High Noon compression tests: Dueling in the Wild West
The title comes from the fact that I am running these tests in a machine that is more suited for cattle rustling than speedy file creation. After a frustrating day of making example files for a client, I realized I could not give the same estimates that I was used to in regards to when compressions would be ready. In an attempt to get to know my new full time workstation, I needed to put it through the paces to figure out what it can be depended on to do. Also I have primarily used Soreson Squeeze and Autodesk Cleaner XL to make all my files so this was a chance to familiarize myself a little more with Apple Compressor and MPEG StreamClip, a program I had never used before but heard amazing things about. After that initial horrendous day of crunching frames, I was moaning on twitter about my experiences and Justin Lee (@justinLanier on Twitter) recommended making a virtual cluster through Qmaster to utilize more of the RAM. I found this article http://www.kenstone.net/fcp_homepage/compressor_multi_cores_stitzer.html
, in addition to many more, so decided to throw this new wrinkle into the mix as well. Overall, I am glad I did the whole test because I have a much better understanding of the limitations of my machine, but really it was a hefty reminder that I need to upgrade. Here we go:
Apple 1.8 GHz Power PC G5, 3 GB RAM
I warned you. This thing has been outdated since the day they went to Intel chips. Interestingly enough that was about 5 months after I bought the thing, but I have spent plenty of happy hours ranting about that and itâ€™s out of my system, I swear. The important thing is it still gets work done. I would rather be slowed down and practicing the craft than waiting for the newest MacPro to finally come out and not editing.
Apple Compressor 3
Apple Compressor 3 using a 2 core virtual cluster
640 x 360 Multi-pass H.264 progressive
I tried to make the parameters the same but the setting differ between Compressor and MPEG StreamClip, so they were matched as closely as possible.
The standard setting of medium quality H.264 at 1300 KB/sec
Resize filter set at Fast (Nearest Pixel)
Deinterlace set at Fast (Line averaging)
In MPEG SC:
The standard quality slider at 50% H.264 at 1300 KB/sec
Interlaced scaling deselected
Deinterlace Video selected
I think I have this pretty close, but if anyone has any optimization tips I would love to hear them.
30 sec 1280 x 720 ProRes 422 (HQ)
30 sec 1920 x 1080 ProRes 422 (HQ)
30 sec 1920 x 1080 DNxHD 145
60 sec 1920 x 1080 ProRes 422 (HQ)
60 sec 1920 x 1080 DNxHD 145
You are probably asking why in the hell I have the Avid DNxHD codec included. Well, the last three years of my professional editing career have been on an Avid Adrenaline PC. A while back I was trying to find a good archival format for all my projects that would work with my G5 at home, and initially it was either Uncompressed or Animation files. These monstrosities were just not working out at 2-3 GB a minute. Well, I discovered I could install the Avid codecs on my G5 and work with DNxHD files in FCP. Badass, problem solved for the most part. There are some gamma shift issues between 601 Avid and RGB FCP, but I find them more manageable than huge file sizes. Initially, my plan was to transcode all of the DNxHD files to ProRes but first I wanted to see how FCP and compressor dealt with the DNxHD native. These files will never be used again for heavy editing; they are just for potential demo reels and client examples. It would be great if I didnâ€™t have to waste the time and disc space transcoding all of these clips to ProRes.
First I took my latest demo reel (shameless plug) in DNxHD and transcoded it to ProRes. I exported Quicktime reference files from a ProRes sequence and from a DNxHD sequence in FCP, then closed FCP. To create some sort of control environment I closed down all running programs other than the one being tested, Microsoft Word, and the Activity Monitor.
So, with a giant cup of coffee in hand and the sounds of my new fiancÃ© watching â€œSay yes to the dressâ€ upstairs, the dueling began:
Round 1: 30 sec 1280 x 720 ProRes 422 (HQ)
Compressor 10 min 49 sec 4.7 MB
Compressor Virtual Cluster (VC) 10 min 06 sec 5.8 MB
MPEG StreamClip 9 min 21 sec 10.3 MB
Round 2: 30 sec 1920 x 1080 ProRes 422 (HQ)
Compressor 10 min 55 sec 4.8 MB
Compressor Virtual Cluster (VC) 15 min 13 sec 5.8 MB
MPEG StreamClip 7 min 18 sec 10 MB
Here I basically stared at the screen wondering, why in the hell did the VC take 5 min longer, I must have screwed something up. So I went back through all the motions and tried it again.
Round 2 REDO: 30 sec 1920 x 1080 ProRes 422 (HQ)
Compressor 11 min 48 sec 4.8 MB
Compressor Virtual Cluster (VC) 15 min 55 sec 5.8 MB
MPEG StreamClip 7 min 27 sec 10 MB
Still don’t understand what is going on with the VC and now the regular compressor method took longer too. Balderdash. This is where I start wondering if I set this VC method up right at all.
Round 3: 30 sec 1920 x 1080 DNxHD 145
Compressor 15 min 27 sec 4.6 MB
Compressor Virtual Cluster (VC) 16 min 20 sec 5.4 MB
MPEG StreamClip 6 min 48 sec 10 MB
What?!?! I must have messed something up MPEG StreamClip.
MPEG StreamClip REDO 7 min 04 sec 10 MB
Guess not. Wow, that is faster.
Round 4: 60 sec 1920 x 1080 ProRes 422 (HQ)
Compressor 19 min 34 sec 10.4 MB
Compressor Virtual Cluster (VC) 33 min 22 sec 11.8 MB
MPEG StreamClip 20 min 21 sec 20.4 MB
Really?!?! MPEG StreamClip is slower now?
MPEG StreamClip REDO 21 min 16 sec 20.4 MB
I don’t know what to say. This has me in disarray.
Round 5: 60 sec 1920 x 1080 DNxHD 145
Compressor 27 min 48 sec 10.1 MB
Compressor Virtual Cluster (VC) 46 min 36 sec 11.8 MB
MPEG StreamClip 21 min 18 sec 20.3 MB
- My computer is not the sharpest blade in the drawer when it comes to compressions and it canâ€™t seem to do exactly the same thing twice. But, I still love it and hope it doesnâ€™t repay by judgments by killing itself in the midst of an edit. Please.
- I sure as hell messed up this Virtual Cluster thing. It does not seem to be helpful at all. Most of this is probably because my G5 doesnâ€™t have the cores to utilize a true speed increase. Also the operator may be extremely confused on all things Qmaster and Virtual Cluster related.
- In terms of speed, MPEG StreamClip wins for the most part. Especially for transcoding clips in the DNxHD codec. Still donâ€™t know what changed when it dealt with the 60 second ProRes test. Compressor definitely does a faster job with the ProRes than the DNxHD. I have not done any batch processing with MPEG StreamClip at this point so I cannot say it wins in that regard, but I certainly will be experimenting with it soon.
- So far I have said nothing about Quality, which is the most important aspect of the final compressions in my mind (this attribute changes momentarily if I am very close to an important deadline and a producer is calling me mean things). In my comparisons the final files were all very, very similar in terms of quality. The one difference was deinterlacing. MPEG StreamClip did a much better job than Compressor (even when I changed Deinterlace to Better and Best). I, personally, am very anal about deinterlacing, so this may be the biggest reason I continue using MPEG Streamclip for clips aimed at computer screen viewing.
There were a couple more tests I put these files though after this series, but I will be sharing those in the next post. Again, any comments or words of advice are immensely welcome. Although I tried to get as scientific as possible, there are a multitude of areas in which I may have taken missteps.
|Print article||This entry was posted by Glen on July 22, 2010 at 12:06 pm, and is filed under Blog, Compression, Tests. Follow any responses to this post through RSS 2.0. You can leave a response or trackback from your own site.|
No comments yet.
about 2 years ago - No comments
Well it seemed as I reached the end of the tests I did for High Noon Compression Tests I had not tortured myself enough. There were still some potential scenarios that I thought I could quickly check off the list and give myself some peace of mind. If you read the results from those tests,…