P4 3.2 Vs Amd 64 3200

[Bear]

Hello all,

I just found a review for 2 xpc at :

http://www.hothardware.com/hh_files/CCAM/shuttle_65g2_85g4(4).shtml

Both machine using a very powerful cpu.

I also want to build a similar machine for encoding purpose. I usually convert a lot mpeg2 files to xvid format (720x480 at 1500k).

It seems P4 got a better performance for encoding to divx and I think same result will apply to xvid too.

So which one do you think is better for xvid encoding?

[Bear]

here is another:

http://www.neoseeker.com/resourcelink.html?rlid=66707

[Sulik]

At similar ratings (3.2 vs 3200), P4 is always better for encoding, mostly because of SSE2 support and HT (do something else while encoding/capturing).

[Tommy Carrot]

Originally posted by Sulik
At similar ratings (3.2 vs 3200), P4 is always better for encoding, mostly because of SSE2 support and HT (do something else while encoding/capturing).

Athlon 64 also supports SSE2, and afaik 64 bit mode will help too in video encoding.

[Prettz]

For the computation-heavy parts of the encoding process that are entirely done in MMX/SSE/SSE2, I think the P4's much higher clock speed makes it perform much better than Athlon64 on these parts (because as far as I know the execution latencies and execution units used for both processors are almost identical, I could be wrong about the Athlon64 though).

[symonjfox]

Maybe Athlon 64 will be faster if using a 64bit OS (Win XP 64) and when there will be Xvid Recompiled to 64 bit.

AFAIK SSE2 of Pentium4 is faster than SSE2 of Athlon 64 from the first tests.

[Defiler]

While neither option is as fast as a similar SMP system would be (though regrettably you can get a pair of Xeons for the price of one Opteron CPU right now.. Insane.), either one of those CPUs would be extremely competent at MPEG-4 encoding.
Personally, I suggest that you find the motherboard you want, and buy whichever CPU fits in its socket. Heh.
If your favorite motherboard has an 875P chipset, get the P4. If your favorite board uses nForce3, get the Athlon64.

[Rober2D2]

I have read in several topics that:
a) divx is faster for Pentium 4
b) xvid is faster for AMD (Although this refers to 32 bit AMD, I don't know about AMD)

[Sharktooth]

AFAIK DivX networks have an AMD 64 optimized DivX codec and it should be bundled with boxed Athlon 64s.

[Defiler]

Originally posted by Rober2D2
I have read in several topics that:
a) divx is faster for Pentium 4
b) xvid is faster for AMD (Although this refers to 32 bit AMD, I don't know about AMD) This is possible, but all I know is that MPEG2Source->UnDot->Crop->Resize->Convolution3D->XviD@640x480 averages about 42fps on my Intel box. :)
I would love to have a dual Opteron, but you can get a pair of Xeon 2.8GHz chips for the price of a single 2GHz Opteron CPU right now, which seems like a poor value to me.
Back to the original topic.. if someone reputable makes an nForce3 motherboard with PCI-X and non-Realtek onboard Gigabit, I would go for the Athlon64 option. Otherwise, I vote P4.

[Sharktooth]

I think the VIA soulution asus adopted (k8v deluxe) is one of the best for the athlon 64 cpus.
If i had to choose i would go for Asus (it also sports a lot of cool stuff like 3com gigabit ethernet controller, firewire and wi-fi support to name some) but im still happy with my Barton 2600+ (@2400Mhz).
Athlon 64 series will reserve some surprises when the 64bit software/codec/OSes will be available expecially in the first quarter of 2004 when AMD will start selling cpus with 1Ghz FSB and faster memories support.
On the counter part Intel will push again the performance with their P4 Prescot but from the few reviews i've read they seems to have thermal problems ... its excessively hot (~70°C for a 2.8Ghz CPU is too hot for me).

[cult]

@Defiler
what is your cpu,memory,xvid settings?

[Defiler]

Originally posted by cult
@Defiler
what is your cpu,memory,xvid settings? Dual Xeon 2.8GHz, 2GB PC2100, and the number I just quoted is with max b-frames = 2, H.263 quantizers, Koepi's 6-24 build of XviD, and VHQ=0.
Changing that to VHQ=4 drops things down to like 14fps. Heh.

[Bear]

Hi all,

I finally got a barbone system which is usng P4 3.2G(http://www.soltek.com.tw/Chinese/product/QBIC/eq3401/eq3401.html). I also have a dual amd 2600 mp machine. I use procoder to encode all my movies.

Yes, the speed of my PC is faster than my mp system. I use both xmpeg and procoder to test it. I encode a mpeg2 movie to xvid, 720x480 1500k. The speed is faster for P4. I don't know why the dual system which running at 2 cpus is slower than a P4. 3.2 G system. May be P4 is running at 800.

You can have a look here:

http://www6.tomshardware.com/cpu/20030923/athlon_64-37.html

P4 3.2 is much bettern than AMD for mpeg4 encoding.

The speed is still slow from mpeg2 to wmv9(720x480 1500k 7-8 fps only). I really don't know why it's much slower to encode to wmv9.

[Sharktooth]

Toms Hardware is not reliable when comparing CPUs.
He's known for his "pro-intel" reviews...
For example he used sse2 optimized IDCT for P4 and generic athlon optimized IDCT for athlon64.
The athlon64 would have performed much better if he used sse2 for that CPU too...
He did not included a test of the AMD64 optimized DivX too...
And he did not specified the software he used is widely optimized for P4 and still not even tought for AMD64...
That said the differences are small and in any case a P4 3200 is a good choice.

[cult]

@Defiler
thank you for answering.That explains the 42fps.I can only reach 12-14fps with vhq4 with my 2700xp

[Leak]

Originally posted by Bear
Yes, the speed of my PC is faster than my mp system. I use both xmpeg and procoder to test it. I encode a mpeg2 movie to xvid, 720x480 1500k. The speed is faster for P4. I don't know why the dual system which running at 2 cpus is slower than a P4. 3.2 G system.

Most probably because encoding a video isn't exactly a parallelizable task - almost all of the work will be done on one CPU, with the other one idling around.

It'd probably go faster if you were to encode 2 videos at the same time (at least if you store each on a different hard drive so the drives won't start thrashing...).

np: Ricardo Villalobos - Bahaha Hahi (Alcachofa)

[Latexxx]

Anandtech compiled Lame for 64-bit Linux and the results were quite impressive (when compared to Athlon-64 in 32-bit mode): http://www.anandtech.com/cpu/showdoc.html?i=1884&p=17

[Sharktooth]

Yep, anand is one of the reliable sources...:)
BTW the performance boost on the 64bit lame is impressive.
Hope that M$ will wake up and speed up the developing of Windows XP for x86-64.

[Defiler]

I would love to have a dual Opteron box, but the price difference between 2xXeon and 2xOpteron is enough to pay for the motherboard and a couple fast SCSI disks. =(

[Sharktooth]

With the introduction of the x48 series the x44 and x46 models had massive price cut.
My dual Opteron box is coming for xmas:)

[Bear]

But will xvid got a 64bit version?

[Latexxx]

When Win Xp 64-bit edition comes, the only required thing will be to compile Xvid with a compiler that supports 64-bit programs. It is also possible to add more 64-bit optimzations but just compiling it will give a great performance boost.

[sysKin]

Originally posted by Bear
But will xvid got a 64bit version? It already has, just compile for amd64 (and use yasm in place of nasm, there is no 64-bit nasm)

Radek

[evade]

I will be upgrading this week to an 3200+ AMD64 chip and board. I currently run linux and compile xvid with gcc and nasm. What can I do to optimize my environment (aside from running a 64-Bit kernel and GCC 3.3) to realize the most improvement.

Is yasm available for linux?

[Danzel]

@evade
reading the yasm page ( http://www.tortall.net/projects/yasm/ ) would of answered your question.

Currently compilable on UNIX (FreeBSD and Linux tested, GNU configure based autoconfiguration), as well as on DOS (using DJGPP) and on Windows (using Visual C++ or CygWin). Unfortunately, the Makefile's for non-UNIX (and Cygwin) are not well maintained. We'll unbreak them for each release.

[evade]

Ah yes a classic case of posting before I think.

I'm sure I will be fine.

I will run some comparsion benchmarks when I get this working.

[crusty]

@Bear

It's quite simple why your mp system isn't faster:
Xvid isn't a multithreaded process.
It just uses one cpu. The other one is sitting there shoving bits around for other processes, like your OS, Disk access etc.
If you want to properly use your mp system, cut the clip your encoding into half and use two processes of Vdub and Xvid simaltaneously. (make sure to use a copy of Vdub out of another directory).

[Bear]

crusty, That's right. I am running 2 Vdub to encode the movies. I found one Vdub only us around 50-59% cpu resources. So I start 2 threads now.

As for mp system, since I use procoder to encode movies in that sys. And procoder claims that they support multiple processors. But still, the speed is more or less with the P4. Actually, I won't buy a dual sys for encoding in the future, a lot of noice, heat...... It's better to stick with P4.

[ToiletDuck]

better to stick with a P4 bah.... I'm on a dual opteron system. 240's running at 1.52ghz instead of 1.4 just by using the software that comes with em. My 3dmark with my 9700pro went from a 4267 on my dual XP 2200+ system to 5603. My gaming alone is slick as glass on anything I play even though I'm only at 1.5ghz. They processors haven't hit 40 degrees celcius yet even when I leave them running all night long crunching and folding. They aren't as hot as XP's in my opinion. I have the K8T800 MSI board and love it. I'd like to see a 64bit download of xvid though :D so I could put on windows 2003 server 64bit. I have it and it seems to work ok. Just nothing 64bit yet to try on it. All you have to do is take 60seconds to fill out a form and microsoft will give it to you for 180days. Since I usually reformat every month or so that's no problem :)

[Bear]

ToiletDuck, will encoding speed a lot faster if using AMD64 on Win 64 platform?

I am also consider to get a AMD 64 to replace my dual amd sys.

[kadajawi]

compile it :) And give us the results compared with the unoptimized version.

[Bear]

But I don't have a amd 64 sys now :)

[crusty]

Perhaps someone could make a good 64bits build of the current CVS and put it up somewhere for download?
I think there are atleast some people out there that are willing to test it on their 64bits system....

BTW: Do you need a 64bits OS to properly use 64bits extensions of a Xvid codec? Since the main coding process is compiled with an assembler, won't it just be enough for the assembler to use 64bits?

[Episode]

Originally posted by crusty
Perhaps someone could make a good 64bits build of the current CVS and put it up somewhere for download?
I think there are atleast some people out there that are willing to test it on their 64bits system....

BTW: Do you need a 64bits OS to properly use 64bits extensions of a Xvid codec? Since the main coding process is compiled with an assembler, won't it just be enough for the assembler to use 64bits?

I think you need 64-bit operating system to enable 64-bit mode on your processor. Opteron and Athlon 64 will work as 32-bit processors under 32-bit systems and I doubt that 64-bit code wouldn't even run on 32-bit OS. And even if you could, it would still run on 32-bit mode. Fortunately, if you have new Opteron or Athlon 64 system, you can now download a trial version of Windows XP 64-bit version for free from Microsoft's webpage (http://www.microsoft.com/windowsxp/64bit/downloads/upgrade.asp). The trial period lasts 360 days, which is more than enough for testing :)

[Prettz]

Originally posted by crusty
@Bear

It's quite simple why your mp system isn't faster:
Xvid isn't a multithreaded process.
It just uses one cpu. The other one is sitting there shoving bits around for other processes, like your OS, Disk access etc.
If you want to properly use your mp system, cut the clip your encoding into half and use two processes of Vdub and Xvid simaltaneously. (make sure to use a copy of Vdub out of another directory).
That would be a good idea, BUT you have to already know the target sizes for the two sections of the clip. That doesn't sound very safe to me, because if you're playing around with the other xvid settings, your settings may cause xvid to either need more bits for one section or be able to encode that section with fewer bits (meaning an undersized file).

[ToiletDuck]

first of all when it comes to ripping I'm the kind of person that has to use the "guides" to make sure I'm doing it right. I follow the instructions. Anything that requires compiling I don't know how to do. Feel free to pm me a message with instructions and I will gladly do a few test and post results of ripping 32 bit and ripping 64bit. I'll dual boot. Also another thing is I just got the xvid 1.0 rc2 and it's just binaries. What do i have to do to update my xvid. If it doesn't come with a .exe on the end of it i'm floating in the water.... I'm stupid.
Duck

[Joe Fenton]

I'm still setting up XP64. Once I reinstall Visual Studio and the Server 2003 DDK, I might try compiling xvid to 64 bit.

A note to developers! The Intel compiler deliberately disallows SSE2 on non-Intel chips! This behavior was noted here (http://groups.google.ca/groups?dq=&hl=en&lr=&ie=UTF-8&oe=UTF-8&threadm=a13e403a.0402091438.14018f5a%40posting.google.com&rnum=1&prev=/groups%3Fdq%3D%26hl%3Den%26lr%3D%26ie%3DUTF-8%26oe%3DUTF-8%26group%3Dcomp.arch%26selm%3Da13e403a.0402091438.14018f5a%2540posting.google.com).

It makes you wonder - the Athlon 64 ties the P4EE in most benchmarks and beats it in many, so why are a few things like MPG4 faster for the P4? Perhaps because of people compiling with the Intel compiler? I would really like to know what compiler the DivX people use as that is one benchmark that P4 people keep crowing over. As noted on the link above, some things are over 20% faster if they can run the code that Intel disallows on AMD processors.

[sysKin]

Originally posted by Joe Fenton
A note to developers! The Intel compiler deliberately disallows SSE2 on non-Intel chips! This behavior was noted here (http://groups.google.ca/groups?dq=&hl=en&lr=&ie=UTF-8&oe=UTF-8&threadm=a13e403a.0402091438.14018f5a%40posting.google.com&rnum=1&prev=/groups%3Fdq%3D%26hl%3Den%26lr%3D%26ie%3DUTF-8%26oe%3DUTF-8%26group%3Dcomp.arch%26selm%3Da13e403a.0402091438.14018f5a%2540posting.google.com).
Now that is pure evil. However, it does not affect xvid much - it's just that icl will not create SSE2 code on its own (from c source). It doesn't mean that xvid compiled this way will not use SSE2 code - whole xvid's assembler stuff is compiled by NASM, not ICL.

The thing that really makes me wonder is whether assembler sources will have to be modified. They do assume that pointers or integers are 32-bit - integers can be forced to be 32-bit in C too (when neccessary), but pointers cannot.

I sure hope we won't have to maitain a second set of assembler functions for amd64...

:)

[Aiorus]

Well, maybe you can check http://www.anandtech.com/systems/showdoc.html?i=1961&p=4 (this Anandtech preview of XP64). It looks pretty impressive considering it was a non 64 bit optimized version, on a beta OD and Beta drivers

[ToiletDuck]

can someone post me a link to where I can find out how to compile my own 64bit version of xvid and such. I'd like to start running some test.

[Joe Fenton]

You can find a list of all the AMD64 resources right here (http://www.amd.com/us-en/Processors/DevelopWithAMD/0,,30_2252_9044,00.html).

You'll find toward the bottom, a porting FAQ, the processor programming manuals, and info on programming for the AMD64 in Windows. If you create a developer account (free), you can get the CodeAnalyst program to profile your 64-bit code.

[ToiletDuck]

is it possible for someone else to recompile one for me then I just download it and use it? If anyone wants to compile a 64bit xvid 1.0 i'd love to have it :D
Duck

[sysKin]

Originally posted by ToiletDuck
is it possible for someone else to recompile one for me then I just download it and use it? If anyone wants to compile a 64bit xvid 1.0 i'd love to have it :D
Duck The easiest solution is just ./configure and make and su and make install. This will, however, activate "generic" (unknown) 64-bit architecture, without any assembler optimizations.

I don't know how to use yasm to compile x86 assembler and whether it works.

Radek

[evade]

I tried to do that and yasm choked but it did build a couple of objects first.

First I ran ./configure, it detected my x86_64 architecture.

I modified the configure generated platform.inc file with the folloiwng lines:

AS=yasm
AFLAGS=-I$(<D)/ -f elf
ASSEMBLY_EXTENSION=asm

ASSEMBLY_SOURCES=$(SRC_IA32)
ASSEMBLY_OBJECTS=$(SRC_IA32:.asm=.o)

and ran make and got the following output:

dave@hammer generic $ make
D: =build
C: ./decoder.c
C: ./encoder.c
C: ./xvid.c
C: bitstream/bitstream.c
C: bitstream/cbp.c
C: bitstream/mbcoding.c
C: dct/fdct.c
C: dct/idct.c
C: dct/simple_idct.c
C: image/colorspace.c
C: image/image.c
C: image/interpolate8x8.c
C: image/font.c
C: image/postprocessing.c
C: image/qpel.c
C: image/reduced.c
C: motion/estimation_bvop.c
C: motion/estimation_common.c
C: motion/estimation_gmc.c
C: motion/estimation_pvop.c
C: motion/estimation_rd_based.c
C: motion/gmc.c
C: motion/motion_comp.c
C: motion/vop_type_decision.c
C: motion/sad.c
C: prediction/mbprediction.c
C: plugins/plugin_single.c
C: plugins/plugin_2pass1.c
C: plugins/plugin_2pass2.c
C: plugins/plugin_lumimasking.c
C: plugins/plugin_dump.c
C: plugins/plugin_psnr.c
C: quant/quant_h263.c
C: quant/quant_matrix.c
C: quant/quant_mpeg.c
C: utils/emms.c
C: utils/mbtransquant.c
C: utils/mem_align.c
C: utils/mem_transfer.c
C: utils/timer.c
A: bitstream/x86_asm/cbp_3dne.asm
A: bitstream/x86_asm/cbp_mmx.asm
../../src/bitstream/x86_asm/cbp_mmx.asm:49: warning: Unrecognized qualifier `data'
A: bitstream/x86_asm/cbp_sse2.asm
../../src/bitstream/x86_asm/cbp_sse2.asm:69: warning: Unrecognized qualifier `data'
A: dct/x86_asm/fdct_mmx_ffmpeg.asm
../../src/dct/x86_asm/fdct_mmx_ffmpeg.asm:76: warning: Unrecognized qualifier `data'
../../src/dct/x86_asm/fdct_mmx_ffmpeg.asm:142: expression syntax error
make: *** [dct/x86_asm/fdct_mmx_ffmpeg.o] Error 1