Log in

View Full Version : Encoding Speed


tiberious
30th June 2003, 23:14
Well this is mainly about encoding speed and efficiency,

I am, encoding futurama s3e1
total size: 190mb
res: 640x464 | 25fps

6 - Ultra High
MPEG
VHQ 1
300
0
qpel and chroma motion on/off
with bframes on/off

(not that the above really matters)

Machine:

dual MP 2200+
1gb 2700 ddr

I have encoded some previous episodes with koepi: 04102002-1
with an average of 40fps taking ~15min per pass, and using 100% cpu.

And now with koepi: 24062003-1 and 19062003-1
encoding with an average 10fps taking ~50mins per pass, and using ~ 60% cpu.

Obviously 24062003 is better quality.

http://www.4ckc.net/tiberious/stuff/encoding.jpg

changing the priority does nothing.

just wonderin :]

KpeX
1st July 2003, 00:14
(not that the above really matters)
This is where you are mistaken :D your settings have a tremendous effect on encoding speed. Xvid Koepi 04102002-1 is the last stable build and IIRC does not include some of the 'unstable' features that you are using in your current 'unstable' build, such as QPel, VHQ, Bframes, etc. Do a search or check out the newbie settings sticky, there's lots of info on how settings affect speed. In many people's opinions the only thing that makes these developer builds 'unstable' is the speed of encoding, which is not always optimized compared to a 'stable' release. Hope this helps.

JohnMK
1st July 2003, 02:29
I'm pretty confident he's changed some settings between his comparison. I haven't noticed an appreciable change in encoding speed. There might as well be no difference at all. I could believe a +/-5% change, but -75%? He's screwed up somewhere.

KpeX
1st July 2003, 02:37
@ JohnMK

That's exactly what I was getting at. I don't believe many of the features in use in this encode like QPel, Bframes, chroma motion, were included in the past stable build that is referred to (04102002). Those features definitely explain the speed difference.

Danzel
1st July 2003, 03:15
I seem to remember there was a post awhile ago from someone else with multiple Processors, they also had less than 100% cpu usage.

VHQ if it was in the old xvid build you were using would be a much different VHQ than the modern one, infact many of the features are running quite different code than in that old build.

With VHQ enabled you can expect not to hit 100% cpu usage, if you disable it I think you will find that cpu usage will increase, but your encoded movie will look worse.
(Thats what the general result was in the old thread anyway)

Danzel.

Acaila
1st July 2003, 18:18
I can confirm what Danzel said about VHQ and multi processors. Enabling VHQ drops the total cpu load, but that's just because it's not optimized for parallel processing. This has always been the case for every build with VHQ, so that has not changed with the recent builds.

I also don't see a noticable speed decrease with the new build.

As a side note, setting VHQ from 0 to 1 can very well explain the decrease from 40 fps to 25 fps, nothing unexpected about that. Turning it up to 4 also drops my speed to below 10 fps, but then it always has.

Selur
1st July 2003, 19:17
"I also don't see a noticable speed decrease with the new build."
Same here on my dualMP1800+,...
(okay, VHQ4 is a bit slower, but just a bit)

Cu Selur

Ps.: got the (nearly) the same cpu usage whith vhq1 and vhq0, but like expected vhq0 ist faster ;)

Pss.: your RAM usage looks a bit too high (748MB), VirtualDubMod shouldn't take that much,...

JohnMK
1st July 2003, 22:48
VirtualDub isn't taking that much, Selur.

Selur
2nd July 2003, 06:44
That's what I thought, so this might be a 'hint' where tiberious could investigate, to find the problem for this speed loss. ;)

Cu Selur

NeVeRLiFt
3rd July 2003, 13:37
Originally posted by KpeX
This is where you are mistaken :D your settings have a tremendous effect on encoding speed.

I agree ;)
The settings have a great effect on speed and quality and people need to understand that for the best quality speed will suffer.

Either live it or find a middle ground with it and decide whats ok for you. My encode times are 20hrs for a 2pass encode but the results are 1cd rip thats damn near perfect :D

Nibor
3rd July 2003, 20:56
@ NeVeRLiFt

So you changed your opionion about the quality vs speed thing?
I remember a few posts of you, saying that the time isn't worth it to get this near perfect quality :D :p
Welcome to XviD :)

manono
4th July 2003, 07:42
Go get 'em Nibor. :) I almost posted something similar earlier, but decided to leave it alone. If NeVeRLiFt has finally seen the light, then XviD must be doing something right. :)

tiberious
5th July 2003, 16:49
fair enough, i just want it to use all my cpu power :]]

geoffwa
6th July 2003, 17:06
Goddamn, I'm hitting 48 hours and that's just for the first pass :D

APF_Gandalf
6th July 2003, 18:03
Originally posted by tiberious
fair enough, i just want it to use all my cpu power :]]
I have the solution: just encode 2 movies at the same time :D

Polarbear
7th July 2003, 04:21
Hello,

I was using xvid default setting (came with 1.03 codec pack) and was encoding at 34-38fps nicely. I tried out some of the other settings, particularly VHQ-4, and I think my encoding dropped to ~5fps...(along with chroma motions and Quarterpel). My questions are these three:

1) Does that seem right? Such a horrendous drop in fps. (AMD 2600+, XP, blah blah blah).

2) I am encoding NTSC movies into around 1.45 gigs, and in this particular case, have a 1,709 kBit/s average bitrate. I simply need to retain around 1.45gigabyte final file size, so is there a need to encode movies at such excruciatingly slow speed (for me)? Primarily, will I notice any BIG improvement in visual quality? Should I just use VHQ-2 or VHQ-1? (I'm not the type that will try to analyze movies frame by frame for visual quality...I simply care about the overall 'feeling' and 'closeness' to the source.)

3) Lastly, at and around 1.45 gigabytes / movie, what's a fairly good general setting that I can use for general types of hollywood movies?


Please offer some suggestions / help,
Cheers,
-Polarbear

Selur
7th July 2003, 06:50
1. Yes, it's the way it's 'ment' to be. If you are out for speed don't use VHQ>1.

2. If you don't mess up with some filter or use some hdtv or higher resolutions, vhq(1) should be okay.

3. You might want to read snowbeach's Xvid guide.

Cu Selur

Polarbear
7th July 2003, 07:18
Thanx Selur for the help,

Talking about a humbling experience in terms of raw CPU power. I was semi-happy with my setup in terms of Divx encoding and when I switched over to Xvid, boy...was it a BIG surprise. Thoughts of investing in a dual MP system looms big over my head now.

Just to nail this in the coffin for me, so to speak, if you or others don't mind. Simply trying to find a generally proven setting which most people seem to agree with for my setup and situation, and stick with it for most of my encodings, which are 99% NTSC motion pictures.

Encodings will pretty much all be 1.45gigs minimum in file size, and go up to 2gigs depending on, essentially, movie duration (ie., 2.5hour movies). Movies will be viewed primarily on trinitron desktop monitor, with occasional TV out viewings. With all this mentioned, should I stick with the following settings? (This is a general "sense" of what I gathered from the various posts / faqs seens around here.)


6-Ultra High
MPEG
XVID
2-Limited Search (This one seems to be very argumentative across the various posts. I feel that 4 is overkill for my situation, since I am allocating quite a bit of size for these encodings and leaning toward 1 or 2, what do you guys say?)
300
1
Use chroma motion
Use Quarterpel
Use B-frame (0/150/100/0) (This one seems very argumentative also. Should I stick with 2 for Maximum B-frame? or 0 or even 1?)

Thank you all ahead of time, and hopefully don't flame me out for this, for whatever reasons (me scared....),

Cheers,
-Polarbear

Nibor
7th July 2003, 08:38
In Koepis new build (don't know if you use it, but if not i'd really suggest to!) B-frame settings of 2/150/75/0 increase quality a lot IMHO.
I don't know if VHQ 2 or 1 makes a big difference, I always set it to 4...
Your other settings should be good for your requirements!

Polarbear
7th July 2003, 09:22
Hey Nibor,

Thanx for the input. When you set VHQ to 4, approximately how long does it take to encode, say an average length ~90 min movie? and of course, what CPU you're using?

Also, I know Divx takes advantage of dual CPU fairly extensively, does Xvid also provide such obvious advantages in encoding? I have a feeling Xvid does not seem to be the case though, after seeing some guy with dual MP getting 10fps under VHQ 4.


Cheers,
-Polarbear

Nibor
7th July 2003, 09:46
Hmmm...
A 90 min movie at a 640x256 resolution and two pass encoding mode with VHQ4, bframes and so on (but no Qpel) it takes about 5 hours for each pass, thats 10 hours for the whole encode.
I've got a laptop with a Pentium 4 Mobile 2GHz processor...

I remember a few tests a while ago I made with VHQ settings. I don't know the values, but the speed drop with VHQ 4 compared to VHQ 1 was at about 30%!
But IMHO the results are worth it for a 1 CD encode :D
I'd recommend you to test for yourself which settings best suits your needs!

Have a nice day!

PS: Hope this post is readable for non-swiss-english speaking people ;)

Polarbear
7th July 2003, 10:47
Thanx for the 'real-life' input.

I first thought VHQ is pretty much the only setting which causes heavy speed reduction when used at higher settings, but from what you mentioned, Qpel also greatly reduce the speed for sure.

With my XP 2600+, Qpel + VHQ 4 + other "stuff", 90min movie will take ~8 hours for one pass, and probably ~16 hours for both passes. I'm going to try VHQ 2 after this encoding and compare both the time and the resulting encoded movies, and see with my case, whether the many extra hours spent for VHQ 4 justify the cause. (Once again, have to take into account that I am encoding strictly at fairly high bit-rate of 1.45gigs - 2gigs per movie...perhaps the size will reduce, if not almost eliminate the possible visual improvements seen by VHQ 1 to VHQ 4, hence making the many extra encoding hours not necessary.)

I hope others who come across this particular thread can be generous enough and also provide their real-world experience in this matter, briefly will be fine. The more cases, the easier it is to attempt to generate a somewhat reliable GENERAL setting for those who cannot dedicate a box solely for encoding purposes.

Cheers,
-Polarbear

(Nibor, your 'sweenglish' is very understandable, don't sweat it!)