Performance differences between Metro and Logic on a G5

[Paul]

I recently bought a dual 2gHz G5 and noticed that there is a big difference in cpu usage between Metro and Logic. My G5 has 1GB internal and for a soundcard I am now using the M-audio Audiophile USB. Buffer size = 256 in both Metro and Logic.

When I play the same music (1 track) with the same sound (from the REFX Vanguard plugin, 44.1 kHz), I get the following values:

% user: Logic between 8.5 and 17.5%, Metro between 48.5 and 55.5%

% system: Logic between 2 and 6%, Metro between 2 and 7%

Why is this difference so big? I'm under the impression that Metro will reach my system's limits far earlier than Logic will.

Am I missing another important setting in Metro that will decrease its cpu usage? Or could it be the plugin?

[Jerm]

Paul,

What version of Metro are you using? Is it the G4 version?

The most important thing here is that in Metro make sure you have no ports assigned to internal. Using any internal ports (including the metronome) will turn on Quicktime Musical Instruments and their nasty CPU eating reverb.

The other thing is that Process viewer (and also the program 'top') does not tell the whole story.

I am also not familiar with Vanguard. Does it have multiple outs? A better comparison might be to simply try audio tracks.

[Scoot]

I'd like to run a similar test on a G5 with 128MB of ram. (if its possible)

Then again I'd just like to have a G5.

[Paul]

Jerm,

I'm using 6.2.1 G4 (OS 10.3.4). I've run a test with ten audio tracks and the results are the same again. When I look at the cpu usage of Logic with these 10 tracks, I see a usage of in between 10.9 and 29%. When running the same 10 audio files in Metro, I see values between 97 and 108%, with an occasional peak of 146%.

I don't think I have assigned any ports to internal (I assume this can be checked by simply clicking the output menu in the tracks window).

Do you have any idea what's going on?

[Jerm]

There's something weird going on because 10 audio tracks should not be anywhere near 100%. I wonder what you are using to measure CPU usage?

I will check with my dual G4 1.25 to see what kind of CPU usage I get on 10 audio tracks.

[Scoot]

I can do 10 tracks on a 450mhz machine.

I would expect a G5 to do 50+?
Metro is dual aware?

[Paul]

It's not that I cannot run 10 tracks, it's just that they take up a lot of cpu. I'm measuring this with the activity monitor that can be found in the Applications/Utilities folder.

Why is it not telling 'the whole story'? What do you mean by that?

[Jerm]

Yeah. I can do more than 10 tracks on my Beige G3 266 too but that is in OS 9.

[Jerm]

...activity monitor that can be found in the Applications/Utilities folder.Why is it not telling 'the whole story'? What do you mean by that?

Well, if you check out activity monitor while clicking in a menu you will see that it reports Metro uses 100% of the cpu. This is because, at the current time, Metro does not use carbon-events fully and uses the older event model. This is because of added compatibility. In the future Metro may move to carbon-events.

Activity monitor also and more importantly, does not tell you how much of the cpu is being used in an audio's real-time task which is the most crucial for audio applications since this runs at a higher priority than UI types of tasks.

A much better program for determining real-time usage is Apple's Shark which is part of the CHUD tools.

[Paul]

Same story with the Powercore Virus. When I'm running that one and a few audio tracks, I get a hard disk too slow message.

Are there any other settings besides the audio packet buffer size that are important in Metro?

[Jerm]

It sounds like you have a powercore? Have you tried disabling the hardware? Supposedly 'virus' is supposed run without any drain on the CPU.

Perhaps the problem is not CPU but rather firewire bandwidth? What exactly is your hardware setup?

[Jerm]

Okay after a little research I have the answers.

I went and loaded 10 unique audio files which is important to avoid any caching that might otherwise be involved. I then launched activity monitor and in Metro I pressed play. Lo and behold just like Paul says the CPU rating goes up to over 100%, which is in itself quite bizarre (and inaccurate of course). Then I went to the tracks window and turned off the vu's and closed the mixer window to turn off the mixer vu's. Still not much change. I then turned off playback scrolling and voila Activity monitor started reporting around 18% CPU.

Then it came back to me. In order to get the kind of Oscilliscope performance when zoomed in to the sample level I had to do some tinkering and this does use up a bit of processing. I think there is some optimizing that I can do here when not zoomed in to the sample level.

Look for improvements in 6.2.1.1

[Jerm]

Oh yeah, and there is also the problem that the frame counter in the transport is running wild. It is tamed, however, in 6.2.1.1... to be pre-released soon. You will see vastly improved CPU usage numbers in activity monitor with 6.2.1.1 and also a smoother playback bar.

[Paul]

That sounds great! I will be looking forward to it. About the activity monitor going over 100%, I would have thought that this was because of the dual processors. You know, 100% per processor...

[Paul]

It's working well now. I loaded the same 10 tracks and happily noticed that cpu usage was even under that of Logic. I don't no how you did it, but you did. Great job!