Compressor 3 Multi-Core Trick

Just wanted to share this with everyone here since there isn't a Compressor forum and this should really be better known.

I'm working on a new Dual Quad core(i.e. 8 core, octo, etc) and wasn't really impressed with the encoding performance gains. I could never really get my cores to bump much above 50% so i did some research.

Basically, Compressor cant really distribute all that processing power to one clip efficiently enough to use of all the power of 8 cores. But there is a trick, called "Virtual Clusters" that can really make your computer fly.

Go to Qmaster Prefs in your system prefs. If everything is greyed out hit Stop Sharing in the bottom right or unlock it if its locked. Choose Share this Comp as QuickCluster with Services.

Under Services check Share and then clicked Managed for Compressor, rendered is only really used for Shake or other apps.

Click Options for selected service and change to 4 instances instead of 1instance. For quads, choose 2 instances. Basically, give each cluster about 2 cores and have a gig or so of memory to go with each.

Give your cluster a name(virtual or something to distinguish it) and keep Include unmanaged services unchecked.

Hit Share and you're ready to go.

In Compressor, when you submit your batch, make sure you submit to that new QuickCluster, NOT This Computer. When you look in Batch Monitor, you will see it split your movie up and is processing away at all different parts of it at once. Plus, your cores will being work much harder.

My best test results are posted below:
This Computer Only -
Name: 09_06pu_downloadProgress.mov
Owner: Scott
Time Elapsed: 0:04:40

Virtual Cluster -
Name: 09_06_downloadProgress.mov
Owner: Scott
Time Elapsed: 0:00:42

Thats right, I took 4 minutes off of a 4:40 sec transcode. Not too shabby...

Aw now that is one for the FAQs if ever I saw it.

Would you mind if I put this under the LAFCPUG FAQs?


Ben



For instant answers to more than one hundred common FCP questions, check out the LAFCPUG FAQ Wiki here : [www.lafcpug.org]

FAQ - that should be a tutorial - great post Scott! I can't believe you took 4 minutes out of 4.40 - that's fantastic. The info certainly needs to be stored somewhere so it doesn't just scroll off the forum and get lost.

Compressor is the unsung hero of FCP Studio 2.

You're essentially talking about distributed rendering to each of the chips in your single computer. For anyone with several of many computers on a network it gets even better! We've set up Compressor to tap into ALL of the computers on our network. For h.264 movies of an hour in length (which we post to the web for review and approval) we've gone from 2 hours plus rendering .... to 6 minutes.

Unfortunately FCP still can't do distributed rendering for FX or quicktimes. Now THAT would be sweet!.

mark

That was great Scott. If you add a few screen shots to this it would make a great tutorial. Wanna do it?

Michael Horton
-------------------

Sure thing, I'll put in the Exporting Section and add some more test results...

Will this work for my core due MBP 17? Anything I should be doing to take a 5 minute 3 track multiclip to H.264 in less than 7 hours?

Jeff

Well, its mostly designed to make Quads and Octos to act like several Dual-cores so starting at a Dual core is pretty much worthless. However, you shouldn't be compressing a multi-clip, you should flatten your multiclip in FCP. Also, do all your compression outside of FCP, its much faster in Compressor and better quality as well.

We have 13 Dual Quad Cores and one Dual-core in our facility and the distributed processing works great when we use Compressor as a stand-alone program (successful about 90% of the time). However, if we export out of FCP using Compressor the process either fails quickly or gets to 100% and then doesn't actually finish. We've tried exporting QT references (current settings) and sending those through Compressor as a stand-alone but it doesn't look as good (interlacing issues, seems to be a lower resolution). Any suggestions on how to conquer the failed distributed processing out of FCP or make the quality better with a QT reference?