QuEnc^n - The multicore frontend for QuEnc v0.9 (Beta)

[Darksoul71]

Hi all,

since I had two weeks vacation there was some time to work on QuEnc^n

Grab the latest version here:
http://rapidshare.com/files/46414427...n_v09.zip.html

If anyone wants to mirror QuEnc^n, feel free to do so.

I can´t really provide a detailed change log since QuEnc^n has undergone a lot of changes. Most of them were to make the encoding process more solid. I´ve added support for custom matrices being used during the encoding process as well as automated adding of AVISynth filters. Thanks to some input from Mr_Odwin over at doom9 (http://forum.doom9.org/showthread.php?t=115471) I was able to get rid of my crappy AVIInfo_CLI tool which I always used to determine the frame count of the source AVS.

For more details refer to QuEnc^n.txt included in the zipfile

Best regards,
D$

[danpos]

@Darksoul71

Well done, mate!

Kept up,

[Revgen]

I keep getting this error message

Line 0 (File "H:\Package\QuEnc^n.exe"):

If (UBound($PID)=0) And @Error Then Exit
If (Ubound(^ERROR

Error: Variable used without being declared

Is there something I'm doing wrong?

[foxyshadis]

Does it do any kind of 1-pass scene-change detection, like ELDER? Even if you just ran the pass on a cut-down script (down that path lies madness though) and on a short segment around where you already expect to cut, it would help set it to a scene-change. But then, ELDER works on many small segments, instead of a few large ones, so maybe it's more important for that.

[Darksoul71]

@Revgen: Which CLI did u use ? If no CLI was given, QuEnc^n will currently error out like this...RTFM

Originally Posted by QuEnc^n.txt

Q: What Commandline Interface (CLI) does QuEnc^n provide ?
A: Basically QuEnc^n doesn´t really have a CLI. It simply takes all CLI params and
"forwards" them to the several QuEnc instances.
The only true parsing that QuEnc^n does is looking for source and target file
because it needs to modify them to make the multiple instances run. QuEnc^n will
also add the "-auto" and "-close" parameter if they are missing since the QuEnc
instances would either not start encoding and/or not close after encoding.

@Foxeyshadis: I dunno ELDER. QuEnc^n simply does what described in the short segment of the ReadMe:
Segmentlength=Frames# Video / Core

[Revgen]

CLI?

Do I use the Commandline Parameter tool to use this?

[Darksoul71]

Originally Posted by Revgen

Do I use the Commandline Parameter tool to use this?

Is water liquid ?
...but feel free to code a GUI which translates the settings of QuEnc to CLI (basically replacing QuEnc GUI).

As I said: Its a simple AutoIt script. May be I´ll find some time in future to add a GUI but QuEnc^n was mainly coded as interface between any calling program (e.g. DVD-RB, DVD2VCD) and QuEnc. At least you are able to use it without something like an ECL generator

Originally Posted by foxyshadis

Does it do any kind of 1-pass scene-change detection, like ELDER? Even if you just ran the pass on a cut-down script (down that path lies madness though) and on a short segment around where you already expect to cut, it would help set it to a scene-change.

Ah, now I understand what you were talking about. Sry, my last post was 4:00AM local time

Originally Posted by foxyshadis

But then, ELDER works on many small segments, instead of a few large ones, so maybe it's more important for that.

This would be my initial impression. Since most people might likely use 2 or may me 4 instances of QuEnc for rather large movies this would result in long segment. Thus limiting the impact of a Segment end within a scene compared to a few hundered scenes within the whole segment.
Nevertheless that is a good point to consider an I´ll put it on my "improvment" list. Thanks.

[Revgen]

Originally Posted by Darksoul71

Is water liquid ?
...but feel free to code a GUI which translates the settings of QuEnc to CLI (basically replacing QuEnc GUI).

As I said: Its a simple AutoIt script. May be I´ll find some time in future to add a GUI but QuEnc^n was mainly coded as interface between any calling program (e.g. DVD-RB, DVD2VCD) and QuEnc. At least you are able to use it without something like an ECL generator

Okay, I wasn't aware of that. I thought it was a GUI frontend for Quenc. I think I know what it's for now. Thanks.

[Revgen]

Okay I've tested it.

It's definitely faster than a than a single thread.

Great Job!

I'll do a more complete test later.

[Darksoul71]

@Revgen:
Thnx, do you own a dual core system ?
I´m really anxious to see numbers on a dual core and a HT CPU with Dual QuEnc vs. Single QuEnc.

Thnx for testing.

[Revgen]

Yes I do.

It's a 4600+ X2 Dual-Core. It doesn't have HT though.

I'll do the tests once I get home.

[Revgen]

Okay here are the results. I timed these manually with a stopwatch.

Quenc 70: 1006 Secs 8.925 Total FPS

Quenc^n: 608 Secs 14.768 Total FPS


This is about a 65% increase.

Keep in mind that this was a 5-minute sample and the additional 30-secs of time for Quenc^n to mux the 2 videos may not affect feature length DVD's as much. I'm guessing a full DVD may give a performance boost into the 70% to 80% range.

EDIT:

Here are the settings I used

H:\Package>quencn -i J:\Progra~1\Dscaler\laker.avs -o J:\Progra~1\Dscaler\laker2
.m2v -b 4500 -maxbitrate 8000 -dc 9 -priority 3 -2 -hq -vbr -noscene -notrell -n
ocgop -interlaced -noextreme -gopsize 18 -maxbframes 2 -noqlb -nocmatrix -aspect
ratio 16:9 -tff

[danpos]

@Revgen

Impressive! This makes me think that I've to buy my MOBO+DualCore CPU ASAP!

@Darksoul71

Hmm, I hope that this result give to you a new stimulaton and so you can to take a look at QuEnc^n 'cousing' - LAN_AQE-RB

Kept up,

[Darksoul71]

@Revgen:
Thank you very much ! Now I have some "true numbers". If you have the time I would be happy if you could provide results for a longer video (> 60 min). TIA...

I´m also expecting something like 70-80%. In very rare cases (very fast dual core / dual CPU systems) I the speed increase could be 80-90% but 200% speed increase is basically unrealistic.

@danpos:
Do not forget that I am basically working on LAN_AQE-RB. Think in little functions and not the "big picture"

[Revgen]

Originally Posted by Darksoul71

@Revgen:
Thank you very much ! Now I have some "true numbers". If you have the time I would be happy if you could provide results for a longer video (> 60 min). TIA...

I´m also expecting something like 70-80%. In very rare cases (very fast dual core / dual CPU systems) I the speed increase could be 80-90% but 200% speed increase is basically unrealistic.

Yer Welcome.

I also was using a 29.97i interlaced source for the test since it was what I had available. 23.976p material could possibly bring greater benefits. I'll try a DVD movie tommorow and see what I can do.

[Revgen]

Okay I decided to use Star Wars Episode 1 WS (NTSC) for a 2hr test.

My Settings:

quencn -i J:\star\star.avs -o J:\star\star.m2v -b 4500 -maxbitrate 8500 -dc 9 -priority 3 -2 -hq -vbr -scene -notrell -nocgop -nointerlaced -noextreme -gopsize 18 -maxbframes 2 -noqlb -nocmatrix -aspectratio 16:9

Here are the results:

Quenc 70:

7815 secs

22.965 True FPS


Quenc^n:

4278 secs

41.952 True FPS


This is about an 82.68% increase over Quenc 70. About what I figured. It would have been in the 90% area if the muxing didn't take about 4 minutes to complete. It looks like the only way we can see a 90% speed increase is if we record long stuff like Lord of the Rings or Gone With the Wind so that any muxing time should hardly affect what is encoded.

Perhaps you could design it to encode in smaller segments and mux these in the background so that by the time the last segments are done the muxing won't take forever. I think this is what ELDER does.

Other than the minor muxing speed quibble, I think this program works quite well. I'll definitely be using it for awhile.

I'll let you know if I find any bugs.

[danpos]

@Revgen

Good testings and well detailed reports.

I definitively have to upgrade my machine to a DC CPU asap!! I did a price search at local market and with about 378 euros (yep, I did the exchange) I can to buy a Pentium D, motherboard and a pair of memories for dual channel usage. I hope to get them soon.

See ya,

[Revgen]

Hmm... I guess prices in South America must be higher than they are here. I can buy an ASUS MOBO, AMD X2 3800+ and Corsair 1 GB Dual-Channel Memory for about the same price ($485) as you're paying in Euros. Low-end Intel Pentium D Combos are in the $315-$370 range.

Perhaps you have a friends or family who can hook you up here?

Good luck with your new system.

[danpos]

@Revgen

As a matter of fact I will pay for them with my local currency, I just did the exchange for the 'euro' people has a clue about price.

And yes, the prices are quite extorsive here and this quotation (377 euros) was got with stores which work with a certain scheme of importation ...

Sometime ago I'd some friend of mine doing post-doctoral in Physics (yep, I'm a physicist too) in USA, so was easy to got them but now I don't know anyone there, so I've to go for my local store ..

Anyway, thanks for yours compliments!

Cheers,

[foxyshadis]

If you wait until July, when Conroe (Core2), Turion X2, and a new rev of A64 X2 come out, the prices on older PDs and X2s should drop even more... not sure how much of a difference it makes by now though. (Do get a 9xx PD if you go that route, 8xx are the older heat-spewing and power-sucking ones.)