FASTER AME Render Times For FREE!

[Steve Kalle]

A little while ago, I ran some quick tests to see how quickly I could get a DVD rendered that had a deadline in 12 hours but my i7 920 PC was going to take 14 hours to encode and I didn't have time to go to my studio PC (12-core) to make the DVD in the morning before the meeting with my client.

The video used was: 1080/30p 35Mb XDCAM EX, 25Mb 1080/60i HDV (in an EX1r) and 1080/30p 50Mb L-GOP from my nanoFlash (on my EX3). I had to use the 25Mb/s mode due to a 32GB SxS being lost by my client (thankfully, it was his card). I also used the main camera (nanoFlash EX3) as a 4th angle when editing with the scale increased slightly above 100% to give the allusion of a 4th camera (clients love this).

I originally edited in a 720p sequence and exported to AME as MPEG2-DVD which was going to take too long to encode. So, I ran some tests by nesting the 720p sequence in a 480p sequence as well as MRQ and RMD on & off. I exported a small section that included cuts with all 4 angles, probably 60s long but I can't recall. Using a GTX260, hardware acceleration was enabled.

To be pithy, here are my results:
6:37 - Sequence=720p / MRQ / RMD
4:58 - Sequence=720p / RMD
1:50 - Sequence=720p
3:23 - Sequence=480p / MRQ / RMD
1:51 - Sequence=480p / RMD
3:05 - Sequence=480p / MRQ

I brought each clip into AE, roto'd the main speaker on each clip and checked the quality in various portions of the clip.

I found NO discernible difference in image quality & sharpness between 720p & 480p and MRQ on or off. Because I had a considerable amount of CC'ing and effects applied to each camera, I found that RMD provided a slightly better image and color reproduction, but I forgot to run the export with 480p and RMD off. Because I can determine that 480p with MRQ and/or RMD is faster than the equivalent 720p, I decided to use 480p without RMD/MRQ to greatly reduce the encoding time.

[Jay West]

Did you possibly mean "MBD" rather than "RMD" options?

For the acronymically challenged, MRQ = "Maximum Render Quality." I thought "RMD" was the acronym for "Red Metadata." It would never have occurred to me to enable it when working with non-Red files like the XDcam and Nano files you are using. I didn't even know that there was such an option in AME ("Adobe Media Encoder.")

That's why I am wondering if, possibly, you were referring to "MBD" meaning "Maximum Bit Depth" rather than RMD?

If so, I did some similar tests a year or so ago, and, like you, basically found that I could see no visual difference between encodes using MBD and MRQ when coming from PPro sequences where I had enabled hardware MPE ("Mercury Playback Engine.")

Checking the MBD and MRQ options only seemed to make rendering and encoding take a muuuuuccchhhhh longer time. My recollection is that somebody (perhaps Randall Leong) figured out that checking these options when exporting from a PPro Hardware-MPE sequence would basically force the processing into software-MPE mode and greatly slow things down. My understanding is that hardware MPE basically gives you automatic MRQ. Maybe I am wrong on this?

Anyway, here is a slightly different question. Are you finding that it will be appreciably faster to (a) nest an HD sequence into a 480 SD sequence and export from there versus (b) simply exporting directly from the HD sequence (or even dynamically linking to Encore from there and burning the project to an ISO image)? While I know that AME does much faster encoding from 480 SD projects, I have assumed that somewhere along the line PPro, AME or Encore would need to render the HD footage to SD, so that the total time savings would be negligible. Is it the case that PPro's format agnosticism allows a speedier export and encode simply by nesting an HD sequence into a 480i or 480p SD sequence?

[David Trayers]

I haven't done extensive testing of all the combinations myself, but I have found that when going from 720p to 480SD for a DVD export, it's a lot faster when I put the 720p timeline on a SD sequence and check the option to scale the video to fit. I found that AME will use the GPU a lot more during export.

When exporting directly from a 720p timeline to DVD, it hardly uses the GPU at all.

So now my workflow is to do all my editing in the 720p sequence, nest it in a SD sequence, scale the video to fit, then export with AME to a MPEG-DVD.

I didn't see any difference in video quality, but I didn't have any titles or captions... I suppose I should make a test for those....

[Steve Kalle]

Hey Jay,

I was correct with RMD - take a gander at AME and you will see the box for "Render at Maximum Depth" (RMD).

In case people reading this do not know what RMD does, here is my quick explanation. When checked, it uses 32 bits per channel Floating Point math to calculate effects rather than the standard 8bits when not checked. If you know about After Effects 32bpc Floating Point compositions, then you will understand RMD in PPro. In AE, the main reason to use 32bpc is for Superwhites and the other reason is 32bpc provides more of a realistic rendering of color and luminance. The best example I have seen of this realistic rendering is when someone filmed a car in front of a green screen and keyed in a night scene for the window. With 8bpc or 16bpc comps in AE, the street lights thru the window were blown out whereas the 32bpc comp had perfectly rendered lights. Here is the link on the Tutorial VIDEO COPILOT | After Effects Tutorials, Plug-ins and Stock Footage for Post Production Professionals

With Premiere, I find RMD slightly increases the quality of color rendition when I have several effects that alter color and/or luminance in some way.

[Steve Kalle]

Originally Posted by Jay West

Did you possibly mean "MBD" rather than "RMD" options?

For the acronymically challenged, MRQ = "Maximum Render Quality." I thought "RMD" was the acronym for "Red Metadata." It would never have occurred to me to enable it when working with non-Red files like the XDcam and Nano files you are using. I didn't even know that there was such an option in AME ("Adobe Media Encoder.")

That's why I am wondering if, possibly, you were referring to "MBD" meaning "Maximum Bit Depth" rather than RMD?

If so, I did some similar tests a year or so ago, and, like you, basically found that I could see no visual difference between encodes using MBD and MRQ when coming from PPro sequences where I had enabled hardware MPE ("Mercury Playback Engine.")

Checking the MBD and MRQ options only seemed to make rendering and encoding take a muuuuuccchhhhh longer time. My recollection is that somebody (perhaps Randall Leong) figured out that checking these options when exporting from a PPro Hardware-MPE sequence would basically force the processing into software-MPE mode and greatly slow things down. My understanding is that hardware MPE basically gives you automatic MRQ. Maybe I am wrong on this?

Anyway, here is a slightly different question. Are you finding that it will be appreciably faster to (a) nest an HD sequence into a 480 SD sequence and export from there versus (b) simply exporting directly from the HD sequence (or even dynamically linking to Encore from there and burning the project to an ISO image)? While I know that AME does much faster encoding from 480 SD projects, I have assumed that somewhere along the line PPro, AME or Encore would need to render the HD footage to SD, so that the total time savings would be negligible. Is it the case that PPro's format agnosticism allows a speedier export and encode simply by nesting an HD sequence into a 480i or 480p SD sequence?

Yes, exporting to a DVD from a 480p sequence is faster than 720p and 1080p sequences. My very first test was exporting to 480p MPEG2 for TV and I found that a 720p sequence was faster to render than a 1080p sequence.

However, I discovered something odd when exporting to SD MOV for broadcast with text in the video (in this case, created everything in AE and DL'd into PPro to add audio and render). The quality of the text was much lower when exporting from 720p versus exporting from a 480p sequence. (Its been a while so I might have this backwards) Exporting to MPEG2 didn't have this large discrepancy in text quality; thus, I always use MPEG2 for SD broadcast TVCs.

[Jay West]

I can't say I've ever run into any problems with print when exporting from PPro to Quitime MOV files from PPro, but I was intrigued to learn of the RMD setting and also intrested to see how well the "nesting" trick might work for me. I made some quick tests when I had a little time this morning. What I discovered for my system was:

(a) the “nesting” trick had a pretty dramatic positive effect when used with dynamic linking from PPro to Encore but only slightly sped up an MPEF2-DVD export via AME directly from the PPro timeline;

(b) RMD imposes only a slight time penalty but --- subjectively, at least --- did not seem to make any difference to an export of either the HD or nested SD files. In my tests, the DVD video that came directly from the HD sequence seemed a bit sharper and cleaner than that from the nested-into-SD sequence, but the difference was subtle and probably would not be noticed by most viewers.

(c) I had assumed that I would need to render the re-scaled video when nested, but that proved unnecessary, There was only a tiny reduction in processing time if I rendered the SD sequence before exporting to MPEG2-DVD.

Obviously, these specific results are only accurate for my particular set-up and the specific video I used in my tests and the effect of RMD (or lack of it) was a purely subjective assessment.

For those who, like me, have never previously noticed the AME check-box for “render at maximum depth,” you can find it when you export from PPro as well as when you open AME directly. The export has to be set for an MPEG2 DVD export. You then look down to “basic video settings,.”and scroll down a few lines to find it. Interestingly, the RMD option does not seem to be available when accessing AME from with Encore after dynamically linking (DL) the timeline to Encore for a DVD project.(I do have Encore set to use AME for encoding.)

Now, for my test results. I used a 3 minute segment from a dance recital I am editing in CS 5.5. (I have not yet had time to install CS 6, yet, and do not want to switch in the middle of a project, anyway. The relative results will probably be the same under CS6, in any event.) This video was shot with three HDV and three AVCHD camera, all at maximum quality settings. All footage was 1080i. (I use that because that is what my HDV cams shoot and because I am editing for DVD, in any event.) The edit sequence was a 1920x1080i multi-cam sequence. Color corrections were applied to all six camera tracks using mostly MPE-enabled effects. (The reason for all the color correction was that a mis-programmed stage lighting controller panel put a blue cast on everything; the video tracks were adjusted in post to give a more natural looking appearance.) There were several PiP and split screen effects (you do that when you’ve got groups of dancers moving to opposite ends of a very large stage and the customers want to see their kids’ faces in the video). Several clips were digitally adjusted for tighter framing. There were two titles. Seemed like this would provide a pretty good test for this subject.

For the tests, the AME export settings were:
NTSC MPEG2-DVD
2 pass-VBR
Quality = 5
MRQ = not enabled
Bitrates = 8 mb max, 6 mb target, 4 mb min.

I started with the rendered, edited 1920x1080i timeline, doing one export with RMB enabled and one without. Then, I copied the segment into a 480i-widescreen SD segment and applied “scale to frame size.” I exported from the unrendered 480 timeline with and without RMD. (I clicked Sequence-DeleteWorkAreaRenderFiles each time to be sure that there were not any secretly rendered files lurking about.) Finally, I rendered the scaled 480 timeline and repeated the exports with and without RMD enabled. Rendering the scaled segment before export took 2:01.

Export timings were started when the progress bar first appeared in the “encoding Sequence” box. Timings stopped when the box disappeared. I ran the tests three times, rebooting between each set, and averaged the results. (I did this because Windows 7 sometimes seems to run faster than it does on reboot and vice versa.) Here are the numbers I got:

Export from 1920x1080i HD timeline to MPEG 2, RMD enabled 4:08
Non-RMD export from 1920x1080i timeline 3:52

Nested to 480i, scaled but unrendered, RMD enabled 3:49
Nested to 480i, scaled but unrendered, (no RMD) 3:36

Nested to 480i, scaled and rendered, RMD enabled 3:49
Nested to 480i, scaled and rendered, (no RMD) 3.34

HD dynamically linked to Encore, time to encode and build ISO 4:53
HD in SD sequence, dynamically linked, time to encode and build ISO 3:03

One final note: I did not have time to try this experiment with any 720p sequences. I rarely shoot in that format because I often shhot multi-cam and only some of my cameras will generate 720p.. But, if I have time this weekend, I may experiment a bit with progressive footage and/or 720 sequences to see if that makes any difference to the results I got.

[Bruce Watson]

Originally Posted by Steve Kalle

I brought each clip into AE, roto'd the main speaker on each clip and checked the quality in various portions of the clip.

Going to AE over dynamic linking is a performance killer at export time. The document called Dynamic Link and After Effects, way at the bottom, says: A linked After Effects composition will not support Render Multiple Frames Simultaneously multiprocessing. Which is just ridiculous -- this is 2012! This is brand spankin' new CS6 we're talking about!

What this means in practice is that AE can reduce your highend 8 processor machine to a single processor machine for the duration of the export, for the frames that AE has to touch. The performance hit is enormous. Just a couple of weeks ago I exported a 3.5 minute clip -- which took 16 hours to render out on my i7 930, with AE limiting the eight hyper-threads to 20-30% while using about 2/3 of the available memory and disk drives that were mostly idle. Unreal bad performance. Because of AE's bad behavior over dynamic links.

I put in a feature request to get this fixed. The more requests they get, the more it moves a request up the priority ladder. I suggest anyone interested in export performance, AE, and dynamic links, make a similar feature request.