Ernest Allen
June 13th, 2006, 10:10 PM
Though I’ve had a couple of years experience editing DV in Final Cut Pro, I wasn’t altogether prepared for the extensive time needed to transcode 1080i60 HDV projects to the H.264 codec—despite warnings contained in some of the threads which I had read at DV Info.net. I had figured on an increase in time by at least a factor of six, given the corresponding difference in pixel counts between DV and 1080 HDV. So when I purchased an XL-H1 from TapeWorks Texas a couple of weeks ago, I jumped right into videotaping a local lecture that lasted about an hour and a quarter, and commenced to cutting/ rendering/ transcoding it via Final Cut Pro on a dual processor 2.5 Ghz G5. I captured the file from the XL-H1 into the Apple Intermediate Codec, and dropped it into an AIC sequence. Just to make things interesting I added a handful of titles with Motion, punched up the brightness of the clip a bit, and kicked in the “Broadcast Safe” filter to clamp the brightest whites. But after sending the sequence directly to Compressor from FCP in order to generate an H.264 file, I was hit with sobering news from the Batch Monitor that it would take around 700 (seven hundred) hours to process this 75-minute sequence! Not possible, I thought; I must be doing something terribly wrong. To make a very long story short, after numerous aborted tries and bouts of self-doubt (e.g. Compressor seemed to restart itself again and again, leading me to the erroneous belief that it was stuck in a loop), I stripped a simple five-minute section of the clip of all filters and proceeded to document the transcoding progress in order to superficially understand how Compressor generated H.264 files. I’d like to now share that experience with DV Info readers and hopefully learn from others what experiences they’ve had with H.264 while working with FCP as well as with other platforms.
First of all, on the dual processor 2.5 Ghz G5 it took Compressor 6 hours and 15 minutes to transcode a 5-minute clip from an AIC timeline to H.264: a 75/1 ratio! What I discovered, however, was that Compressor, with its stock setting for producing a 60-minute H.264 movie, required four passes to achieve its goal. At 30 frames/second, five minutes yields 9,000 frames. The first pass took about 1 hour 20 minutes, processing an average of about 112 frames/min. The second took 1 hr 47 minutes, at a rate of about 80 frames/min. The third was similar to the second: about 1 hr 46 minutes at 85 frames/min. Finally, the fourth pass was similar to the first, taking about 1 hour 22 minutes at about 109 frames/min. The final pass was also the most interesting for two reasons: it produced the widest variation in frames/min (from around 111 to 128 at any given moment), and around every 28 frames the frame count would flash the term “audio.”
Six hours and 15 minutes is one hell of a long time to transcode a 5-minute clip. But what is really maddening about FCP 5 is that the H.264 setting gives offers a “time left to process” and percentage-of-work-completed numbers that bear little resemblance to the real figures at any given time—except at the finish line. Shortly after the beginning of the first pass, Compressor indicated that 14 hrs 10 minutes remained. At the beginning of the second pass an hour and 20 minutes later, the indication was that about 12 hours remained, and that Compressor had completed 10% of the job. One hour 40 minutes later, at the beginning of the third pass, 11 hours 16 minutes officially remained, with 21% of the work completed. At the start of the fourth pass, at about an hour 46 minutes later, around 10 hours were purportedly left, and 40% of the job done. Thereafter, the times rapidly fell to zero and the percentages rapidly increased to 100%, offering a welcomed sense of relief. The movie looks great.
Now, it is possible to create a custom setting in Compressor based on its 60-minute H.264 default setting, and then to uncheck the multi-pass box to significantly reduce transcoding times. I haven’t yet done this to compare the quality between one-pass and multi-pass H.264 movies, however, so don’t know yet what I might find.
Based on my experience with the 5-minute clip, I figure that my entire 1 hour, 15 minute sequence will take four non-stop days to transcode. To lose the services of my fastest computer for this long will require a juggling of projects—perhaps even a short vacation on my part. I’m definitely looking forward to FCP 6 in hopes of some speed improvement, praying for a quad-processor Macintel desktop machine to show up soon in the marketplace, but am also most interested to learn how many forum people are using, for example, Sorenson Squeeze on existing Macs to generate H.264 files, or how other platforms are generally behaving vis-à-vis H.264. It is also possible, of course, that there is something egregiously wrong with my workflow, in which case I’d be happy to hear what others have to say about that.
First of all, on the dual processor 2.5 Ghz G5 it took Compressor 6 hours and 15 minutes to transcode a 5-minute clip from an AIC timeline to H.264: a 75/1 ratio! What I discovered, however, was that Compressor, with its stock setting for producing a 60-minute H.264 movie, required four passes to achieve its goal. At 30 frames/second, five minutes yields 9,000 frames. The first pass took about 1 hour 20 minutes, processing an average of about 112 frames/min. The second took 1 hr 47 minutes, at a rate of about 80 frames/min. The third was similar to the second: about 1 hr 46 minutes at 85 frames/min. Finally, the fourth pass was similar to the first, taking about 1 hour 22 minutes at about 109 frames/min. The final pass was also the most interesting for two reasons: it produced the widest variation in frames/min (from around 111 to 128 at any given moment), and around every 28 frames the frame count would flash the term “audio.”
Six hours and 15 minutes is one hell of a long time to transcode a 5-minute clip. But what is really maddening about FCP 5 is that the H.264 setting gives offers a “time left to process” and percentage-of-work-completed numbers that bear little resemblance to the real figures at any given time—except at the finish line. Shortly after the beginning of the first pass, Compressor indicated that 14 hrs 10 minutes remained. At the beginning of the second pass an hour and 20 minutes later, the indication was that about 12 hours remained, and that Compressor had completed 10% of the job. One hour 40 minutes later, at the beginning of the third pass, 11 hours 16 minutes officially remained, with 21% of the work completed. At the start of the fourth pass, at about an hour 46 minutes later, around 10 hours were purportedly left, and 40% of the job done. Thereafter, the times rapidly fell to zero and the percentages rapidly increased to 100%, offering a welcomed sense of relief. The movie looks great.
Now, it is possible to create a custom setting in Compressor based on its 60-minute H.264 default setting, and then to uncheck the multi-pass box to significantly reduce transcoding times. I haven’t yet done this to compare the quality between one-pass and multi-pass H.264 movies, however, so don’t know yet what I might find.
Based on my experience with the 5-minute clip, I figure that my entire 1 hour, 15 minute sequence will take four non-stop days to transcode. To lose the services of my fastest computer for this long will require a juggling of projects—perhaps even a short vacation on my part. I’m definitely looking forward to FCP 6 in hopes of some speed improvement, praying for a quad-processor Macintel desktop machine to show up soon in the marketplace, but am also most interested to learn how many forum people are using, for example, Sorenson Squeeze on existing Macs to generate H.264 files, or how other platforms are generally behaving vis-à-vis H.264. It is also possible, of course, that there is something egregiously wrong with my workflow, in which case I’d be happy to hear what others have to say about that.