Apple Snow Leopard for Video Producers, Part 2
In the last edition of Final Cut Pro Insider, I detailed the new performance-related advances in Apple's new Snow Leopard. In this issue, I share some benchmark tests with a Mac Pro and Mac Book Pro comparing Leopard (OS X 10.5) to Snow Leopard (OS X 10.6) in 32-bit and 64-bit modes. I ran three sets of tests: one for Apple Final Cut Pro, one for Adobe Premiere Pro, and one involving streaming encoders Sorenson Media Squeeze and On2 Technologies Flix Pro.
The main question that I tried to answer was, Should you expect significant performance boosts if you install Snow Leopard? The overall answer is yes in some configurations, but overall, the results were mixed and depend upon whether you're running Snow Leopard in 32-bit or 64-bit. As a reminder, Apple shipped Snow Leopard with two kernelsone 32-bit and one 64-bitand it runs the 32-bit kernel by default on all computers except certain Xserve workstations to maintain compatibility with devices with 32-bit drivers.
To run the OS in 64-bit mode, hold down the 6 and 4 keys when booting your Snow Leopard-upgraded Mac. It will remain in 64-bit mode for subsequent startups until you hold down the 3 and 2 keys while booting. To determine which kernel is running, check System Profiler (click About this Mac, then More Info) and click the Software profile.
With this as background, on to the tests.
With Final Cut Pro, I tested with three projects. Briefly, the "Loose Strings" project was a single-camera widescreen DV shoot, and I rendered one song about 4 minutes long to the YouTube format using Final Cut Pro's Share option. I then created a QuickTime reference movie and produced the same file in Compressor with Qmaster enabled. The "Recital" project involved about 10 minutes of video shot in AVCHD at 1080i and rendered into H.264 format for Blu-ray and MPEG-2 for DVD using Final Cut Pro's Share option.
In 64-bit mode, Snow Leopard was faster than Leopard in all four trials, though the difference was slight in all tests except for the recital to MPEG-2 encoding, where Snow Leopard reduced encoding time by 27 percent. Overall, Snow Leopard produced an average 10 percent reduction in encoding time in all tests. In 32-bit mode, the results were less impressive as Snow Leopard was slower than Leopard in two of four teststhough on average, Snow Leopard reduced encoding times by 3 percent.
With the dual-core MacBook Pro, Snow Leopard's results were inauspicious, especially if you're using Qmaster, where Snow Leopard slowed encoding by 44 percent in 64-bit mode and a whopping 531 percent in 32-bit mode (I ran the 32-bit test multiple times to be sure the time was correct). I have no explanation for the results, but if you're running Snow Leopard on any dual-core computer and encoding using Qmaster, you should experiment with 32-bit and 64-bit modes to determine if your results are consistent with mine.
Looking at both tables, if MPEG-2 is a relevant format for you, then the $29 you have to spend on Snow Leopard may deliver significant bang for the buck, particularly on an eight-core computer running 64-bit mode. In addition, it appears that 64-bit mode is generally more efficient than 32-bit mode, so you should run some comparative testing and see if it's faster for you, and to determine whether you can live without any 32-bit peripherals while rendering. All it takes to revert to 32-bit mode is a reboot, so it may be worth your while to edit and render in 64-bit mode, then return to 32-bit to print, scan, or perform other tasks with 32-bit peripherals.
Not to beat a dead horse, but the difference in encoding time between the first two tests convinces me that producers on a deadline should eschew Final Cut Pro's new Share encoding option. As noted, these tests compared two roads to producing the same fileone using Final Cut Pro 7's new Share option and the other encoding a QuickTime reference movie in Compressor with Qmaster enabled.
The difference in encoding time is so dramatic that it makes me want to go back and retest (don't worry, I did). True, I didn't include the time it took to create the QuickTime reference movie in the calculation because it took only seconds. Still, overall, on both computers, the disparity in encoding time could easily be the difference between making a deadline and missing it by a mile.
Given the title of this newsletter, I was unsure if I should include data on other programsparticularly Premiere Probut I felt that more data was better for those trying to predict how Snow Leopard will perform with programs other than Final Cut Pro.
I produced three projects on the Mac Pro with Premiere Pro: a 14-minute, three-camera, mixed-format shoot (two HDV, one AVCHD) rendered to MPEG-2 and Blu-ray-compatible H.264; a dual-DV camera shoot rendered to MPEG-2; and a 4-minute, single-camera AVCHD shoot rendered to MPEG-2. Though Snow Leopard proved faster than Leopard in 32-bit and 64-bit modes, the difference was modest and a bit scatteredparticularly in 64-bit mode where the disparity ranged from 11 percent faster to 18 percent slower. I wouldn't swear off Snow Leopard if I'm a Premiere Pro producer, but I wouldn't expect a significant across-the-board bump in performance either.
As you can see in tables 4 and 5, I also tested performance with Sorenson Squeeze (MPEG-2/H.264) and Flix Pro (VP6) on the Mac Pro and MacBook Pro. On the Mac Pro, the performance difference was very modest with the exception of H.264 encoding, which dropped by 10 percent.
On the MacBook Pro, Snow Leopard rocked, with an average 15 percent reduction in encoding time in 64-bit mode and 13 percent in 32-bit mode.
Given the disparity of these results, it's tough to generalize them. For example, if you're producing H.264 video files with Squeeze on a Core 2 Duo computer, you should run, not walk, to buy the upgrade. But whether you'd see the same performance boost with Telestream Episode Pro or Episode Engine is anyone's guess.
In eight of 10 groups of tests, Snow Leopard was faster than Leopard, but in one of the two instances where Snow Leopard was slower, the difference was dramatic. Overall, I would guess that most producers can install the update and expect to see very modest speed increases, but you definitely shouldn't expect significant, across-the-board performance boosts. You should also experiment with 32-bit and 64-bit modes to see which delivers optimum performance with your typical projects.