Xvid encoding speed on new Dual Core Chips

[gte024h]

In case anyone was wondering how much faster the dual core chips are at encoding, I have some first hand information.

New Hardware:
Athlon 64 X2 4600+ (2x2.4Ghz Manchester cores)
nForce 4 based motherboard at stock settings
Radeon X700

Previous Hardware:
Pentium 4 3.0C (3.0GHz Northwood core w/Hyperthreading)
Intel 865 based motherboard at stock settings
Radeon 9800 Pro

Same RAM (1GB DDR400), drives, case, and power supply between the two setups.

Software (same on both setups):

Windows XP Pro SP2 (32 bit) w/all updates and latest drivers
VDubMod 1.5.10.1
Avisynth 2.5.5
DGDecode 1.4.4b4
Xvid 1.1.0b2
Lame ACM 3.96.1

I encoded some Star Trek Voyager episodes (~45 minutes, full screen) and recorded the time between the start of the first pass and end of the second pass (as reported by VirtualdubMod) on each setup.

The AVS file used for the test:

loadplugin("dgdecode.dll")
loadplugin("decomb521.dll")
mpeg2source("path to file.d2v",cpu=4)
telecide(order=1,guide=1,vthresh=50,blend=false)
decimate(mode=3,threshold=2.0)
lanczos4resize(640,480)

Xvid settings were AS@L5, 2-passes, 988kbps, Lame was set to encode at 128kbps CBR on the second pass.

The final results:
Pentium 4 @ 3GHz = 80 minutes
Athlon 64 X2 4600+ = 41 minutes

There it is... the same procedure on the same input files with the same software where the only difference is the CPU, motherboard, and video card. I can't wait to see how much faster the 64-bit software will make things (still waiting on 64-bit drivers for my TV Wonder Pro).

[MeteorRain]

someone just told me that AVS don't perform well on dual core as it didn't be optimized on SMP

[Kopernikus]

Xvid is not SMP optimized, too. So the gain with dualcore is probably because AVS and XviD each can use one core for themself.

[Sharktooth]

exactly. it you try with x264 or other threaded encoders, the gain will be much higher.

[gte024h]

Well looking in the task manager each core is averaging about 90% usage. I think AviSynth and VDubMod are using a core, and Xvid is using another. On the second pass, the CPU usage stays closer to 100% because Lame is also working. So on the first pass there are 3 threads, and on the second there are 4 (not including all the other Windows junk running).

An SMP capable version of Xvid would operate very well on dual core if you aren't using any other stuff like AVS or Lame, but looking at the current usage I don't think SMP would make much difference. 64-bit on the other hand could be huge... I dunno.

http://parametric.org/~gte024h/xvid.jpg

[squid_80]

64-bit versions have already been done, there's not much of an increase since the code is already very optimized. The instructions are the same, there's just more registers - it's like adding more wheels on a car, they're not going to help it go faster.

[gte024h]

That's too bad about the 64 bit, but maybe between SMP and 64-bit the overall performance will be improved a bit more. I'm going to try a little overclocking and tweaking as well, so maybe I can get more speed that way.

I would be curious to see how the dual core Pentium 4's do. But I've spent enough on my setup, someone else will have to try that.

I know there are a lot of hardware sites out there with benchmarks (including xvid), but they never encode exactly the same way I do and they never include comparisons to older hardware. Usually they compare to other equipment currently on the market. So if you want to upgrade from old hardware it's hard to see how much faster the new stuff would be.

[708145]

http://www.funknmary.de/bergdichter/projekte/index.php?page=ELDER
gets near perfect speedup on up to 8 CPUs.
Just works on Linux right now since I don't bother to port it to windows for now. I tried once and it is PITA.
Once I get an X2 myself I'll have enough motivation to port it I guess :)

bis besser,
708145

[Pookie]

This app might help -

http://www.mlin.net/SMPSeesaw.shtml

http://www.7-max.com/