Should encoding times change so drastically?

[gonzalimator]

Hey,
I'm new to encoding for DVD and since my only experience so far has been with xVCD and CVD, I'm not sure what kind of difference in encoding time is normal.

DivX to CVD, 2-pass VBR(avg. bitrates 1500-2500) = 10-12 hours (2hr movie)
DivX to DVD(NTSC), 2-pass VBR(avg.4594, max.9000, min.2000) = 21 hours (2hr movie)

I've never seen my machine take so long. Is this normal?
I'm using TMPGEnc on my Win2k partition.

I also frameserved from VDub (for resize) and the AVI had two audio tracks, AC3 and DivX audio. Could this have slowed down VDub? Is it the change in resolution?

[fmctm1sw]

I remember on my old machine (celeron 433), I was averaging about 5.5 hours per 1 hour source to CVD at about 3500CBR. No filtering, resizing or anything. Just straight up. I assume your using the 600MHZ machine in your profile?

[gonzalimator]

yes I am.

So is the DVD video encoding time average (for my machine) or what?
Could one say that the times are normally longer for DVD?

[fmctm1sw]

yes I am.

So is the DVD video encoding time average (for my machine) or what?
Could one say that the times are normally longer for DVD?

I never dared full D1 resolution on it...

[flaninacupboard]

sounds about normal, afterall full dvd resolution is quadruple VCD resolution and double CVD resolution, so a doubling of encode time is about right. a faster machine would help

also, i would change your minimum bitrate from 2,000 to 0. no need to use 2,000kbps on a black screen. if you're worried about your player choking on 0 then use 500kbps.

[VCDHunter]

a 633 celeron I would avoid any kind of DVD encoding until you get a faster machine. It takes long enough on my xp2000

[flaninacupboard]

you should get better than real time with tmpgenc unless you're using lots of filters, using CCE it pushes just over 2.5X on my xp2000.

[VCDHunter]

you should get better than real time with tmpgenc unless you're using lots of filters, using CCE it pushes just over 2.5X on my xp2000.

No filters but a straight avi to DVD conversion with Motion Search Precision set to High took me around 6hrs for a 3 1/2 hr film. This was with 23.97 fps (internaly 29.97) and 3:2 Pulldown when playback enabled.

[reboot]

CPU speed and bus width are directly related to processing/encoding times.
CPU make also seems to be significant.
A 633mhz machine, with a 133mhz FSB, is what I would consider below minimum spec for video encoding. Sure, you can do it, but like you've found out, it takes about 3 hours (or more) to encode one hour of video.
A 1ghz machine ratio would be about 2:1
A 1.5ghz machine would be about 1.75:1
A 2ghz maching would (should) be better than 1:1 (wider/faster bus).

For comparison, a 2.4ghz Intel takes twice as long as my 1.85ghz AMD.
I know that doesn't seem to make sense, however, AMD chips are simply faster at video encoding than Intel, and I have several friends that can back that up. Your mileage may vary.
If anyone has any REAL evidence why Intel are so slow at this, please post links.

[flaninacupboard]

vcdhunter, i guess that's 2 pass then? in which case that is about 1X.

reboot, this webpage seems to emphatically disagree with your comments. http://www6.tomshardware.com/cpu/20040322/prescott-14.html an Athlon FX-51 is beaten by a P4 2.8

i found the FX-51 for sale at £468, and the P4 2.8 for £123.

if we go up a step the FX-53 i found for £581 and the slightly quicker P4 3.0 for £157.

of course if you peek here http://www6.tomshardware.com/cpu/20040322/prescott-09.html
the athlons kick ass, but for video encoding i'm afriad it's all about the ghz, and intel win hands down.

[reboot]

Interesting arguments. Now if only any of that were true in real life.
I have two systems here that compare in hardware/drives/ram/etc.
AMD XP2500+ and Intel P4 1.8ghz.
Given the same avi, in the same software, on the same OS, at the same bitrate, at the same blah blah blah, the Intel will take 3 hours and 43 minutes to encode a 2 hour avi. The AMD will take 1 hour and 17 minutes.
Please explain why?
I have friends that spent a ton on Intel 2.4's and up, and my lowly $73 Athlon (on a cheap MSI board) still outperforms.

[flaninacupboard]

because an XP2500 runs at around 2.3Ghz, so has a 500Mhz lead on your P4. plus a P4 1.8 runs on a 133bus while the 2500 runs on a 166bus. and as i said, it is all about the MHZ, so if the amd chip is quicker, it's going to win. a fairer test would be a 2500 versus a P4 2.26 running on 166 bus.

oh look, the intel wins:

http://www6.tomshardware.com/cpu/20030217/cpu_charts-25.html

Don't get me wrong, i use AMD chips and always will because they offer good value for money, and supporting AMD encourages competition, which is good for everyone. it's just a fact that P4's are better at sequential tasks like encoding.

[reboot]

What?!?
Look again. The Athlon XP 2500 runs at 1837mhz, give or take.
The single biggest difference is the bus speed of 166mhz, instead of 133mhz. So, given that the Athlon has only a 20% faster bus speed, why does it encode video at ~120% faster?

Benchmark anything anywhere, and real life situations are going to contradict it. Benchmarks are for selling points, not for actual usage.

[flaninacupboard]

well toms don't sell stuff, so they're not making the results up for profit.

I'm going to venture a guess on this one though. the P4 core is not as good as the P3 core, the P4 requires high clock speeds for good performance. the amd core can do more work per clock than intel. as we go higher up the amd scale the disparity between model number and clock speed increases. so at the top end an xp3200 (amd think is comparable to p4 3.2) is much slower than a p4 3.2. however lower down the scale the increased work per clock is more important. if i had all the figures to hand i'd try and draw it out on a graph

[reboot]

Now you're making sense
After a bit of my own Googling, I found the same conclusions.
At the lowest range of processors, in the 600mhz to 1ghz, Intel wins. AMD had almost nothing that could keep up.
From 1ghz to ~1.95ghz AMD wins. The CPU's just do more work faster than a comparable Intel.
Above 1.9ghz Intel wins again, mostly by sheer bus speed until we get to the 200mhz point, then CPU clock speed takes over.

[gonzalimator]

Thanks for all the replies!

Just to clarify... I have a P III 600. I'm not trying to brag but I think somebody said something about a Celeron 633.

I would avoid any kind of DVD encoding until you get a faster machine.

I think that if your machine can do it, why not do it? I'll get a new machine later this year. So now I know that my encoding time is normal.

'also, i would change your minimum bitrate from 2,000 to 0. no need to use 2,000kbps on a black screen. if you're worried about your player choking on 0 then use 500kbps.

Thanks for the tip.

Thanks to everyone!

[VCDHunter]

vcdhunter, i guess that's 2 pass then? in which case that is about 1X.

No CQ, 1 pass. I don't like the sound of this. If we have almost identical machines shouldn't we have very similar encode times ? I get nowhere near realtime for DVD encodes.

[RoopeT]

XP 2400+:

avi frameserved via avisynth to CCE SP 2.50 ~ 1 hour movie encoded in 30 minutes (CBR or CQ VBR)

DVD frameserved via avisynth mpeg2dec plugin & DVD2Avi project ~ 1 hour movie encoded in ~ 45 minutes

TMPGenc takes several hours for one hour movie using high quality motion estimation.

Frameserving from VirtualDub is A LOT slower than frameserving with avisynth.

[reboot]

gonzalimator, now you know. I also have a PIII 600, but I've loaded it up with 768 meg of RAM, and a stripped down version of XP. Doing enocdes on it is an overnight project, especially on 2pass VBR. It does the job, but only that job. No multitasking at all. I would say that single pass CBR encode would take about 15 hours for a 2 hour avi. CQ probably longer, and 2pass is a 2 day event

Note: My XP2500+ does slightly better than realtime encodes, 2pass VBR to DVD. Mainconcept is even faster with the same settings (bitrate/2pass/etc.).

[bugster]

If you are going to compare encode times, at least make sure you are comparing like with like. I have not yet seen in this thread a full list of anybodies settings to compare against.

For TmpGenc in particular, motion search precision, CBR or CQ or 2pVBR are important, but also make sure the source material is at least pretty similiar (DV is pretty standard, divx is not, but even the amount of movement in a scene affects the encode time).

For encode time comparisons between 2 Pc's to be really meaningful, the software used has to be identical (including specific versions) and the settings identical too if the comparison is to be of any real use, otheriwse its apples to oranges.