Massive performance drop - Bad Axe 2 Q6600 & After effects CS3

[the-negative]

I know I'm gonna get a LOT of flak for writing this, but:

 

Try Xbench. If the PD has higher results than your Q6600, you need to boot into Windows (OS X is too much of a wash here) and see. I don't see Natit hogging 50+% of idle CPU time, too. (Have you got MenuMeters? Good tool to check for idle usage)

[mifki]

beamsync does nothing for raw speed. HPET is barely used in the patched kernels, im saying it is your kernel. Try going for a bare minimum patched kernel, only sse3 and auto-fsb. No eed for sse2 and all the other junk. Also, openGL will have no effect whatsoever, unless you have a Nvidia quadro card. As the OpenGL option is really only used for quick renders and real-time previews. cpus=4 has no effect any more either. xbench is {censored}, its a useless benchmark. the core 2 quad should work better regardless. I say do a clean install, only select what you need, dont have 100 little apps running in the background (and other useless widgets). also make sure you cpu fsb is correct (check this with a sysctl)

 

also natit isnt a driver, it only does its work at boot up. and is never used till the next reboot. look at activity monitor and check idle cpu usage.

remove any speedstep kexts you might have installed as they are useless for a desktop. copy the preferences file from your mb to your quad and see if anything changes

and if all that fails, do a clean install, and select only what you need.

[rhalls]

I agree with Kiko, it seems you have a probelm with your install.

I did some tests a few mintes ago:

30 seconds AfterFX File with not very much stuff, some videos, text and so on, from an older project, rendered uncompressed. AfterFX 8.01

Q6600/2.4 GHz/BX2/4GB/10.4.9: 198 seconds

Core2Duo/2.13 GHz/Asrock/3GB/10.4.8: 301 seconds

MacBook, Core2Duo/2 GHz/Apple/2GB/10.4.10: 447 seconds.

 

So the Quad does what it should on the BX2 - its not so fast a one could expect, but there is a lot of coordinaton to do for using 4 cores, which does not result in double speed.

[solaar]

Cheers people!

 

True say. I'll do a couple of fresh installs on my test HDs with Jas versions /semthex only. I've got two installs right now, uphuck 10.4.9 which I upgraded to 10.4.10 koolkal and the 10.4.10 kalyway Intel sse3 only. They must be both dodgy then as I definitely only installed the barest minimum.

 

rhalls, is there a particular reason why you didn't go 10.4.10 on your hacks?

[rhalls]

Yes, theres a very good reason why i didnt go to 10.4.10: Both machines are working very well. So why should i upgrade???

[solaar]

You've got a point there rhalls.

 

All right, I installed Jas 10.4.9 Intel SSE3 only/ semthex kernel with definitely the barest minimum, meaning just the basic install plus natit 0.2.

I ran my AE animation and it finished in a devastating 72 minutes

 

I'm totally stumped for now but will do another test with a different config when I have time again. I will post my results here.

 

Otherwise I will use my lil Macbook just for AE rendering lol

 

BTW what's the deal with the BIOS upgrade on the bad axe everybody's talking about?

 

Thanks again folks for all the replies. Much appreciated.

[solaar]

I did a few more test installs. I think I covered about everything that was suggested here but the result remains the same - very poor AE render performance.

 

The geekbench result was consistent with my previous 945/ D940 rig - 2500 (D940) vs 4700 (Q6600). Even if these benchmark testers may be rubbish, at least I get a result that seems to be consistent in comparison.

 

The last resort is to install XP and run a few more tests. The problem is that the results will most likely be conclusive and definitely not sufficient for an RMA.

 

btw I had to install this hacked AppleSMBIOS kext to be able to start AE. Without it AE 8.0.0 final crashes right at startup. On my old system I never installed the kext but AE beta would run without it. Is there a chance that could be it?

 

What would you do in a situation like that? Keep doing trial and error or cut your losses?

[rhalls]

Hmm. hard to say... But in my case, i would just get another mainboard.

I have some usb problems with my ASrock Board, that may result of my old BIOS - but with a newer BIOS, the CPU Fan is much louder (bad Fan Control). So i have to decide to workaround my USB Problems or to get a loud machine. My cinclusion: I bought a Gigabyte P35 DS3 Board and will change the board. I am goin to sell the ASrock Board, so my money loss is minimal.

I think for you it would be best to just test in windows if your board is o.k. eventually for a windows install, sell it on ebay and get a new one. Maybe you will consider buying one of those recent P35 Boards, they are chaper than the BX2 (i know, the BX2 CAN run smooth and without problems, but its an "old" board)

[solaar]

Cheers mate. I was also thinking of RMA-ing the mobo but...

...here are some more test results.

 

I installed XP SP2 on a 40GB antiquity I found in a drawer. Then I put AE CS3 on it and ran my render process.

---> 12 minutes

 

That's what I'm talking about. At least it should prove that my mobo and processor are ok.

 

Now it all makes sense.

D940 - 30 min.

Macbook 2ghz T2500 - 18-20 min.

Q6600 - 12 minutes (in XP)

 

The question remains, how do I get the same performance out of this bloody thing in OSX?

I definitely want to avoid to go back to XP just for one app. I already felt a bit dirty after installing it

[rhalls]

How big is you AE File? Or: Can you create a small test AE File that you can upload somewhere, so i can test it on my equipment, for checking out our performance differences? Its odd what your results are...

[solaar]

Thanks man.

Check your inbox. I PM-ed you.

 

The file is really tiny, 50KB or so.

[kle500]

Dear sollar, i am reading about your problem really carefully and although i can't think of something that causes this, i think it would be very educating for all of us , if you could upload your test file (on any upload service e.x. www.rapidshare.com), so we can give a try and post our results.

 

On my machine, having 2 HDD, 10.4.10 (XxX 10.4.10 dvd) on one and XP Sp2 on the other, i see big difference in performance and lag, although i have changed many many distros and kernels.

 

Waiting to test your file!

[solaar]

All right kle, will do.

Rapidshare is anonymous, isn't it?

I would usually put it on our domain server but well... you know where we are here and murphy's law...

 

here you go

http://rapidshare.com/files/63577836/ParticleTest.zip.html

 

Bear in mind that this is a particle animation from just one frame, green screened and only 320x240. That's why the file is so small. The strain on the CPU is massive though. So in the beginning you will see just a black frame but no worries, it takes a while for the particles to build up.

 

Thanks to everybody who'd like to do some testing. Much appreciated.

It can perhaps provide some valuable clues...

 

Just one more thing.

Hello mods, I hope this doesn't drift too much off topic for the hardware forum but I reckon it would be very interesting to know how the AE rendering behaves on a quad Xeon Mac Pro. Just in case a proud owner cares to do a test...

 

cheers

[rhalls]

O.K., i tested the file....

My Core2Duo Machine @ 2.13 GHz was finished in about 21 minutes. My Quad... I stopped it after 40 Minutes cause i needed the machine for some 3d rendering.... wasnt patient let it finish.

It would be interesting if a MacPro 4 core has the same problem...

It seems to me, as AFX 8 has problems with some of its effects on mac. i dont think its a OSx86 problem, i bet it persits on MacPros, too.

If anyone here who reads this has a macpro, would be nice if he tests the file.

[marliwahoo]

Why not try xbench and post the score ?

[solaar]

Why not try xbench and post the score ?

 

Benchmark scores seem to have no utter relevance in this. As I mentioned before, in geekbench I had 2500 on the D940 & 945 mobo and 4700 on the Q6600 & bad axe.

xbench was 120 for the D940 if I remember well and 160 for the Q6600.

So the benchmarks look consistent in relation to each other. The render times in AE don't.

 

As rhalls says it seems to boil down to a problem specifically with AE8 on a quad core in OSX. (Perhaps even more specific than that.)

 

O.K., i tested the file....My Core2Duo Machine @ 2.13 GHz was finished in about 21 minutes. My Quad... I stopped it after 40 Minutes cause i needed the machine for some 3d rendering.... wasnt patient let it finish.It would be interesting if a MacPro 4 core has the same problem...It seems to me, as AFX 8 has problems with some of its effects on mac. i dont think its a OSx86 problem, i bet it persits on MacPros, too.If anyone here who reads this has a macpro, would be nice if he tests the file.

Thanks a lot man. Much appreciated.

 

The 21 minutes are really acceptable for your Core 2 duo. I might have to correct myself, it could have been 20-23 minutes on my Macbook but it's very consistent.

[solaar]

I found a thread in the adobe forums where somebody is complaining about poor performance in CS3 on a Mac. It seems like the one 'defending his ground' is most likely an adobe rep. IMHO there's a lot of BS being written in that thread.

 

http://adobeforums.com/cgi-bin/webx?14@@.3bc466c7

 

There is probably something system-specific or project-specific in your case. For example, there are a few effects (Particle Playground, Timewarp and other time-related stuff) which turn off multi-processing because by their nature things would slow-down instead of the opposite.

I beg your pardon?

[rhalls]

"There is probably something system-specific or project-specific in your case. For example, there are a few effects (Particle Playground, Timewarp and other time-related stuff) which turn off multi-processing because by their nature things would slow-down instead of the opposite."

 

Nice saying. Was it one of the adobe stuff who said that? Cause on the Dual Cores it seems not to be turned off (multiprocessing) but on the quads it seems so.

 

But your problem seems to be, hmmm..., not solved but maybe explained. Not your board, not your OS, AFX is to blame. Doesnt really help but may be fixed in the future.

[kle500]

I did that test and guess what?

On my Osx the test finished in exactly 21 minutes and 35 seconds.

 

Now the funny part:

I did the same test on my other Hdd on the same machine, but on XP, with same AE CS3 and.............

Finished in 12 minutes and 33 seconds.

 

What's up with that?

Can't understand that big difference.

Rendered Output (Render Que)using the default settings in both platforms.

 

Will do the same test in my other Xp machine and post results latter.

But the biggest challenge will be if someone with a MacPro could post his results.

EDIT: On my other XP machine, the render finished at: 14 minutes and 37 seconds

 

This machine's specs:

Intel DG965

Core2Duo E6300

3x1gb 667 Kingston Value Ram

Onboard GMA950

5 HDD's

 

On all setups i used the same AE settings with Multiprocessor enabled.

So it looks like a MacOs AE problem

 

Regards

[solaar]

Cheers mates. Nice one kle.

It confirms my results.As rhalls said before, I would now also bet that a Mac Pro will show the same poor performance unless bootcamped into XP.

 

Let's hope this is fixed in 8.1 or whichever release comes next. I'm actually glad I didn't pay yet for an upgrade from CS2. If we can find somebody with a Mac Pro who comes to similar poor results as the Q6600 I will go the 'official' way and drop a trouble ticket at Adobe.

 

It's weird though that nobody has ever 'formally' complained in a way that Adobe puts it in their FAQ. What do all the commercial studios who invested in Intel Mac Pros do? One thing is sure, I will pay for neither CS3 nor a Mac Pro until this is fixed

 

Nice saying. Was it one of the adobe stuff who said that? Cause on the Dual Cores it seems not to be turned off (multiprocessing) but on the quads it seems so.

I suppose it was a staff member, at least a forum mod. Doesn't make sense, does it. How come it works properly in XP?

 

The thing is, this is just a 4 second clip at 320x240. Now imagine the same at hi def and longer duration.... When it finishes rendering you'll have already forgotten what your initial idea for that clip was

Not even the 'home enthusiast' has that much time to waste.

[rhalls]

Maybe you will check out if Apples Motion 3 will do the job better for you. If it has a similar particle effect.

If i were you, i would do AfterFX projects only in windows.

But wait: Ever tried that with a VM-Ware emulated Windows? VM ware uses multicore systems (parallels doesnt) which would prevent you from rebooting.

[kle500]

If we can find somebody with a Mac Pro who comes to similar poor results as the Q6600 I will go the 'official' way and drop a trouble ticket at Adobe.

 

What do all the commercial studios who invested in Intel Mac Pros do?

 

When it finishes rendering you'll have already forgotten what your initial idea for that clip was

Not even the 'home enthusiast' has that much time to waste.

 

You have absolutely right on the above.

Let's keep this topic fresh and i am sure, sooner or latter someone will come up with a MacPro result.

 

Have you tried other plugins performance?

 

One Question though:

Why my E6400 has the same render time as your Q6600 ?

 

Maybe you will check out if Apples Motion 3 will do the job better for you. If it has a similar particle effect.

If i were you, i would do AfterFX projects only in windows.

But wait: Ever tried that with a VM-Ware emulated Windows? VM ware uses multicore systems (parallels doesnt) which would prevent you from rebooting.

I don't think Motion follows the same road as AE, so if someone is used to work with AE, he may have a little problem on doing the same things he used to.

 

On my machine, parallels have quite some difference in performance compared to real XP boot.

But i can't be 100% correct as i feel my Hack a little sluggish in everything, except benchmarks

[solaar]

One Question though:

Why my E6400 has the same render time as your Q6600 ?

What rhalls mentioned before, I've noticed the same. In a quad, AE never exceeds 50% CPU usage per core. I've seen it spike 200% on my Q6600 but never above that. It's almost like they've coded a hand brake into it...

 

On my dual core Macbook I see spikes of 160% and sometimes a bit more, so you get around 80% (which I could live with on a quad).

 

On top of that, looking at the discrepancies between OSX and XP, I can only assume the OSX version must have been poorly developed.

 

I'll definitely try the VM thing. Would have been nice to have it native but well... one day hopefully.

 

What I just don't get is that quad core technology is nothing new. What about quad G5? I know it's a different platform and everything but before I release something and want people to pay for it, I test it properly and make sure it's worth the money...

[kle500]

And why the heck http://www.barefeats.com/octopro4.html

came to the conclusion that AE CS3 is worth the upgrade to 8 core MacPro ???

Have you tried the "TotalBenchmark" test ?

 

I will try it and post later on

[solaar]

No I haven't tried it yet.

 

I'm not a tech head so my question is naive: can code be written and optimised for just one specific processor type - xeon in this case - and will perform poorly on a different type although it *might* even have superior performance?

 

(To be honest, that would be a totally exotic concept for me.)

 

Man, we gotta find a helpful Mac Pro owner...

 

correction: I mentioned earlier a 35 minutes render on the Q6600 in OSX. My mistake, it was 1hour 35 min. The 50 minutes I mentioned before are wrong too. I must have stopped the process at about 50%. Sorry about the confusion... Fact remains, things are disproportionally slow in OSX with a Q6600.