2.3.13 BETA Conversion speeds

Jun 5, 2013 at 4:07 AM
Hello all!

I was wondering if anyone had looked into average conversion speeds with the latest beta builds. I understand the rboy1 has done a great job with optimizing the profiles to be speedier, but for some reason my system isn't really seeing the 25% to 50% increases in the profiles I use (MP4 Normal & Fast, MKV High). Here's a breakdown:

I am running Windows 8 on an Athlon II X4 640 chip @ stock clocks, 4GB DDR3-1600 RAM, AMD HD6570 video card with the latest Catalyst drivers. I am converting MPEG2 WTV files (Comcast isn't sending H.264 streams in my area yet).

Also note, when I am running the latest Handbrake direct download, it is running ~25% faster on average for a given file than the version that is bundled with MCEBuddy (just the Handbrake portion, ignoring the comm removal, metadata, moving, etc. etc.).

So, I guess this all comes down to: is my system simply ridiculously underpowered processor-wise (I know the video card can handle the transcode, as it never goes over 20% usage during any conversion), or is there a setting/filter/codec pack/driver that will help me see the awesome gains rboy is working so hard for?
Jun 5, 2013 at 6:05 AM
Yeah, I downloaded the latest beta from the Early Access folder about a week ago and noticed my conversions are not at all faster. In fact, they seem to take longer. I have an i3-2100 on a Windows 7 machine. It takes MCEBuddy about 3.5-4 hours to convert a 1 hour HD TV show using the MKV High Quality profile. The only change I've made to the default profile is to add "FixedResolution=true".
Coordinator
Jun 5, 2013 at 7:10 AM
@Wilky that actually makes a huge difference in teh profile processing time, FixedResolution will increase the BitRate and resolution which takes longer to encode.

The performance increase I had measured for the latest 2.3.13 BETA is v/s 2.3.12 on my iCore7 system (using 1 concurrent conversion for benchmarking).

Also point to note, in the Latest BETA the profiles are tweaked so that Normal and Fast and SIGNIFICANTLY faster than High Quality. here is a rough comparison,

Fast - x minutes
Normal - 2x minutes
High Quality - 4x minutes

ie. High is 4 times slower than Fast. I ONLY recommend High Quality for OUTPUT video which are > 720 pixel width (i.e HD output NOT HD input). For 720p and lower OUTPUT I would recommend Normal is good enough.
In the latest BETA the HighQuality profiles really tweaked for MAXIMUM HD quality at the EXPENSE of speed. If you want a good balance use the Normal profile, for pure speed us the Fast profile. (all this as of the latest BETA).

@belatukadro - what version of Handbrake are you using?


Jun 5, 2013 at 6:09 PM
Edited Jun 5, 2013 at 6:11 PM
Thanks for that information, it is very helpful. And boo on testing on a Core i7, of course that thing is gonna fly :-P I'll probably invest in Haswell for my rebuild this winter, assuming that Kaveri isn't comparable for video transcoding (considering that they'll finally have hUMA running, it very well might, and likely at a much lower cost).

I'm using Handbrake version 0.9.8 (2012071700), using all the default Normal settings. Also, single conversion jobs only on MCEBuddy, my poor little quad core can barely handle that let alone concurrent jobs. I'll do a test tonight with a sample .ts file -> .mp4 that's a clip from one of my unprocessed .ts commercial cuts on both the latest MCEBuddy 2.3.13 and the standalone handbrake, compare them, and share a link to the log files here. I have a feeling that for some reason the standalone is utilizing OpenCL acceleration while the version included with MCEBuddy is not, but why would be a huge question. I know you tweak your version somewhat, but the underlying conversion engine should work the exact same, right?

As for the output profiles, I will try tweaking my conversion profile to use MP4 Fast @ 720p max width. If I disable cropping, will that fix the resolution & bit rate the same as using FixedResolution=True, or simply skip the crop step in the conversion process? All of my recordings are 1080i, so I shouldn't need to crop them to scale properly to 720p.
Coordinator
Jun 6, 2013 at 1:10 AM
Fixing resolution has different meanings

FixedResolution=true keeps the original resolution and defined profile bitrate.

Skip cropping skips cropping video

Max width limits the maximum width but adjusts the bitrate to compensate for the new width. The bitrate specified in the profile is baselined at 720 pixel width. If the output resolution is higher than 720 then the bitrate is scaled up appropriately, if the output resolution is set to lower then it is scaled down appropriately. If the output resolution is 720 then the profile defined bitrate is used.

All this as of the latest BETA

Jun 6, 2013 at 2:43 AM
Thanks for the explanation, it should help things greatly.

Well, I performed the test conversion this evening, and was floored by the results. I decided to go ahead and update the AMD Catalyst drivers to the 13.4 version that went live last week, and the difference is, in a word, dramatic. The standalone Handbrake converted a 15m file in ~13m. The MCEBuddy 2.3.13 MP4 conversion took 5 minutes TOTAL. It looks like the issue was the MCEBuddy version not being hardware accelerated, as the average framerate before the update was ~35fps, while after the update it was ~88fps.

Moral of the story: always at least try the newest drivers kids. You can always rollback if necessary.

Thanks again for the phenomenal work you've done already, and I really look forward to what goodies you have in store for us.
Jun 6, 2013 at 6:31 PM
Edited Jun 6, 2013 at 6:31 PM
Well, I spoke too soon. It would seem that the 88 fps log was a complete anomaly, and I am back down to 40 fps during ~10 more tests, with the original test files and others.

New moral: one test does not a conclusion make. :-(

Also, I noticed that in the MKV Normal and MP4 Normal (at least), even when cropping is disabled mencoder is still scanning for crops. It only takes ~5 minutes or so, but if that data isn't being used for anything, it is still time and power wasted.
Coordinator
Jun 7, 2013 at 2:18 AM
Speed improvement depends heavily on the number of commercials.

If there are no commercials, just conversion it wil be a 5% to 10% improvement due to tweaking of profiles
If there a lots of commercials you will see a 30% to 50% improvement depending up on the amount of commercials sicne now commercials are removed BEFORE the conversion.

Regarding the skipping cropping, yes you're correct it's a bug I'll fix it in the next beta build tomorrow


Jun 8, 2013 at 4:20 AM
I do care about quality of the output file. If all my input files are HD recordings, how much of a difference will I notice between the MKV Normal and MKV High profiles in the quality of the output file?
Coordinator
Jun 8, 2013 at 5:48 PM
AGAIN, please read my comments carefully. INPUT DOES NOT MATTER, output resolution does. If you output resolution > 720 pixels then it may make some difference, if it is 720 or lower there will be no difference.