Performance benchmark: transcoding on a 2.93ghz unibody MacBook Pro

[Paul Ramsbottom]

I have just started examining Red's software, in advance of a camera purchase, and was curious about ProRes transcode times, as I've read so much, often negative, commentary on this. I spend my professional life examining Mac OS X apps, and I have to say I have been very pleasantly surprised.

I did this short examination based on my particular work-flow needs (using ProRes422 rather than 422(HQ) for example) so it is an entirely selfish study. Your mileage and the relevance of this to you might vary.

I absolutely have to be able to transcode 4K to ProRes422 in the field, and bought the fastest MacBook Pro I could, the optional 2.93ghz 17". The OS is currently 10.5.8 and is installed on the stock 7200rpm 2.5" drive.

Alongside the MBP, I've build a small portable RAID that is connected via the Expresscard slot. This consists of two chinese made enclosures branded Supabox, which have a build-in RAID chip-set. These allow me to RAID 0 two drives in each and then RAID 0 both of them together, effectively allowing me to RAID four 2.5" drives on one two-port eSata card without needing a port replicator (everything I am doing is geared towards a set-up that I can hike into the Himalayas, Andes etc., so every little box and doodat, every power supply (and a redundant copy of each) needs to be thought about.

I'm going to need to work at extreme altitudes and I'm just not liking the performance stats that I am seeing from many SSDs right now. That will improve over time but for now I am using 80GB 5400rpm Seagate EE.25 high altitude SATA drives. These are the only rotational 2.5" drives that are currently made for extreme altitudes and conditions, and I have successfully used them at 17,100ft (which is actually above the stated rating).

Using AJA's System Test, and benchmarking the disk Read?Write with a 2GB 2048x1556 10bit RGB simulation, I get a stable 160MB/s write and a slightly slower 150MB/s Read.

For my first transcode test (and please be kind, because this is my first test and my first post - no warranty implied) I took the 4K .R3D file from here: http://www.ramjetfilms.com/peter/BIL...01_1125JE.RDC/ placed it on my portable RAID and performed a standard ProRes422 transcode using v.21.0.0 of Red Alert.

I was very surprised at how well this system performed. It's a small 10 second clip of course but it chewed through it in a touch over 2 minutes. Time is money of course but I was expecting much, much worse on anything less than a RED Rocket™™ equipped 8-Core Nehalem.

This was so much better than I though, that I can't help but think I must be doing something wrong. I'd be grateful for any feedback or questions. I'd also be grateful for access to any larger test files, so I can do some more tests.

Thanks...

[Uli Plank]

That sounds OK. I use a similar configuration on the road, my MBP is even weaker. But I have a nice small Chinese box called Data Tale connected to my express card, which can work as a RAID-5 (fast and secure) with 4 drives.

In most cases but very low-light shooting you get excellent quality at "Half High" decoding, and if you use Clipfinder for transcoding, both of your CPUs can chew on one clip each.

If you need larger files, send me a PM.

[Paul Ramsbottom]

Just some extra data

8GB of 1067 DDR3 RAM

Silicon image chips sets in both the RAID boxes and the Griffin eSata card I am using.

Repeated the test tonight with the same 4096x2304 .r3d file. This time did a full res debayer to a rec709 color/gamma file, and ProRes422(HQ) to boot just to stretch things a little.

I understand the pain a little better now: 21'.59" for the same 10:20:00 file.

Still better than I thought though.

Surprised at the relative file-sizes, only knocking off about 60MB from the original. Not sure what I was expecting here though.

[Uli Plank]

As I said, a full decode is not needed very often on the road…

ProRes is not optimized for file size, but less load in FCP and visually lossless quality (the latter is disputable, as always).

You can get a quite useable offline version by transcoding at 1/4 debayer to 1280 by 720 in ProRes LT. Small files and fast transcode. Later you can re-conform with Clipfinder and transcode for offline quality.