SHOOTOUT x264 new optimized builds: G5 vs. Core2Duo

[PhilG5]

here's an updated version of this posting of mine comparing an optimized build of x264 to the build used in the most recent version of ffmpegX (0.0.9x r2) and a reply to this posting written some days ago

i have downloaded from SVN the sources for build 618 and made two binaries

one optmized for Intel Core (2) Duo and one optimized for G5

both machines are running at nearly the same clock speed (30 MHz are negligible) so results are nicely comparable. They both have 2 GB of RAM and are running a fully patched and up-to-date OS X 10.4.8

here are the results:

FILE: Lord of the Rings 3:02 mins VOB

SETTINGS: x264 iPod 640w setting - 900 kbit/s - CBR - 2-pass - i4x4 - AVC Level 3.0 - QMin=15 - QMax=51 - no crop

MacBook Pro Core2Duo Dual 2.33 GHz

custom build (618)
pass 1: 43.35 fps
pass 2: 49.25 fps

ffmpegX 0.0.9x r2 build (569)
pass 1: 39.58
pass 2: 43.22

PowerMac G5 "Antares" or "late 2005" Dual 2.3 GHz

custom build (618)
pass 1: 24.76 fps
pass 2: 26.53 fps

ffmpegX 0.0.9x r2 build (569)
pass 1: 19.57 fps
pass 2: 21.35 fps

RESULT COMPARAISON

Core 2 Duo vs. G5 (both optimized): Intel Core2Duo is more than 80% faster than the G5

Core2Duo (build 569) vs. Core2Duo (custom 618): the custom build is 11% faster

G5 (build 569) vs. G5 (custom 618): the custom build is 25% faster

DISCUSSION

the Core2Duo is the clear winner. at the same clock speed, it encodes nearly twice as fast than the latest G5 processor. in my opinion this is due to

- 4 MB of shared L2 cache so even 1 single core can use all of it (the "Antares" G5 offers only 1MB for each core separately)
- MMX/SSE/SSE2 x264 is known to be heavily optimized for these Intel SIMD extensions whereas the Atlivec/VMX support is not so well implemented, especially in older x264 builds (the latest build offers a noticeable speed boost, possibly due to better Altivec/VMX support?)

the good news for G5 owners: with optimized and recent builds of x264 you are able to boost your encoding performance by 25%!!!

this again points out the necessity for heavily tuned and up-to-date x264 binaries if you want to maximize your encoding performance.

[bucho]

Any chance of sharing those binaries? Especially for the Core2Duo.

[PhilG5]

damn...file upload is not working in this forum anymore... any idea where to put it so you can download it?

[bucho]

Just sent you my email address in a PM. Perhaps I should just learn how to 'brew' my own binaries.

Thx.

[Scallywag]

damn...file upload is not working in this forum anymore... any idea where to put it so you can download it?

How about one of those free download services like RapidShare or Sendit or MegaUpload?

[PhilG5]

ok i've uploaded a binary that contains BOTH the intel core 2 duo and G5 optimized build.

wow file upload is back again!!!


x264-intel-g5.zip

[edoud]

Will it also go faster on a G4 ?

[PhilG5]

sorry, i have no G4 around here for testing but since i explicitely built the power pc binary for G5 i have some doubts the binary will execute at all on a G4...

the only thing you can do is try it out...i don't think it's gonna hurt your machine (but don't blame me for anything )

[edoud]

it does'nt work and my iBook is totally down !
(jocking )

[rumplestiltskin]

ok i've uploaded a binary that contains BOTH the intel core 2 duo and G5...snip

Phil,

Thanks for your efforts. I've DL'd your binaries and will try the Intel version on my early '06 iMac. I'll post a comparison that should confirm your results.

Is this something you want to have major include in future releases of ffmpegX? If so, might be good to send him an eMail so he's clear on your intentions for the code.

[caozinho]

Great work Phil and thanks for the comparison.

From what I've seen on my Dual2GHz there is even a bigger boost in performance when encoding HD content. It seems that the bigger the picture, the more benefit is taken from the optimized compiling.
I didn't clock the difference, but your build really ignited the afterburner on my 720p encoding.
This performance gain allows me to choose a better motion estimation method and still the whole process is definitely faster than using the stock build.

Cheers, C.

[PhilG5]

nice to hear that HD encoding works even better... mabye there IS some hope for the PowerPc platform after all

Is this something you want to have major include in future releases of ffmpegX? If so, might be good to send him an eMail so he's clear on your intentions for the code.

yeah what i really miss in ffmpegX ist multithreading...and it worked some versions ago for x264(!!!)...now, even when x264 is compiled with pthread support, it only uses 1 single thread and therefore runs only on 1 core. ffmpeg usually takes around 50-60% of the second core, so there are free ressources that could very well be used to boost the fps...
and what's the status on ffmpegX 1.0.0? was anounced ages ago...

[caozinho]

@Phil

Hmm, looking at my processes while encoding, x264 uses around 120% while ffmpeg takes around your mentioned 50-60%. So adding all other small tasks running in the back my system doesn't waste any resources at all.
I'm running a dual CPU, not dual Core. I don't know if there is a difference, just for your interest.

Cheers, C.

[PhilG5]

ummm..maybe it's an intel only issue...i believe you are working on a ppc machine? so maybe the multithreading issue is intel only...which would make them look even better when multithreaded maybe

[rumplestiltskin]

Where are you placing the "new" binary; inside the Resources folder contained within the app?

[PhilG5]

yes that's the correct path

ffmpegX.app > Contents > Resources

you may rename the original x264 binary in case you might need it later

[rumplestiltskin]

Just finished a rough test comparing the "original" and "optimized" x264 binary. I duplicated ffmpegX and switched the x264 binary found in the Resources folder with the new "optimized" binary.

I then ran the same AVI through both the original and modified versions of ffmpegX using exactly the same encodings (x264 640w for iPod, 550kbps, 128kb audio, all other details left as is). End result: No significant differences. The optimized binary produced a file that was .3MB smaller (407.7MB/407.4MB). Encoding times: 1:11 for the standard binary; 1:10 for the optimized binary.

Examining some frames showed the standard binary producing slightly better quality (fewer artifacts) but it took viewing the 640x352 frame at 2X size in order to spot this. I'll emphasize that the differences would not be noticeable on your TV.

I must be doing something wrong because (unless I misread the original post) I should see shorter encoding time.

Advice and suggestions welcomed!

Test Mac: Intel iMac 2.0GHz CoreDuo (Early 2006), 2GB RAM.

[curtis]

Do you think the next optimization will improve speed for G4 Mac too ?
because on my 1,42 GHz it's a little slow

[apollo8fan]

PhilG5:

Thanks for the build. While I don't have hard physical numbers, the performance boost on my 1.8GHz G5 is certainly enough to notice. I'm gonna keep your build in my arsenal!

Curtis:

Major's probably already compiling everything for the G4, hence the requirement for that processor over the G3.

[cvkB]

I think this thread is the perfect example of why the G5 will never fully realize its potential. Everybody compiles for G4 to keep compatibility high, yet the future holds intel, so which devs are gonna stop and make a 'special' G5/intel universal binaries of the their application? Not many. This is sad. Granted, I am hopeful. Leopard IS bring full 64bit support for the 970 architecture as far as I can tell. All G5's may see a boost at that time. My fingers are crossed.

Thanks PhilG5

[rinconj]

Can you also post x.264 quicktime component optimized for G5 and core duo, core 2 duo too?
I haven't been able to build X.264 on either G5 or core duo yet. Thank you so much!

ok i've uploaded a binary that contains BOTH the intel core 2 duo and G5 optimized build.

wow file upload is back again!!!


x264-intel-g5.zip

[reallynotnick]

I just installed this and all I can say is WOW. I got a 20% speed increase (7:22, down to 6:03), at 848x480 2000kbs, with 128kbs mp3 audio. Thanks SOOO much, now when the xbox 360 adds H.264 support on 5/7/2007, I will be able to whip these files out.

Looking at this forum post, it is kind of old. By any chance is there an even newer version?


This really needs to be included in ffmpegx.

[fatespawn]

Phil,

I hate to keep bumping the topic as I'm sure you have other things on your plate, but have you ever made any progress on the quicktime component or an optimized G4 version? I'm trying to eek some life out of my poor "aging" PPC computers.

I've used your x264 plugin since the day you originally posted and have loved it. Makes a big difference on my G5. Thanks for the work.

-fate

[reallynotnick]

I hate to revive an old thread but I really loved this optimized version and figured there had to be some improvements in the past year to the x.264 codec that would warrant for a new update.

How hard is it to make a G5 optimized version, because I would not mind being able to do it myself when a new version came out as clearly it is not going to be packaged with a new version of FFmpegx any time soon.