Log in

View Full Version : XviD for Dual Cores


veagles
4th August 2006, 14:53
I have dual core processor. Latest build of XviD uses only one core, where as DivX uses both. Please post a link for the multithreaded version of XviD. I cant find it anywhere. I searched the forum and links given there are dead.

Sharktooth
4th August 2006, 15:20
just use the celticdruid build and set the appropriate number of threads.

BigDid
4th August 2006, 17:49
Hi,

In case you didn't find it yet: http://celticdruid.no-ip.com/xvid/

Actually very interested in buying an AMD dual-core; would you share some informations like: what model, nominal speed or overclocked, any other comments...
Thanks

Did

Blue_MiSfit
6th August 2006, 02:02
The 3800+ X2 is a total bargain right now. They are about $150 for the 939 or AM2 platform at newegg. These chips have a clock speed of 2 GHz, and generally overclock very well, sometimes hitting 2.6 on the stock air cooling with some voltage boost from what I've read.

On the higher end, the 4000+ X2 is just under $400, and also runs at 2 GHz, but it has 2 MB of cache (1MB x 2).

Still, Conroe (Intel Core 2 Duo) really is the better choice at this point in time. They overclock like an RB26DETT engine, and will be very affordable. Of course, theyre not available yet :)

Stock 2.4 GHz hitting 5 GHz+ on water cooling..... freaking unbelievable...

LordIntruder
6th August 2006, 13:16
I have dual core processor. Latest build of XviD uses only one core, where as DivX uses both. Please post a link for the multithreaded version of XviD. I cant find it anywhere. I searched the forum and links given there are dead.

It is quite normal, you HAVE to set it up yourself to 2 threads.

In the Xvid configuration window, at the bottom click on "Other Options".

Here you have a line:

"Number of Threads" which is set on "0" by default. Put 2 and now you will encode on 2 threads. :)

foxyshadis
6th August 2006, 13:46
0=use process affinity. In other words, automatic.
>0=specify exact # of threads.

Most likely he was using 1.1 instead of 1.2+.

Koepi
7th August 2006, 06:33
The SMP-alpha-build on my site also works multithreaded (1.2.-127). Find it following the link in my signature. Please post feedback!

Cheers
Koepi

Yanaran
7th August 2006, 10:32
I'm using that one Koepi and as far as I can tell it works great, 98-99% cpu used when encoding. I have the threads set to 2 manually (didn't know 0 would auto detect).

squid_80
7th August 2006, 12:09
People, please don't take high cpu usage as a sign that you're getting good performance. It's actually possible for the codec to stall completely while cpu usage shows close to 100%. Compare fps or total encoding time if you want to quote valid figures.

sysKin
8th August 2006, 06:05
People, please don't take high cpu usage as a sign that you're getting good performance. It's actually possible for the codec to stall completely while cpu usage shows close to 100%. Compare fps or total encoding time if you want to quote valid figures.
And if someone doesn't believe - try more threads than CPUs in XviD: output will be the same, CPU usage will go up, but encoding speed will go down.

I'm not sure if 0 threads means auto-detection just yet. "Load defaults" should fill the number using auto-detection.

squid_80
8th August 2006, 06:23
Dropping the priority in the middle of an encode can do a similar thing.

About that sleep(0) call: There's a function called SwitchToThread() that does something similar, but unlike sleep(0) it yields to lower priority processes. Unfortunately it's not available on anything below NT4.0 i.e. win95, 98 or me.

Koepi
8th August 2006, 07:14
Win 9x (the extended DOS-windows-versions) don't support multiprocessors anyways, so I wouldn't even think about using that OS on a highend machine.

Cheers
Koepi

Yanaran
8th August 2006, 13:18
It's actually possible for the codec to stall completely while cpu usage shows close to 100%.
Well I though everyone who has used windows knows you can max a cpu without doing anything useful. ;) I guess you aren't allowed to give positive feedback without some valid figures as you call it, so I apologize for that. :p

Settings: Q=3, AS5, MPEG quant, B-VOPS 2/1.62/0, Packed, ME6 VHQ1 VHQ for bframes, chroma optimized, 240 max I-frame, trellis

1280x544 test clip:

XVID 1.1 final: 157s ~50% cpu
Koepi's SMP 0 threads: 155s ~50% cpu
Koepi's SMP 2 threads: 140s ~85% cpu
FPS increase 100*157/140-100=12%

Same clip and settings, with spline36resize(704,368) in avisynth (for dvd-rw):

1.1 final: 73s ~85% cpu
Koepi's 2 threads: 70s ~97% cpu
FPS increase 100*73/70-100=4%

Good enough for me! Keep up the good work :)

sysKin
8th August 2006, 13:26
Good enough for me! Keep up the good work :)
What CPU is that?

Yanaran
8th August 2006, 13:38
Pentium 4 630 Prescott

sysKin
8th August 2006, 14:26
Pentium 4 630 Prescott
Ah, a HT ~ phew, I was worried it was some dual-core :)

LordIntruder
8th August 2006, 23:20
0=use process affinity. In other words, automatic.
>0=specify exact # of threads.

Most likely he was using 1.1 instead of 1.2+.

No, "0" means dual-core capacity won't be used and not it is automatic. I have a 4200+, if I use "0" I have no gain if I compare with an old build that is not multi-threaded, the encoding time is the same, the % used is 50% no more. If I set "2", I get a real gain, the encoding time is much faster. I tried to set "3" and "4" and I get no gain and like sysKin said, it is even slower.

Unless you use a specific build were "0" has been coded to be automatic? As for myself I use Koepi build 1.2 under Windows XP Pro.

BigDid
8th August 2006, 23:35
... I have a 4200+,...
Hi,

As stated above (http://forum.doom9.org/showthread.php?p=859420#post859420), I am very much interested in live infos from AMD64 x2 users, specially 3800 and 4200.

Merci/ :thanks:

Did

foxyshadis
8th August 2006, 23:54
No, "0" means dual-core capacity won't be used and not it is automatic. I have a 4200+, if I use "0" I have no gain if I compare with an old build that is not multi-threaded, the encoding time is the same, the % used is 50% no more. If I set "2", I get a real gain, the encoding time is much faster. I tried to set "3" and "4" and I get no gain and like sysKin said, it is even slower.

Unless you use a specific build were "0" has been coded to be automatic? As for myself I use Koepi build 1.2 under Windows XP Pro.
I noticed after I tested and saw the results. Weird, I thought I had a build with CPU detection (I've updated several times since so it's hard to say), I'm not sure if it was customized or if I was just mistaken. >.> Maybe I haven't noticed since avisynth usually restricts encode speed.

It wouldn't be hard to just plunk x264's new code almost verbatim into it though.

sysKin
9th August 2006, 04:17
As stated above (http://forum.doom9.org/showthread.php?p=859420#post859420), I am very much interested in live infos from AMD64 x2 users, specially 3800 and 4200.

Merci/ :thanks:
Hmm what kind of info do you need? Multithreded code was being developed on an X2 4200+ you know ;) Speedup is from ~30% on very fast settings (1st pass) to ~80% on slow settings.

squid_80
9th August 2006, 05:58
It wouldn't be hard to just plunk x264's new code almost verbatim into it though.
Other way around - the code to auto-detect cpus has been in xvid since feb and the same code was added to x264 just recently. The VFW frontend just doesn't use it correctly.

BigDid
9th August 2006, 17:58
Hmm what kind of info do you need? Multithreded code was being developed on an X2 4200+ you know ;) Speedup is from ~30% on very fast settings (1st pass) to ~80% on slow settings.
Thanks,

So, the more processor dependant, the more speedup, if the app is multi/hyper-threading optimized I suppose.
Actually I gather infos on the 2 above models to take a buying decision when they will be available again (out of stock for my 2 possible french onlines stores :( ). Also I try to get stats on the overclocking possibilities.

So far I have more infos on the 3800 (before the prices cut-down) than on the 4200.
The 3800 seems definitely capable of going up to 2400; above being stepping dependant. Problem is I should not get improvement on my actual 3000@2450 for mono-threaded apps if maxxed @2400 !
Hence the 4200, less stats than the 3800, begins @ 2600 which suits me IF confirmed by other users; see here:http://forum.doom9.org/showthread.php?p=859073#post859073

Main goal will be (xvid) encoding but, from what I have read, the system and any app could benefit; hence my call to users experience.

Did

pelle412
10th August 2006, 13:54
Are there any benchmarks out there that shows the performance of both the Pentium D processors and AMD X2 using the multithreaded alpha version of XviD?

moon1234
13th August 2006, 11:06
Are there any benchmarks out there that shows the performance of both the Pentium D processors and AMD X2 using the multithreaded alpha version of XviD?

That is not really a good comparison. The better comparison would be to the new Core Duo 2 processor.

Emp3r0r
14th August 2006, 20:13
That is not really a good comparison. The better comparison would be to the new Core Duo 2 processor. A comparison of all three would nice as there is likely more users with the Pentium D right now than the Core 2 Duo. Also, since Core 2 Duo isn't widely available at decent prices, some could still be interested in the Pentium D 900 series as it is much cheaper and still overclocks well (4+ghz).

BigDid
14th August 2006, 20:22
Are there any benchmarks out there that shows the performance of both the Pentium D processors and AMD X2 using the multithreaded alpha version of XviD? and A comparison of all three would nice as there is likely more users with the Pentium D right now than the Core 2 Duo...
Hi there,

Maybe this link :http://tomshardware.thgweb.de/2005/11/03/benchmark_marathon_82_cpus_von_amd_und_intel/page76.html
or this one: http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2736&p=8
can suits your needs?

Also a very informative thread here: http://forum.doom9.org/showthread.php?t=112621

By the way, no news from Inc; I would'nt like to disturb him so I'll wait a bit more before upping or PM :)

Did