Newer versions of DoItFast4U do not utilize both CPU's in a dual processor setup?

[newguy007]

It seems that the newer beta versions of DoItFast4u are somehow slowing the CCE encoding time by half. I use CCE v2.66 on Win2k on a dual 1.2GHz Athlon MP, 512 DDR ram, Tyan S2468. It's as though only one of my processors is doing the actual work since it now takes twice as long to encode, but the performance window reads 60% usage for BOTH CPU's. I can't see what's wrong. I usually had had a speed of around 1.8x real-time, but now the highest it'll go is 0.9x. I've been updating DOITFAST4U with the auto-update feature -- I haven't made any other software or hardware changes. Is anyone else having a similar problem? Please help :(

[69Mws]

Try an encoding job without using DIF4U, but I doubt much that it would be any difference. I don't know how DIF4U should change the way CCE encodes.

My guess is that there were maybe options activated, you didn't use before, like deinterlacing and stuff. That slows down encoding a lot.

Greetz
69Mws

[newguy007]

I tried to encode the movie manually with cce and the speed was the same (~0.9). What's odd though is that I tried to encode a different movie with doitfast4u/docce4u and the speed came back to normal, ~1.8x. So I guess it's really dependent on the movie. I'll try to find out what exactly causes the decrease in speed.

[D3s7]

your biggest drop in encoding speed is in the .avs

If it's interlaced it has to telecide / decomb the source or you get issues.

That will drastically drop the speed of encoding

[Eyes`Only]

Unless you are using CCE PRO (Most users use SP) I don't see how you expect to encode using two processors. If you are using CCE PRO, I guess there's an issue, but let's get that fact straight first ;)

Also you're using AMD. As far as I know, there's never been a version of CCE that can use dual-AMD CPUs. Even the PRO version only supports PIII and P4. Cheaper chips need not apply, I guess.

[newguy007]

Sorry, I should've been more clear -- I'm using CCE SP v2.66.

Maybe that was the case for previous versions of CCE SP, but the one I'm currently using supports dual Athlon MP processors. The two separate CPU Usage History graphs in the Task Manager average to about 60% when encoding. And 1.8x RT for a single 1.2 GHz CPU is highly unlikely.

[Kedirekin]

In my experience, you typically get between 55% and 75% CPU usage on a dual-CPU rig when encoding in CCE via AviSynth.

I've always assumed that AviSynth and CCE are running in different threads. With a slow avs (lots of filters), AviSynth uses 50% (all of one CPU), but because it's slow, it bottle-necks throughput down to the point where CCE only uses 5% - and it shows as 0.4 RT in CCE. With a fast avs, AviSynth uses the same 50%, but supplies CCE with data fast enough that CCE uses 25% - and that shows as a five-fold increase in encoding speed (2.0 RT).

IIRC, you'll see CCE really scream if you encode directly from an AVI (with or without SMP).

[Eyes`Only]

makes sense to me, because like I said before, I KNOW that CCE SP doesn't support dual-CPU. If the second CPU is being used for avisynth, well that's all good, but that's NOT CCE SP using the second CPU. I've always actually assumed SP stood for Single Processor anyway.

[Kedirekin]

You know, thinking about this a little further, we don't need a multi-processor enabled CCE. That's not where the bottleneck is.

What would really increase speed on an SMP rig is a multi-processor enabled AviSynth (assuming of course the CCE and AviSynth really do run in separate threads).

On second thought, that probably means we'd need an MP enabled mgp2dec and MP enabled decomb filter too. Never mind.

[da franksta]

@eyes'only
lots of respect for all the great work around here, but i have to disagree

cce pro is provided WITH hardware, unlike cce sp. cce pro supports dual systems, but is not provided with one (anymore). since it's build to encode in realtime, it needs at least 2 p3 800's to do this.
all this, however, does not mean cce sp (or any other cce version for that matter) does NOT support dual processors. on the contrary. sp stands for software package (i.e. hardware not included) and uses dual processor systems just as well.

that being said, kedirekin is absolutely right about what he says about avisynth. who knows one day it will be multithreaded...

[Eyes`Only]

hey if i'm wrong then i'm wrong. But I still don't think it's optimized for dual CPU. It may use it, but it'll use it in the same way that every other software app does, which is by optimized libraries. I really really doubt the programmers have taken the time to optimize the use of SP with multi-processors, and if they have, it's because they left code in from their PRO version, which is optimized for mult-Pentiums, not Athlons. Anyway it's a little off-topic, as the original subject of this thread was umm... trying to blame my app for the slowing of CCE, which is wrong. Especially since my app doesn't even run CCE (yet...)

[da franksta]

:)

[Kedirekin]

@newguy007 or da franksta,

Are either of you willing to create a 1 or 2 minute, 720x480, HuffyYUV AVI clip (from any source you want), encoding that with CCE and seeing what speeds you get? I vaguely remember getting 3.0 RT on my old 1GHz dually. I'd be willing to bet newer rigs can get between 4.0 and 8.0 RT. I'd do it myself, but my dually fried about 1¼ years ago.

If you're not willing, that's fine, but I know I'd be curious to hear the results, and I bet others would be too.

If my math is correct,
- 60%-50% = 10% (percent used by CCE in newguy007's first post)
- 50%/10% = 4.5RT/0.9RT (CCE speed if CCE could use 100% of one processor)
if you get over 4.5 RT that is strong evidence that CCE SP is using both processors.

Unfortunately we can't conclude anything if you get under 4.5 RT - that could just mean that decoding HuffyYUV isn't fast enough to load down CCE to the point where it uses more that 1 whole processor.

On a side note: I used to get a max of 1.25 RT on my 1 GHz dually with CCE/AVISynth. That was when I was seeing 75% processor usage (50% for AVISynth, say 23% for CCE, and 2% for system overhead). ~2.7 RT should have been my max if CCE was single processor only. If my recollection of 3.0 RT encoding directly from an AVI is correct, then that is evidence that CCE SP is MP enabled.

[da franksta]

all this has been covered elaborately in the past here (http://forum.doom9.org/showthread.php?threadid=26451) .

I got an e-mail from Cinemacraft tech-support confirming that CCESP categorically DOES support SMP: "CCE is designed by multi-thread and supports dual processors. We recommend you a dual-CPU for high speed MPEG2 encoding as almost two times faster. However, please understand whole processing speed must be less than two times, because it depends on decoder and HDD as well.

This is why we also recommend a PC having two HDDs. If your PC has
only one HDD, it is a bottleneck in whole performance. You should
save a source file and an MPEG file into each HDD separately".



as you can see, the matter is well documented

i hope any of this helps

[Eyes`Only]

Sweet! That's all I wanted was documented proof. I concede. Now I want to build my SMP box that much more. :D

[rkorkie]

newguy007,

I don't know if you have been following the change log for DF4U lately, but there were issues with the detection of FILM or NTSC. With versions 1.1.7 and 1.1.8, most FILM movies were being diagnosed as NTSC, even though they were film, thus your .avs script was using telecide and decimate, which slows down encoding by about half. If you have the latest version of DF4U 1.1.9, the detection has been solved for now, and your .avs scripts, as D3s7 pointed out in a post below, should not be using decomb unless necessary.

Hope that helps. It seems the dual processor discussion was a bit of a red herring, and it probably was leading you in the wrong direction.

[timekills]

The dual CPU question can be quickly (if not completely accurately) confirmed by opening the task manager on an SMP system (and, no, I don't mean HT-enable P4s (bleh))

I get around 80% usage on both CPUs - whether AMD MPs (2400s) or P4 Xeons, by the way Eyes` Only - by default. It is considerably faster than, for example, TMPG with multi-threading enabled. Which also benfits significantly from SMP. Although my good friend Hooz from 2CPU disagrees with my CCE vs TMPGEnc opinion... :D

Clearly the hard drive bottleneck is a huge factor. On my uni P4 box I use the speed increases drmatically by using two hard drives. Obviously I use multiple hard drives on my dual boxes.

Back on topic, sort of, DIF4U, as has been discussed, obviously has no impact or bearing on the CCE speed, as it just feeds the info to DoCCE4U or ReAuthorist.

:sly:

[Eyes`Only]

Hmm didn't I already concede?

For those that didn't get it--

con·cede : to accept as true, valid, or accurate

[timekills]

Uhhh. Damn. I knew I shouldn't have thrown away my Webster's after grad school.

I concede your concession came well before my logorrheic affirmation of SMP's abilities. :D

[Eyes`Only]

touche. I see your logorrheic affirmation and raise it a loquacious outrance :)