Processors tests - XviD

[Suchy]

Hi.

I began mini project.
The reasons, why I made it, are:
- answer for question "How XviD 1.2 works on various CPUs?"
- answer for question "Which CPU is the best suited for XviD (not theretically, but in practice)?"
- answer for question "Which solution is the best way when You want buy processor for compression (for ex. if You want buy CPU in midde cost range, which one will be optimal or if You shoul overclock it)?"


Test results, instruction (actually in polish only, but images are included), needed software, etc., are here (http://www.videoaudio.pl/forum/index.php?showtopic=1586)
Please, test Your CPU with settings form instruction (link to instruction (http://www.manganime.avx.pl/Xvid_testy/instrukcja.rar)).


Comments, questions and opinions can be wrote here (in this post) or here (http://www.videoaudio.pl/forum/index.php?showtopic=1587)

Source for tests is here (http://www.manganime.avx.pl/Xvid_testy/Sample.avi)

[shon3i]

Using full processing mode are not change anything, just degrade speed

[foxyshadis]

The best CPU for xvid is a 8-core 9.6 GHz UltraSparc T1. A few of them in a farm if you can swing it. In general, 1.2 works better the more speed and more cores you throw at it, although it does have certiain optimizations for x86 and x64 processors.

As for best CPU price/performance, there are a number of dual-core chips out or coming out in the next two months. Check out benchmarks and overclocker forums and make up your own mind.

[Suchy]

Yes, but still there are group of users whose can't buy dual core CPU, because they haven't enough money (for ex. me :) ).

P.S.
I this test, P4 (one core but HT) is faster (about 35 second faster in 25 sec source) with Num. of thread = 2

[shon3i]

@foxyshadis, that is best CPU for everything

[Suchy]

LOL

P.S.
I don't ask how much is cost ;)

[foxyshadis]

Best budget CPU for encoding is the best budget CPU for everything else (although a large cache size and fast memory also helps encoding a lot more than some other apps): sempr0n/A64, X2 3800+, whatever you can afford, or a PD 920 after the next round of cuts. As usual, check pricewatch (http://www.pricewatch.com/motherboard_combos).

Apparently budget Conroes are coming, too, in the $200-250 range, even cheaper for single cores. Right now Conroe appears to be the ultimate x86 chips (engineering samples have OC'd to 4.6GHz and broken benchmark records), but we'll have to see whether that holds for encoding when they arrive.

Really, the presence of SSE2, large cache, or a second core are the only major differentiators in how xvid seems to perform across various chips, in my experience. I'm trying to put together a benchmark suite to find out for sure though. ;)

[edit]
Ah, found a good test of current chips.
http://www.matbe.com/articles/lire/298/pc-desktop-le-core-duo-face-a-la-concurrence/page16.php

[Suchy]

Oh... this test is interresing. Thanks

P.S.
Hmm AMD 64 3800+ is better than AMD 64 4800+ ?
Werid..

[shakey]

The 4800 is better than the 3800. The numbers given for each cpu are encoding times, so lower is better. The 3800 does beat the 4600 however. If I remember correctly, one of them has 512kb cache/cpu, the other has 1mb, which could be the reason.

[foxyshadis]

I don't know if that was just an anomaly and they swapped 3800 & 4400 in that test, as all the other tests show it slower, but you never can tell where something will show unexpected awesomeness.

(3800 is 2.0/1M, 4400 is 2.2/2M and 4800 is 2.4/2M.)

[FabioMTPX]

Actually, for XviD, forcing the program to use MMX/3DNow! optimizations makes the program run faster on K7 family (From Thunderbird, Palomino, Thoroughbred, to Barton). On XviD 1.1.0, 2 - 10% speedup is observed on my machine, when forcing XviD to use MMX, Extended MMX, 3DNow!, and extended 3DNow!. I have AMD AthlonXP 2500+ (no OC).

It may relate to how XviD code is written, but it may also relate to how K7 executes MMX/3DNow/SSE instructions. On K7/K8 the processor can execute 2 64-bit MMX loads per clock cycle, but it can only execute a 128-bit SSE load every 2 clock cycles. For more details, there are some detailed benchmarks on www.digit-life.com. The articles are quite old, but they provide some interesting insights into the processor architecture.

[squid_80]

Actually, for XviD, forcing the program to use MMX/3DNow! optimizations makes the program run faster on K7 family (From Thunderbird, Palomino, Thoroughbred, to Barton). On XviD 1.1.0, 2 - 10% speedup is observed on my machine, when forcing XviD to use MMX, Extended MMX, 3DNow!, and extended 3DNow!. I have AMD AthlonXP 2500+ (no OC).
That doesn't make sense. All functions using those instruction sets should be used automatically for an AthlonXP. Note there are no SSE functions, only SSE2 which wouldn't be used in your case.