YouTube Video Codecs, Resolutions & Data Rates

[Brian Iannone]

I just recently learned that YouTube supports some more advanced video codecs like ProRes, so I decided to do a few tests (okay... A LOT of tests) to see exactly what codecs and resolutions it supports.


YouTube Supported Codecs

[TABLE="class: grid, width: 500"]
[TR]
[TD][/TD]
[TD]1080p
(1920x1080)[/TD]
[TD]4K (4096x2304)[/TD]
[/TR]
[TR]
[TD]MainConcept H.26410 Mbps (Level 4.1, Main Profile)[/TD]
[TD]Yes[/TD]
[TD]n/a*[/TD]
[/TR]
[TR]
[TD]MainConept H.26450 Mbps (Level 5.1, Main Profile)[/TD]
[TD]Yes[/TD]
[TD]n/a*[/TD]
[/TR]
[TR]
[TD]MainConcept CUDA H.264[/TD]
[TD]Yes[/TD]
[TD]n/a*[/TD]
[/TR]
[TR]
[TD]Photo-JPEG[/TD]
[TD]Yes[/TD]
[TD]Yes[/TD]
[/TR]
[TR]
[TD]Apple Animation[/TD]
[TD]Yes[/TD]
[TD]Yes[/TD]
[/TR]
[TR]
[TD]Apple Intermediate Codec[/TD]
[TD]No[/TD]
[TD]No[/TD]
[/TR]
[TR]
[TD]DNxHD 220 10-bit[/TD]
[TD]No[/TD]
[TD]n/a**[/TD]
[/TR]
[TR]
[TD]DNxHD 220 8-bit[/TD]
[TD]Yes[/TD]
[TD]n/a**[/TD]
[/TR]
[TR]
[TD]DNxHD 145 8-bit[/TD]
[TD]Yes[/TD]
[TD]n/a**[/TD]
[/TR]
[TR]
[TD]Apple ProRes 4444[/TD]
[TD]Yes[/TD]
[TD]Yes[/TD]
[/TR]
[TR]
[TD]Apple ProRes 422 (HQ)[/TD]
[TD]Yes[/TD]
[TD]Yes[/TD]
[/TR]
[TR]
[TD]Apple ProRes 422[/TD]
[TD]Yes[/TD]
[TD]Yes[/TD]
[/TR]
[TR]
[TD]Apple ProRes 422 (LT)[/TD]
[TD]Yes[/TD]
[TD]Yes[/TD]
[/TR]
[TR]
[TD]Apple ProRes 422 (Proxy)[/TD]
[TD]Yes[/TD]
[TD]Yes[/TD]
[/TR]
[TR]
[TD]QuickTime Uncompressed 8-bit 4:2:2[/TD]
[TD]Yes[/TD]
[TD]Yes[/TD]
[/TR]
[TR]
[TD]QuickTime Uncompressed 10-bit 4:2:2[/TD]
[TD]Yes[/TD]
[TD]Yes[/TD]
[/TR]
[/TABLE]

*MainConcept H.264 and CUDA H.264 encoder does not support 4K.
**DNxHD codecs do not support 4K.

Personally, I find these tests to be very interesting. And surprising. Normally, we export a video from our NLE to H.264 and then upload that to YouTube only for it to get recompressed again by YouTube's encoder. That means that the video is being compressed (and destroyed) twice. Whereas, if you export from your NLE to a codec like ProRes or DNxHD, you can reduce the loss of quality by ensuring the you're delivering a high quality file to YouTube. YouTube will then take that high quality file and compress it down to H.264, but at least you're doing everything you can on your end to receive the highest quality image possible from your YouTube videos.

So, what difference will uploading a 4GB file encoded in Apple Animation make when compared to uploading a 32MB file in H.264 at 10 Mbps?


Apple Animation

MainConcept H.264 at 10 Mbps

Original Frame

View Original 4K Uncompressed Version (10MB)

See the peak signal-to-noise ratio section below for additional tests.


Disregarding the gamma shift, the H.264 version has noticeably more blocking in the gradients and more compression artifacts everywhere. At first glance, the H.264-encoded one may appear sharper due to the image being darker, but zoomed in, you can see that the lines are worse than the version encoded in Apple Animation.

This specific video was about 27 seconds long. The Apple Animation file was more than 4GB. Converting all your videos to Apple Animation from your NLE to upload to YouTube would be inconvenient, to say the least, due to the enormous bandwidth that would be required. Using another lossless compression method like DNxHD or ProRes would be a much better choice. Both the DNxHD 220 8-bit and Apple ProRes 422 (HQ) versions of this video were nearly identical in file size at 722MB and 721MB.

That also means faster exports from your NLE to the file that you'll be uploading to YouTube. Normally, your NLE should be set to render to a format like 8/10-bit Uncompressed or ProRes/DNxHD. If your exporting a file in the same format as what you've been editing in, your conversions are (typically) going to be much faster. Many people aren't aware of this, but converting to H.264 is a very CPU-intensive process. By the way, I know this may come up, so I'll address it now... GPU-accelerated H.264 encoding is not at a point yet where its quality can compete with that of multi-pass CPU-processed video conversions. GPU-powered encoding methods such as Nvidia CUDA is limited to 2-pass encodings. The definition of "multi-pass" CPU encoding is between three and six passes. So, there's quite a substantial difference in quality. But, it is getting better.

YouTube supports resolutions up to a maximum of 4096x2304, which is 4K with an aspect ratio of 16/9. (That's larger than Quad HD or "4K HD.") When you watch a video on YouTube, you have the option to watch it at different resolutions, and ultimately different levels of quality. YouTube takes the video that you upload and converts it to each of those files. YouTube's encoder sets a different data rate for each version according to the file's frame rate and resolution.


YouTube Encoder Data Rates

[TABLE="class: grid, width: 500"]
[TR]
[TD][/TD]
[TD]23.98 fps[/TD]
[TD]29.97 fps[/TD]
[/TR]
[TR]
[TD]4K (4096x2304)[/TD]
[TD]Results on 1/22[/TD]
[TD]Results on 1/22[/TD]
[/TR]
[TR]
[TD]1080p (1920x1080)[/TD]
[TD]3.75 Mbps[/TD]
[TD]5 Mbps[/TD]
[/TR]
[TR]
[TD]720p (1280x720)[/TD]
[TD]2 Mbps[/TD]
[TD]2.5 Mbps[/TD]
[/TR]
[TR]
[TD]480p (854x480)[/TD]
[TD]825 Kbps[/TD]
[TD]1 Mbps[/TD]
[/TR]
[TR]
[TD]360p (640x360)[/TD]
[TD]550 Kbps[/TD]
[TD]725 Kbps[/TD]
[/TR]
[TR]
[TD]240p (400x226)[/TD]
[TD]300 Kbps[/TD]
[TD]325 Kbps[/TD]
[/TR]
[/TABLE]

So, watching a video on YouTube at 1080p will, of course, show you a higher quality file than if you select 480p. But... If the original file that was uploaded to YouTube had a frame rate of 30p, it will be of "higher quality" than if it were a 24p file. Basically, the higher the frame rate, the larger the data rate that YouTube encodes your video in.

YouTube encoder quality: 1080p30 > 1080p24

Keep in mind however that YouTube encodes using VBR. So, there will always be a fluctuation in the data rate. The table above is the most accurate data I've been able to put together.


Peak Signal-to-Noise Ratio Tests
This is a continuation from the first test above comparing a frame encoded in Apple Animation and MainConcept H.264 at 10 Mbps.

Test 1
Difference between uploading a video at QuickTime Uncompressed 10-bit 4:2:2 and MainConcept H.264 at 10 Mbps.


Loss in quality from original and YouTube-encoded QuickTime Uncompressed 10-bit 4:2:2 file:

Loss in quality from original and YouTube-encoded MainConcept H.264 (10 Mbps, Level 4.1, High Profile) file:

Difference in compression artifacts between YouTube-encoded Uncompressed 10-bit and MainConcept H.264 file:

That's the loss of quality you get when you upload an H.264 file at 10 Mbps compared to uploading an Uncompressed 10-bit file. The difference is substantial, both visually and "technically."


Also, for those of you who are curious about the animation that I'm using as the test video, it's a Cinema 4D rendering of a city fly-through. I did a version of the animation at 1920x1080 for the 1080p tests. Rendering took just a little less than two hours in Maxon C4D NET Render across three machines (an 8-core Mac Pro and two PC workstations overclocked at 123%). It's a 16-core, 12.92GHz processing cluster. As I type this, the cluster is rendering the 4096x2304 version of the animation that I'll be using for the 4K tests.


EDIT:
Due to the character limits in this forum, I've had to post the additional tests in a separate post in this thread. See here: http://www.reduser.net/forum/showth...amp-Data-Rates&p=925175&viewfull=1#post925175

 

[Paul Harding]

That's great info I have had a lot of trouble with YouTube destroying my uploads I normally use H264, I'm in the uk and the upload speed is down to .05 which Has become unworkable as it takes so long and then says there's an error, not sure what's going on.

 

[Claus Mueller]

Just downloaded two videos (uploaded 4K 2:1 23.98 fps ProRes422 LT) from my YouTube channel to check the data rates. One is now 9.7 Mbps, the other 12.4 Mbps.

 

[Kemalettin Sert]

downloading is different i guess..same for vimeo.it never buffers the actual video size

 

[Brian Iannone]

downloading is different i guess..same for vimeo.it never buffers the actual video size

Actually, it must buffer the precise size of the video. Buffering is the process of transferring the video data from the web server to your computer. So, in order to watch the entire video, the entire video file must be transferred to your local machine. You can then redirect that data to calculate the total video file size and then read the data rate of the individual tracks in that file.

Vimeo's encoder is very interesting... I'll be writing about it a little later once I finish these 4K tests on YouTube. But, I find YouTube's support for HD formats to actually be better than Vimeo's (surprisingly).

 

[Claus Mueller]

downloading is different i guess..same for vimeo.it never buffers the actual video size

Maybe...
Downloaded another two videos (uploaded 1080p 23.39 h.264), one is now 2.7 Mbps the other 4.0 Mbps. It gives you roughly an idea about YouTube data rates, 10-12 Mbps vs 3-4 Mbps...

 

[Brian Iannone]

Maybe...
Downloaded another two videos (uploaded 1080p 23.39 h.264), one is now 2.7 Mbps the other 4.0 Mbps. It gives you roughly an idea about YouTube data rates, 10-12 Mbps vs 3-4 Mbps...

By the way, how long are your videos?

 

[Claus Mueller]

By the way, how long are your videos?

2-4 minutes

 

[Rosario Balistreri]

great thread

 

[luigivaltulini]

hey Brian,
nice thread, i wait you test about youtube 4k

 

[Les Dittert]

Interesting, thanks for testing this Brian.
Can a user do an h264 compression and have youtube use that one size 'as-is' for the highest res video ? That would avoid a recompress for your best video.

-Les ( I'm the guy that was at the BandPro thing a while back )

BTW: I got my two Scarlets a week ago, and have some 'interesting' info about them vs a Canon still camera, Private message me if you are curious.

 

[Rosario Balistreri]

I think that the test should be done considering a 2560x1400 max resolution...
It's the bigger resolution that you can use on youtube.
Obviously 2.5K files are smaller then 4K ones and so you can upload faster.
We have to understand if uploading ProRes 422/422HQ is still better then H264 at 30-40Mbit

 

[Brian Iannone]

Interesting, thanks for testing this Brian.
Can a user do an h264 compression and have youtube use that one size 'as-is' for the highest res video ? That would avoid a recompress for your best video.

-Les ( I'm the guy that was at the BandPro thing a while back )

BTW: I got my two Scarlets a week ago, and have some 'interesting' info about them vs a Canon still camera, Private message me if you are curious.

Hey Les! It's great to hear from you. And congratulations on receiving your Scarlets! I am indeed very interested, so I'll contact you.

So, are you suggesting that the user compress an H.264 file and specify to YouTube for that file to be used as the highest resolution version of the video, without YouTube recompressing the file? Because that would be fantastic! But unfortunately, that can't be specified to YouTube. That is something, however, that they should seriously consider allowing. That would greatly reduce the loss of quality since most users on their local machines compress videos using multi-pass conversions. YouTube compresses using the fastest possible method (1-pass VBR). (Of course, I'm sure everyone here already knows that.)

I did some quick tests with Vimeo (I'll be doing some more thorough tests later), and I personally find their upload system to be flawed. YouTube provides a maximum of six resolutions to users when watching a video:

- Original (anything between 2K and 4K; maximum resolution of 4096x2304)
- 1080p
- 720p
- 480p
- 360p
- 240p

Therefore, the user can choose what quality they want to watch based on their Internet speed, computer capabilities, monitor size, etc. (which I think is great). But, Vimeo only offers two:

- "HD" (maximum resolution of 1920x1080)
- "SD" (typically between 640x360 and 848x480)

Now, the problem with Vimeo is that they support 1080p. Meaning, you can upload a file in 1080p. The trick is, you have to select the, "Display this video in 1080p" checkbox under the video files settings. By default, it's unchecked. So, by default, even if you upload your video in 1080p, it will be displayed at 720p (which Vimeo refers to as the "HD" version).

While that's not too bad of a system, most users are going to go under the video file settings and notice that little checkbox. Which is why if you browse around the Vimeo site and take a look at the resolution that most of the videos are being displayed at, it's 1280x720 even though the user uploaded a 1080p file.

So, that's just an observation I've made. I really should have put that info in my main post though... Oops.

I think that the test should be done considering a 2560x1400 max resolution...
It's the bigger resolution that you can use on youtube.
Obviously 2.5K files are smaller then 4K ones and so you can upload faster.
We have to understand if uploading ProRes 422/422HQ is still better then H264 at 30-40Mbit

Do you mean that you think I should conduct the tests with a video file at a resolution of 2560x1400? For the 1080p test, I used a video file at 1920x1080, but I'll be doing some 4K tests today with a 4096x2304 file. I'm not completely sure what you meant in your post, but are you suggesting that I do the tests at 2.5K instead of 4K?

Just want to be sure. =)

 

[jake blackstone]

Interesting discussion.
I wasn't aware, that youtube supports Prores uploads. Obviously this is much better, than uploading H264 and then let youtube transcode it again to H264. So, to test it I did two uploads-one Prores and the same material in H264 at same resolution from the same source. Unsurprisingly, Prores upload ended up looking better, than H264. Here are both examples:
Prores http://www.youtube.com/watch?v=ynrGtPmgJ90
H264 http://www.youtube.com/watch?v=htCyvK3bW1k
Herde is the same file on Vimeo at the same resolution
http://vimeo.com/29028959

 

[Doug Beatty]

Great thread, Brian.

Jake, that ProRes Youtube file kicks the crap out of the Vimeo HD file. I wonder what's going on under the hood at Vimeo. I'm watching these full screen on a Dell U2711 and the difference in quality is very noticeable.

 

[Brian Iannone]

Interesting discussion.
I wasn't aware, that youtube supports Prores uploads. Obviously this is much better, than uploading H264 and then let youtube transcode it again to H264. So, to test it I did two uploads-one Prores and the same material in H264 at same resolution from the same source. Unsurprisingly, Prores upload ended up looking better, than H264. Here are both examples:
Prores http://www.youtube.com/watch?v=ynrGtPmgJ90
H264 http://www.youtube.com/watch?v=htCyvK3bW1k
Herde is the same file on Vimeo at the same resolution
http://vimeo.com/29028959

Great thread, Brian.

Jake, that ProRes Youtube file kicks the crap out of the Vimeo HD file. I wonder what's going on under the hood at Vimeo. I'm watching these full screen on a Dell U2711 and the difference in quality is very noticeable.

Why does the Vimeo one look so... Not very good? Simple. It's not really 1920x810.

Those of you who are watching the video on a monitor that has a resolution of 1920x1080 or above will be able to understand what I'm talking about. Jake, I see that you uploaded your video to Vimeo at 1920x810. The video that Vimeo is showing however, is 1280x540. Why? Because Vimeo tricks you by forcing you to check a specific checkbox under the video file settings if you want your video to be shown at 1920 pixels wide (and by making you buy Vimeo Plus, too).

This is the checkbox I'm referring to:

By default, it's turned off. There's a quick way to check whether or not a video is being displayed at 1920- or 1280-pixels wide, but you need to have a monitor with a resolution of 1920x1080 or higher. Play the Vimeo video, make sure "HD" is on, click on fullscreen, and then turn scaling off by clicking, "Scaling is on." That way, the Vimeo player will show you the video at actual size (1:1).

This is what you'll see if the uploader of the video did not check the "Display this video in 1080p" box:

Vimeo shows you a small resolution file and then blows it up to fill your monitor...

That's why the YouTube versions of Jake's videos look better. Because the YouTube ones are being shown at 1920 pixels wide while the Vimeo one is being shown at 1280 pixels wide.

So basically, does Vimeo support 1080p? Yes. Do they make it easily available for users to display their videos at 1080p? No. YouTube, on the other hand, both supports 1080p and makes it easy to for users to upload and display their high resolution files. Especially now that YouTube supports resolutions up to 4K, I think YouTube is going to become a more popular choice for filmmakers to show off their work.

By the way Jake, that version of your video that you uploaded in ProRes really does look excellent. I have a question about the H.264 one you uploaded on January 3rd... What was the data rate?

Also, an update on my 4K tests... Still rendering the test video from C4D. I had a minor delay since one of the machines in the processing cluster disconnected from the server this morning and caused the entire render process to fail. So, I'm starting over on some portions of the animation and it should be done by 11 PM.

 

[jake blackstone]

No, both Vimeo and Youtube are the same resolution. Yes, I just re-uploaded the file in question on Vimeo just two days ago for the exact reason you described for non Vimeo Plus crowd and also because without Vimeo Plus there is no mobile version created. So, without Vimeo plus iphone can't play those embedded videos. I use Vimeo embedding on my web site, so I just upgraded to Vimeo Plus a few days ago. You can upgrade your videos to higher resolutions without re-uploading by simply checking the checkmark "upgrade to highest quality", but for mobile version you must re-upload it as a Vimeo Plus user.
I used 1.5Mb data rate for H264 Vimeo upload. Jan, 3rd is the old upload. As I said, I re-uploaded it two days ago for stated reasons.

 

[Brian Iannone]

No, both Vimeo and Youtube are the same resolution.

Are you saying that both the Vimeo and YouTube version are 1920x810?

 

[luigivaltulini]

Interesting discussion.
I wasn't aware, that youtube supports Prores uploads. Obviously this is much better, than uploading H264 and then let youtube transcode it again to H264. So, to test it I did two uploads-one Prores and the same material in H264 at same resolution from the same source. Unsurprisingly, Prores upload ended up looking better, than H264. Here are both examples:
Prores http://www.youtube.com/watch?v=ynrGtPmgJ90
H264 http://www.youtube.com/watch?v=htCyvK3bW1k
Herde is the same file on Vimeo at the same resolution
http://vimeo.com/29028959

Hey Jack,
we say that there is' a big difference, both in detail at the chrominance in favor of ProRes.
How were the two mb h264 vs ProRes files?
can try to do an upload of uncompressed 8-bit QuickTime in to see if it works?

thanks


[FONT=arial, sans-serif]
[/FONT]

 

[jake blackstone]

Are you saying that both the Vimeo and YouTube version are 1920x810?

Yes