anandtech has a cpu benchmark that uses x264 software encoder to test encoding speed, and you can see AMD Ryzen processors are faster than Intel : https://www.anandtech.com/bench/CPU-2020/2777

 

Note though that hevc software encoder may use AVX extensions which are more powerful on Intel, so Intel processors may do a bit better than AMD on hevc encoding, BUT with the downside of Intel processors being way more power hungry when using AVX.