Real World Video Compression

17 August 2008

Encoder Shootout – A Comparison of 4 Mac Compression Tools

Posted by andybeach under: Ramblings; Research; Software .

 

I was recently hassling a new friend(let’s call him Jeff) via email over some quality issues I was seeing in a video podcast he is publishing.  The issue isn’t really a super big deal, but I’m trying to help him resolve it so the production puts its best foot forward.  We are still trading emails about his overall workflow, as I don’t think its the encoder application he’s using, but some minor setting that is introducing the issue i noticed.  

In the course of discussing his problem however, I learned he has been using the El Gato Turbo264 1.3 ($99) to create his final encodes.  His reasoning for using it was primarily the speed bump he got over an export from Compressor (he is using a Macbook Pro for production).  El Gato uses its own proprietary USB hardware with a specialized DSP (Digital Signal Processing) chip to offload (and speed up) the encode process).  The Benefit as marketed by El Gato in a nutshell is to either a) speed up the encode on older machines or b) offload the encode process on newer machines to keep you working.  I had checked this product out briefly when it first came out, but it had been awhile and I wanted to see what both the quality and the encode times were like.  For a baseline, I compared it to the other encoders we discussed in our various e-mails – Apple’s Compressor 3.0.3 (included with Final Cut Studio), Telestream’s Episode Pro 5.0.1 ($999) , & Techspansion’s Visual Hub 1.3.4 ($23.32).  I was primarily just curious about the encode times, however I did want to at least marginally take quality of the encoded files into account, as who cares how fast the compression is if the quality is garbage.  

Methodology

I used the same file in all four applications with default settings or templates.  I encoded the clip to two different common file types – Apple TV and High Quality iPod/iPhone.  Since I have access to both a Mac Pro and a Macbook Pro, I encoded the same files on both systems.  My Macbook Pro is a 2.33 GHz dual core with 2 GB of RAM running 10.4.11.  My Mac Pro is a 2.66 GHz quad core with 5 GB of RAM running 10.5.4.  Each system had the same versions of the 4 encoding apps as fresh installs with default templates or default settings (VH doesn’t really have templates per se).  The clip used is a DV-NTSC clip from Zoom In Online that I also used in writing RWVC (this is the original posting at ZIO).  The clip is approximately 6 minutes long.  Although it is letterboxed in the source, I chose not to crop or customize the settings in anyway as I didn’t want to introduce any additionally processes into the encode pipeline.  I ran each encode separately with all other applications closed so nothing could take away from the encoding process.

The results after the jump!

Speed Results

Below is a table outlining my results.  I rounded each time to the nearest minute rather than present minutes & seconds (over 30 seconds I rounded up; under 30, I rounded down).

Macbook Pro
	Turbo264	Compressor	Episode	Visual Hub
DV Clip to ATV	6	15	19	8
DV Clip to iPhone	6	21	13	7
Mac Pro
	Turbo264	Compressor	Episode	Visual Hub
DV Clip to ATV	6	9	17	4
DV Clip to iPhone	5	17	12	4

As you can see from the results, on the Macbook Pro, the Turbo264 was the clear winner, coming in at realtime for the encode for both versions.  Surprisingly however, Visual Hub came in second at fairly close to real time.  Both Episode and Compressor did about the same coming in at 2 to 4x depending on the encoder and setting.  Interestingly enough, each took a great deal longer on one encode, but not the same one (compressor struggled with the iPhone clip, while Episode choked on the Apple TV one).  Arguably, I can guess that both Episode & Compressor are doing multi-pass encodes by default, while Visual Hub is doing a single pass.  I’m not sure about Turbo as its offloading the encode process to the external DSP, but i’ll throw a guess that it’s single pass until i learn otherwise.  Since we’re only concerned with speed in this part of the test, i wont speak to whether going multi-pass actually benefits and makes the extra cycles worthwhile (we’ll compare the images further down to make that call).  

Moving to the Mac Pro, the Turbo’s times were almost identical (they were slightly faster) while Visual Hub’s nearly halved.  This makes sense as Turbo only knows to offload the processing to the same stick, while VH was taking advantage of the additional cores on the Mac Pro to encode faster.  This is the result I actually expected to get, so I wasn’t surprised.  I was surprised by the results of both Compressor and Episode again however as they did not drop as dramatically.  Infact with the exception of Compressor encoding for Apple TV, the encodes were only a minute or two faster on the 4 core Mac Pro.

Based on just these results, the Turbo would technically be the winner if you were on a laptop and encoding, but I’m not sure the extra $80 in price would be worth the few minutes saved, so I’d have to think long and hard before I picked.  If, however, you are on a Mac Pro, save the $ and run Visual Hub.  As for Episode & Compressor, both were considerably slower, so if speed was my only consideration, i’d probably leave them out.  That said, there is a wealth of other formats and options packed into both, especially Episode Pro, so while they disappointed speed wise, they should wholly be ignored or discarded.  

Image Results

Speed to finish is only one consideration though – quality is very important as well.  Note I didn’t say it was more important.  The balance between quality and speed is personal and often dependent on other factors in your project.  If its just a rough cut you are sharing with clients, quality probably doesn’t matter much, but speed does.  If however, this is the final encode to be used to author a DVD or present to a larger crowder, it may make sense to sacrifice speed for the quality of the end results.

Below are a series of screen shots I used to compare the images side by side of the Apple TV encodes made on my Mac Pro.  I specifically wanted samples that showed a typical headshot, live video mixed with graphic titles, graphics by themselves, and a still image being panned (an effected used repeatedly in the edit.  Quality wise off of just an eyeball comparison, i can tell you the encodes made on the Macbook Pro match the same settings on the Mac Pro (as I anticipated they would).  The finished results from encoder to encoder however are fairly different, both in file size and image quality.

Each clip is named after its encoder to make life easier.  To see the full size screenshots, click the image.  I guess the most immediate and obvious element is that all encoders but the Turbo264 maintained the current aspect ratio.  The Turbo dropped its output to 610 x 458. In the final screen shot i manually scaled up the Turbo image to match the others so i could compare the quality of the deinterlace of the mesh background (i’ll talk more about this in a minute).  

If you notice (probably easier if you click through to the original shot) the images are slightly different.  each of the encoders is handling the Pixel Aspect Ratio slightly differently.  From a computer screen, the Turbo clip plays and looks the best, though it is worth noting that when all four are loaded on a Apple TV (that is the target profile after all) they playback with no image distortion.  I thought Visual Hub actually did the worst of the four here for playback on a computer screen as its image looks very stretched.  As I mentioned though, it looked fine when played back through my Apple TV, and its ipod file looked fine on a computer screen, so i won’t complain too much.

It is hard to tell in the small pictures here, but if you look at the original shots of all three, you’ll see that both Compressor and Episode did a worse job of deinterlacing the clip than Turbo264 and Visual Hub.  This is perhaps most obvious in the final image which is a still image being panned.  In the image is a mesh background that introduced a slight interline twitter (a slight artifact brought about because of the fine grain of the original image and the encoder being unable to fully reproduce it).  Both Turbo264 and Visual Hub did a very good job of handling the fine mesh background.  I played all four clips at the same time and screen-capped them to further show this section, as the artifacting is more noticeable in motion.  Click here to play the movie in a new window.  Keep in mind that this is a re-encode of screen captured material that had already been encoded, so there is some quality loss, however the difference in the image playback is still obvious.

 

When viewed on a television via my Apple TV, both the image quality of the Visual Hub and Turbo264 clips were superior to the Episode Pro & Compressor clips.  Episode does have multiple de-interlace options and I know for a fact if i changed the settings I could have improved the quality of the final image, but this test was about default settings, so I left everything as is.

File size wise, the clips were again, quite different from each other, with the Turbo files generally being the largest across all files, while the Compressor files for Apple TV were twice as large as the nearest file.  This really reflects the difference in opinions on encode settings more than anything.  I’m not crazy about wasting bits unnecessarily, especially for files destined to be share on the Internet, so i tend to like the Visual Hub files the most here, as they consistently are the smallest.

Conclusion

In summary, for quick, fairly reliable “no thinking” encoding, both the Turbo.264 and Visual Hub are excellent choices.  If you are speed conscious and working on a Macbook Pro, the Turbo is an obvious choice, but if you are cost conscious or have already invested in a Mac Pro, go with the cheaper solution.  Episode Pro & Compressor can do a good job and have their place in productions, but you will most likely want to customize the settings you are using.  I don’t want to know either too hard here – they just weren’t the best tools for the job.  There are plenty of occasion when I’ve put both into duty to do functions that neither Turbo.264 nor Visual Hub were appropriate for (that’s right girls, you’re pretty in your own special way).

 

 

 

4 Comments so far...

Leave a Reply

You must be logged in to post a comment.