Split thread from marks430 re TMPG

[marks430]

[ddogg edit: Thread split from original as it was off topic and I replied without noticing. My bad.)

@DDogg

I don't have the problems with CCE crashing, but I saw your earlier post to mkanar about testing with tmpgenc and I did a little test on my own using 1.0.7 build 1.

I have a dual 1GHz P3 with 512MB PC133. I wanted to see the difference in encoding times between the CCE and tmpgenc since it supports multi processors. I did an encode of a few chapters using both methods (CCE 3 pass VBR) (TMPGEnc CQ) and there was very little difference in the total SVCD creation times. The length of the chapters encoded was about 11 minutes. Using CCE, DVD2SVCD finished in 55 minutes 7 seconds. Using TMPGEnc, DVD2SVCD finished in 55 minutes 39 seconds. Processor usage during CCE was between 50 and 55 percent. Processor usage during TMPGEnc was mostly at 100 percent.

I haven't had a chance yet to burn the resulting images for quality comparison, but the bin file sizes are considerably different. CCE 207,516 KB and TMPGEnc 155,499 KB.

[DDogg]

marks430, while I know testing is a major pain in the backside I would suggest you do a full encode. I would expect that a few chapter encode using TMPG CQ mode used CQ 100. That would not give a fair comparison as the object is to place an extended amount of video on a disk. The CQ value is altered via D2S's compression tests to "fit" the encode on the amount of disks select (preferably 2 disks for test purposes). If the encoded source is a short test chapter the resulting calculated CQ will not be indicative of the full encode using TMPG.

If you do get the chance to run the full test please consider using these test settings:
http://forum.doom9.org/showthread.php?s=&threadid=18665

[da franksta]

cce and tmpgenc BOTH support dual processor systems. CCE is faster. if you want more details let me know.

[chainsaw135]

Yes if you us them as stand alone encoders they will both use the full cpu. If you us the avisynth script which dvd2svcd uses then cce will only use half of the cpu usage.

[da franksta]

technically, even if you use avisynth cce still uses both cpu-s. you don't notice this 'cos it is held back dramatically by avisynth. avisynth, of course, does NOT support dual cpu-s.

[chainsaw135]

"technically it uses both cpu's" thats fine, but with it going through the avisynth script you will not get the same performance as if you just went straight through cce. I mean i wasent trying to teach a course in technicality, just wanted mark to understand that avisynth was his bottleneck.

[marks430]

@DDogg

I have no problems doing the test as you suggested. I just downloaded and installed build 3 and have set all the settings as you suggested. Once the encode is complete, I will post back here with the particulars of the encode.

The encode should complete sometime during the night so I will post the results in the morning. Any particular info about the encode that might be useful to anyone?


@da franksta & chainsaw 135

I understand that the bottleneck in the encode using CCE is actually avisynth. I have noticed that when doing encodes using DVD2SVCD and CCE, I have less than 55 percent processor utilization. Since the latest version of DVD2SVCD supports TMPGEnc I thought I would test the speed and quality in comparison to CCE. During this test, I had 100 percent processor utilization. But if the speed of the encode isn't any faster, I probably will not use this method since I am very satisfied with the quality using CCE and 3 pass VBR.

Thanks for the info though.

Mark

[chainsaw135]

What i've noticed is that tmpgenc is getting better then ever,it still has the problem with blocks in certain parts of a movie that are kind of annoying. Tho cce has issues nothing like those blocks i see in tmpenc, time wise is not an issue with me never has been just the quality. I'm looking forward to the tests and results you post.

[Abend]

On my "Test machine" I can monitor each app and CPU it uses.
Avisynth always uses one CPU, actually in multiproc you're even loosing some time because tread is switched from CPU to CPU by OS
No way to multitread even with quite sophisticated tools that came with M$ datacenter and Unisys (the maker of "testbench") :-)
CCE scales to 2 CPUs on its own and can be made run on more with abovementioned tools. When frameserving with Avisynth, CCE basically stays idle most of the time and waits for input
TMPG before 2.5 scales just great up to 32 CPU
Can't wait for 64 Itaniums that should arrive in 3 month time :-)

[chainsaw135]

Yea it will be nice when we can back up a dvd of like 2 hours in less then 15 minutes:) Kinda like we do with our pc games or programs.

[DDogg]

Can't wait for 64 Itaniums that should arrive in 3 month time :-)

Ok, just rub it in! LOL

[DDogg]

@marks430

Any particular info about the encode that might be useful to anyone?

Just the stuff mentioned at the bottom of:

http://forum.doom9.org/showthread.php?s=&threadid=18665

[Abend]

Originally posted by DDogg


Ok, just rub it in! LOL

It would be itchy :-)

[marks430]

The TMPGEnc encode has completed. I encoded the NTSC region 1 anamorphic DVD "Joe Dirt" from VOBs previously ripped using internal routines. The length of the video was 1:30:53.

Total time for SVCD creation, 8 hours 33 minutes 23 seconds.

Encoded at aspect ratio 16:9
Force Film was activated.
Audio set to 192 with no downsample.
Resize method set to Bicubic with b value 0.00 and c value 0.55
Max bitrate set to 2520, Min set to 755 and Max average set to 2230
Between 61 and 100 minutes use 2 CD of 800 MB.
TMPGEnc Rate Control Mode set to Constant quality (QC)
Motion search precision set to Highest quality
DC Component Precision set to 10
Use floating point DCT and No motion search... both checked
Soften Block Noise unchecked

Video Quality Value Used was 85.51

Ended up with 2 bin/cue files.
CD1 was 825,644 KB and CD2 was 796,515 KB

I am currently doing the same DVD encode using CCE and 4 Pass VBR with my usual settings. I will post those results when it is done.

I have saved all files from this encode so if anyone wants additional information, let me know.

Mark

EDIT: Results also posted in http://forum.doom9.org/showthread.php?s=&threadid=18665

[DDogg]

Your coherent, well ordered report is MUCH appreciated. I am sure we are all looking forward to the next installment.

EDIT: Please post results in:
http://forum.doom9.org/showthread.php?s=&threadid=18665

[marks430]

The CCE encode is now finished. I encoded the same VOBs as above. NTSC Region 1 "Joe Dirt" anamorphic.

Total Time for SVCD creation, 8 hours 47 minutes 48 seconds.

Encoded at aspect ratio 16:9
Force Film was activated.
Audio set to 192 with no downsample.
Resize method set to Simple Resize
Max bitrate set to 2530, Min set to 300 and Max average set to 2230
Between 61 and 100 minutes use 2 CD of 800 MB.
Video Encoding Mode set to Multipass VBR 4 Pass
Image Quality Priority set to 17
Anti noise filter checked and set to 2
Linear quantizer scale checked
Upper field first checked
Zigzag scanning order checked
Progressive frames checked

Average Bitrate 2190

Processor Usage during CCE was around 65 percent

Ended up with 2 bin/cue files
CD1 was 825,575 KB and CD2 was 826,577 KB

While this one was cooking, I burned the images from the TMPGEnc encode and I have to admit, they looked pretty good, very little difference from the ones I had done before at 3 pass VBR. TMPGEnc could be a viable alternative to those who experience the CCE crashes.

I do like the fact that with CCE, I still have some processor left over to surf the web, check email, etc., without significantly slowing down the process. When TMPGEnc was running I had 100 percent processor usage so any other use of the computer would be stealing time away from the encoding.

I will keep these files as well (for a while) so if you want more info, let me know.

Mark