Logic 9 on 12-core Mac Pros

Will Logic take advantage of all 12 cores on the new Mac Pros? I heard it will not...

Will Logic take advantage of all 12 cores on the new Mac Pros? I heard it will not...

 

http://www.logicprohelp.com/viewtopic.php?t=60323&highlight=

No answer in that thread...

No answer in that thread...

 

Did you actually read it?

 

From that thread...

 

A Quote from Ski here on a post I recently made:

 

IMO, don't get too excited. Apple's hardware has always been ahead of their software. And by the time Logic gets written to take advantage of those cores (if ever?) the new computers will be obsolete.

 

To be honest, I couldn't agree more. Buying a 12 core MP and running Logic, would be like buying a Ferrari and running it on 3 stroke...

 

I'd say the new 12 core has some 'killer specs' for sure. But until the software is coded to properly take advantage of all those cores..I can't see the point in getting one (for Logic, anyway).

Sure, I read the thread. I don't see anything other than conjecture, nothing definitive.

Sure, I read the thread. I don't see anything other than conjecture, nothing definitive.

 

http://macperformanceguide.com/Reviews-MacProWestmere-Cores.html

Sure, I read the thread. I don't see anything other than conjecture, nothing definitive.

 

You're right, there is a lot of conjecture there, and that's because there is nothing definitive to report. This stuff is all too new.

 

The only time you'll get "definitive" is when these things have been out for a few weeks or months and people have stress-tested Logic on it to see how it works. ONLY then will you have an idea of what to expect.

 

Apple's past performance is, more often than not, a great guide towards future performance; disbelieve that at your potential peril (i.e., by literally buying into the idea that you'll get full performance from a multi-core mac, only to be disappointed down the road). And if you want further proof/portents of this, go over to Apple's own Logic forum and read some of the threads there, particularly what Mike Connelly has to say on the subject.

 

http://discussions.apple.com/forum.jspa?forumID=1201

Thanks ski. Yes, the reason I started the thread was to see if anyone had any additional information. I'd heard from a v. good source that Logic will not support 12 cores for quite some time, and I was hoping for some clarification, elaboration.

http://macperformanceguide.com/Reviews-MacProWestmere-Cores.html

 

Good link -- thanks!

Good link -- thanks!

 

You're welcome!

That thread doesn't answer the question, it's all speculation and most of it was posted before the new machines were even shipping.

 

Now the machines are out, and while there isn't much, there have been a couple six core users who have done the stress testing and reported results.

 

Unfortunately, it looks like not only are the six core machines not an improvement, but the hexes are performing WORSE than quad machines at lower clock speeds. This is using the "evan" benchmark test - the new six cores only get 45-47 tracks while slower quads have got over 50.

 

The benchmark session (it's an interesting one to run on your own system if you're interested) and results/discussion:

http://www.gearslutz.com/board/music-computers/371545-logic-pro-multicore-benchmarktest.html

 

Logic results on the 6 core (for the record I think his conclusions are wrong, but his benchmark result is useful):

http://macperformanceguide.com/Reviews-MacProWestmere-LogicStudio.html

 

More discussion about the topic:

http://discussions.apple.com/thread.jspa?threadID=2541408&tstart=0

 

 

In short, it looks like Logic wasn't coded to run on six cores and having those two extra cores actually causes it to run worse for whatever reason. So Logic definitely needs an update to run on that machine to run properly (it already needed an update to fully use the eight core machines but this is much worse).

 

I haven't seen info from anyone running Logic on a 12 core yet. If anyone tries it (or finds info online), any info would be appreciated.

That thread doesn't answer the question, it's all speculation and most of it was posted before the new machines were even shipping.

 

Which thread? Three threads have been referenced in this thread previous to your post.

In short, it looks like Logic wasn't coded to run on six cores and having those two extra cores actually causes it to run worse for whatever reason. So Logic definitely needs an update to run on that machine to run properly (it already needed an update to fully use the eight core machines but this is much worse).

 

"It looks like" - that's the key. Still speculation, nothing definitive.

Reports are slowly coming in from users on the Apple forums, and opinions are varied.

Which thread? Three threads have been referenced in this thread previous to your post.

 

Sorry about that, the first thread linked in the thread.

 

"It looks like" - that's the key. Still speculation, nothing definitive. Reports are slowly coming in from users on the Apple forums, and opinions are varied.

 

The only speculation at this point is WHY it is working so bad on six core machines. When someone actually runs Logic on the machine and tests the performance, that isn't speculation, that is an actual result and unless you're saying that they did something wrong and their results are somehow invalid, I don't see why it wouldn't be definitive.

 

What other results have you seen from Logic users running on six core machines? I have only seen a couple and those had very similar results - what results have you seen that have been different?

 

Any links would be appreciated, thanks.

Which thread? Three threads have been referenced in this thread previous to your post.

 

Sorry about that, the first thread linked in the thread.

 

Thanks for the clarification Mike.

What other results have you seen from Logic users running on six core machines? I have only seen a couple and those had very similar results - what results have you seen that have been different?

 

Any links would be appreciated, thanks.

 

Check out this link:

http://forums.macrumors.com/showthread.php?t=1000550

 

It says the one logic 9 test they did at macperformanceguide wasn't even a correct test, the guy wasn't even a real Logic user. He just put 46 instances of the same thing before the CPU maxed out.

Those are the type of test that get referred to as "real tests".

So one test by an engineer, not even a Logic user, and suddenly a 6-core is crap??

Come on...

Looks like you're just not familiar with that test and didn't bother to read the details about it.

 

He didn't just start putting on plugins, that's a standard benchmarking session that many users have used on many machines. You can find it linked here along with many results and much discussion.

 

http://www.gearslutz.com/board/music-computers/371545-logic-pro-multicore-benchmarktest.html

 

Since he used the exact same session that many Logic users have used for this purpose, I don't think it's relevant how much Logic experience he has since it looks like he ran the benchmark properly. People are entitled to their opinions, but I don't agree with the one post in the thread you linked who criticized the test (I get the impression he didn't understand it either). Of course it's different than real world use, but the point is to be able to compare the relative power of different machines running Logic. You can even download the benchmark and try it out yourself if you want to better understand the situation.

 

If you read the thread you linked, another user ran the same test on the same machine and got an identical result, 45 tracks worked and an error with 46.

 

You are correct that there isn't much data yet. But you have given no reason to doubt that what we do know is valid.

 

Whether the 6 core is "crap" - if this machine only gets 45 tracks with this particular test but a quad running at 2.26 gets 50, I wouldn't consider that particularly good performance, would you?

 

In general, wouldn't you expect a machine with two more cores AND a much higher clock speed to perform better? If that faster machine performs worse, you don't think that's an issue that should be fixed?

 

Earlier, you said that "opinions are varied". Are there any reports posted from users running six cores with Logic who are getting measurably better performance than on a quad? Or are those "varied opinions" just people speculating?

Just saw a report from a third 6 core user. He was able to get a few more tracks (54, which is better but users have been able to get that many on a quad 2.66), but again six cores were displayed but only three had any activity on them.

 

http://www.gearslutz.com/board/5725935-post632.html

 

I'd also say something is fishy with Logic when it shows eight CPU meters on a quad but six on a six core.

Might be OT, but FWIW, I just downloaded the "Evan" benchmark test, and I can't help but question what makes this test even remotely valid as a benchmark?

It provides a means to compare performance between different machines, and can be a useful troubleshooting tool. Obviously it's not going to be exactly the same as real world use but it's difficult if not impossible to do a test with a bunch of different plugins and get a result that is comparable between different machines. In general, the relative performance between different machines as shown by this benchmark has been close to the relative performance when doing real world sessions - if a machine can do twice as many plugins as another with this session, it can generally do about twice as many with a mix of plugins.

 

No benchmark is ever going to make exact predictions about real world use since real world use is going to be different for every user, but it's a good ballpark figure.

 

And using the six cores as a specific example, if a six core machine at 3.33 is doing the same or worse as a quad at 2.66 with this benchmark, it shows something is wrong and lacking performance across the board is likely with Logic on the six core machines. Is it possible that there's something about the six core machines that only is a problem with this particular plugin or session? Sure, anything is possible in theory but that's extremely unlikely.

 

There are other ways to do benchmarks on Logic, I have done a couple of my own but they haven't been used as widely as the Evan one. It's probably possible to do better ones, if you have an idea for a better way to do it, create a session or suggest the idea so someone can try it out.

 

So what result did you get from the benchmark?

Hi Mike,

 

Thanks for your reply.

 

I think the test is unrealistic in the sense that it's unlikely that anyone would put exactly the same reverb on multiple channels. Same comment regarding the ring shifter. Both are obvious wasters of CPU power in the way they're implemented in that session. If you were to install a singular instance of both reverb and ring shifter on auxes and bus the signal from each track to them you'd get exactly the same sonic result. I'd think this approach would vastly increase the track count. Next, playing exactly the same sample at the same time across multiple instances of EXS isn't something that would likely ever happen in the real world.

 

As far as track count, I tried it out on my PPC Quad, and got to play 14 tracks without making changes. When I did what I said above regarding the ringshifter and the reverb (uninstalling them from all instrument tracks as well) my track count doubled, and on occasion I was able to get upwards of 31 tracks.

 

Since no two pieces of music would ever be exactly the same, I don't see the point in any one test project being a valid benchmark. My feeling is that multiple benchmark tests are needed, each taking a different approach, some taking an unrealistic approach musically and engineering-wise (such as Evan's test) and others which are more realistic or at least varied. For example, how about a test using instances of both EXS and Sculpture, using standard engineering practice of bussing to a reverb instead of having individual plugs on each channel?

 

Another good test might be for assessing the performance of audio playback. Take a single stereo audio track and have the file duplicated and saved multiple times, each under a different name (for as many instances as tracks that you want to test playback with). This way during playback you'd be assured that Logic is handling playback of discreet files. Or, of course, it could be playback of multiple different files.

 

No benchmark is going to be perfect. Goes without saying. But with Evan's test we're only testing one aspect of playback (EXS) in a context that's not musical and shows poor engineering practice. More meaningful results would come from running multiple tests each designed to stress a different (or multiple) set of criteria.

 

As to trying to establish a more rigorous testing procedure, I would imagine that some people might balk at having to stray beyond the constraints of a single test, looking for more immediate gratification. But IMO, a multi-thousand dollar investment in a computer requires more investigation than that. And it's to this point that I'll reiterate that historically, multi-processor Macs and versions of Logic that can take advantage of them all have been two ships passing in the night. That's not to say that the more recent multi-core Macs aren't an exception, but with the advent of these new computers, in my mind I'm thinking "here we go again".

Well put, ski.

My point is, like you said, every user is different, and then people refer to 1 test online, and suddenly the 6-core has a bad rep.

For example pull up a session with around 80 or more tracks of audio and midi with and without 3rd party plugins (another huge factor), then compare. That's more meaningful and more real to me.

There are so many variables, one test or benchmark is not gonna mean anything.

When the 8-core mac pros came out almost 2 years ago, we had a session of 80+ tracks that a G5 couldn't handle after 4 bars into it.

The mac pro just yawned thru it. And we could play it with a 32 buffer and no peaks whatsoever.

Thanks triplets.

 

FWIW, I'm currently running a project with 231 tracks: audio tracks, tons of sampler instruments including EXS-24, Goliath, Kontakt 3.5, three space designers, tons of automation, a bit of EQ here and there, and a few [gulp] highly edited Apple loops. All running on the same moldy-oldy computer I wrote about above -- G5 Quad, PPC -- all without a hitch. Now, that's not to say that all tracks are playing at the same time, but that's precisely my point.

Looks like you're just not familiar with that test and didn't bother to read the details about it.

 

It's easy to critisize, much harder to contribrite.

I think the test is unrealistic in the sense that it's unlikely that anyone would put exactly the same reverb on multiple channels.

 

Of course it's not a real world test. But it does compare how logic handles CPU use on various machine and allows comparisons. Obviously if you use different plugins you may get a slightly different result, but this test gives a decent idea of the CPU performance differences between various machines. The session is intentionally designed to max out a machine for testing purposes.

 

Since no two pieces of music would ever be exactly the same, I don't see the point in any one test project being a valid benchmark.

 

No benchmark is going to be 100% indicative of how actual sessions are going to run, but this one does give a good ballpark idea. Just because it's not going to be perfect doesn't mean it's not useful to some degree. Certainly more useful than not comparing at all.

 

I completely agree that more benchmarks would be even more useful, I'll dig out the ones I've used in the past and if you have an idea for a more "realistic" one I encourage you to create a session. You can do one bussing to a reverb but if you don't max the machine out I'm not sure how you are going to measure the comparison. Audio playback is a good idea for a test, but cpu isn't going to make much difference, disk speed is the biggest factor for that.

 

And it's to this point that I'll reiterate that historically, multi-processor Macs and versions of Logic that can take advantage of them all have been two ships passing in the night. That's not to say that the more recent multi-core Macs aren't an exception, but with the advent of these new computers, in my mind I'm thinking "here we go again".

 

While I agree with that, I see a difference here. In the past, more cores have provided little or no improvement. In this case, more cores (along with much higher clock speed) actually make performance worse, which I haven't seen before, and isn't the case with any other apps on the 6.

 

ou said, every user is different, and then people refer to 1 test online, and suddenly the 6-core has a bad rep.

 

You're right, it's just one test. It's possible that coincidentally this just happens to be something that runs worse on 6 cores than on 4. But should ANY Logic session run worse on 6 than 4? And for the record, I don't think anything is wrong with the six core (since it runs great on every other app that has been tested), the evidence we have so far points to Logic not supporting it (which would just be a Logic fix).

 

It's easy to critisize, much harder to contribrite.

 

I completely agree, but I think your post is directed at the wrong user. I have contributed more data about this issue than anyone else here, with links to real use of the software on the machine. Those disagreeing haven't posted a single example that contradicts these facts, just criticism.

 

Here's another bit of data to add to the discussion.

 

On this particular session, multiple users have got 45 tracks running in Logic 9. On the gearslutz thread, someone tried the same session on Logic 8 on this machine and got 75 tracks.

 

You still think that doesn't point to an issue with L9?

The session is intentionally designed to max out a machine for testing purposes.

 

Using an unrealistic example of music production? It's called a fait accomplis.

I just downloaded the "Evan" benchmark test, and I can't help but question what makes this test even remotely valid as a benchmark?

Putting aside for now the issue of whether a project like this is real-world or not.

 

If the point was to stress test the cpu and look strictly at the multi-core performance, then no, this project isn't a good benchtest example. Each EXS instance has virtual memory enabled, but I can't determine whether the notes/samples are long enough to properly exercise that feature. It kind of throws an undetermined variable into the whole thing. If the original intent of the project file was to test virtual memory and RAM performance under multi-core conditions, then the project is nowhere near being comprehensive enough. Another variable is that multiple EXS instances share resources to some degree, so your not actually stacking up finite cpu requirements.

 

A project that consists of tracks hosting sustained notes of real-time synthesis being processed through real-time (not convolution) effects, would be a much better measuring stick for cpu handling and allocation.

 

What I'm saying is that the evanlogicmulticorebenchmark project doesn't jive with good scientific methodology if pure real-time cpu performance is what you're trying to measure.

I'm just putting these results out there so that people are aware of them.

 

If you don't find any value in them, then by all means ignore them. People need to decide for themselves how useful they are.

 

While I acknowledge that they are a limited test case, they are the only results we have so far about the new machines, so they are useful to me. And for me it raises a red flag when any app runs worse on a machine that has more cores and a higher clock speed. It also raises a red flag when version 9 of an app gets 60% of the performance of version 8 of that same app on the same hardware (regardless of the specifics of the session).

 

Along the same lines, the quad machine shows 8 cpu meters in Logic (and uses all of them) while the 6 core shows six meters and only uses four. That's completely unrelated to the session being run and points to a problem with the app.

 

It's absolutely possible that this particular session is an exception and that most other things in Logic run better on the new machines. And that's easy enough to demonstrate, just create a session and run the same one on both a quad and six core machine. So far I have yet to see any feedback from any Logic users on a six core machine saying that they are getting better performance on the same sessions than on a quad.

 

So if you think the six core machines are performing well with Logic, it's easy enough to demonstrate that, just find someone with one and have them run a session then compare the same session on a quad. Heck, just find someone with one of these machines and have them see if they can get ANY combination of plugins to use more than four cores.

Fader8, you make some interesting points. There may a bit of variation in the way that EXS works on different machines, but compared to the other plugins in the session, the amount of CPU power used by the EXS is a tiny portion (it is a very small instrument, and the midi part isn't many notes). The intent of the session was not to test virtual memory in EXS, the EXS was just used as a sound source so the other plugins on the channel would have audio to process. Personally, when I have done similar tests I have used Klopfgeist as the sound generator in order to minimize sample playback as part of the equation. I'll have to test it out, but I'll bet that if you switched all the exs to klopfgeist the results would change very little. Might even be similar with audio on each track (but that would bring in disk streaming, although unless the system has a really slow disk I doubt it would still make much difference on such a small number of tracks in most cases) - the cpu required for any of those is dwarfed by that required by the plugins on each track.