Logic Performance Thread
Since EmApple Logic is a great way to test the performance of a mac in real-time media apps, we can use it to see where everything stands. Here's my results so far for number of Platinum Verbs on a stereo track (spilling over onto a stereo bus).
Settings: Logic 4.7.3 in OS 9.2
Mac AV in universal trackmode
no delays
larger process buffer
larger disk buffer
ibook 700: 8-9 PVs
Ti 550: 8-9 PVs
G4 tower: 9 PVs
Ti 500: 12-13 PVs
Settings: Logic 4.7.3 in OS 9.2
Mac AV in universal trackmode
no delays
larger process buffer
larger disk buffer
ibook 700: 8-9 PVs
Ti 550: 8-9 PVs
G4 tower: 9 PVs
Ti 500: 12-13 PVs
Comments
i'll get to it as soon as i can - i've got visitors and work piled up.
but i will match the specs and have a go at it. especially now that i've learned a trick which should allow me to max the PVs on both processors of a dual. it's not a trick which reflects real world performance, as logic's multiprocessor support is pretty weak at the moment, but it should show power that is still available in duals and yet to be harnessed by logic.
i suspect that with the improved performance settings, all of my numbers will go up a couple of instances, but we'll see.
i ran a quick test tonight and was able to achieve the following Logic peformance on X using 5.3 - same settings you posted - on my dual 1.25.
without trickery i got 27 PlatinumVerbs.
WITH trickery, i got 45 PlatinumVerbs and 1 soft synth.
for a description of this trickery, please visit this thread:
<a href="http://www.osxaudio.com/forums/viewtopic.php?p=473#473"; target="_blank">dual support in logic</a>
IF logic would ever get true dual processor support, then the 45 PlatinumVerbs would be a daily reality for dual 1.25 users.
i'll try to get more performance soon, but the weekend keeps getting more away from me.
Right now I get the result above for the ti500. But OSX and logic 5 performance losses could account for the low ability of the 667.
OS X does load balancing just about as well as an OS can do load balancing, but it's not possible to split a single thread of execution between processors on any platform. Apple's claims of "automatic" multiprocessor support simply mean that the OS transparently schedules any running processes and threads across all of available processors as evenly as possible; no explicit support for multiple processors is necessary, as it was in OS 9.
However, you do have to thread. Threading is something of a black art, mostly because few of the languages in common use (including all of the C-derived languages) have any native support for parallelism. It's something that has to be designed in carefully. Not only can you not just "tack it on," but unless the code has been very carefully factored (roughly, organized into pieces that have few - ideally, no - dependencies on other pieces) any attempt to reengineer threads into a big, single-threaded app will be a large-scale undertaking not unlike a full rewrite. On the other hand, if the code is factored (as the game Giants: Citizen Kabuto was) then threading it won't be that hard (Omni did it with their OS X port of the game).
Fortunately, half of the reasons not to go to the trouble of threading Logic vanished when they canceled Windows support - threading on 2000 and XP carries a stiff penalty on single-processor systems (for no reason that I can imagine), and the vast majority of Windows systems are single processor (I'm not sure whether those OS's inherited NT's asymmetrical MP support, where one processor ran the OS and the other ran all applications, but if so that would be another reason...). The other half will vanish when Logic drops OS 9, whose support for threading is, uh, suboptimal.
It'd be great if someone kept a master list of this on a site, and people could add to it.
Would also love to see a track count list using 7200rpm drives as a benchmark.
<a href="http://www.digitalnaturalsound.com/logic_dsp/perform.shtml"; target="_blank">DNS performance charts</a>
thanks for the explanation. i've been asking all over in hopes of getting just that. now i can only hope the Logic code won't need a full re-write to thread properly. also gotta hope that apple sticks with duals when they start with the new 970 - because if they are going back to single processor units i fear that emagic (being owned by apple) won't have any real incentive to really optimize for duals. ah well, it's out of my control - thanks again for the explanation.
giant
sorry for the delayed response - i was unable to login at appleinsider (or even email sysadmin here - kept getting sent back to me) for some reason until today.
anyway, with similar settings using 5.3 on both OS X and OS 9, i was able to get the following stereo PV results:
Ti 667 (2002 DVI model) - 14 PVs
Dual 867 - 18 PVs
Dual 1.25 - 27 PVs
the OS DIDN'T matter for PV performance. but it DOES seem to matter for other plug-in performance - and might be a reason for you to go to Logic X (and this improvement also shows how a simple PlatinumVerb test, like the one you linked to, can be misleading regarding realworld Logic performance).
it turns out that SOMETHING has been optimized within Logic for OS X - perhaps someone here has an idea what?
let me explain:
in response to my PV tests, another user created the following session in Logic 5.4 on OS X:
[quote] As an experiment, I took a drum session I had, 6 mono audio tracks 1 stereo, 24/44.1, and on each of the mono tracks, I loaded up, all mono/mono:
3 Multipressors
1 Enveloper
4 Compressors
1 Fat EQ
1 Overdrive
1 Auto filter
1 ModDelay
1 Gold Verb
1 Limiter
1 Pitchift II
I also added 5 stereo Platinumverbs to the mix.
(95 plugins, new Dual 1Ghz)
Only then did I get the CPU performance meter near the top, but the whole thing played back very happily and the system was still responsive. I could even open up AppleWorks with Logic playing in the background and type quickly without a single glitch in the audio.
This was all with the buffers for the MobileIO set to 128...
I then opened the session in 5.3 under OS9. THE SONG WOULDNT EVEN PLAY FOR HALF A SECOND <hr></blockquote>
so, seeing these results, the user figured that 5.4 had dual support.
so i matched his test session (i actually was able to get 7 audio channels full of plugs because i'm using a dual 1.25 where he was using a dual 1 Ghz) and opened TOP and found that Logic 5.4 still WASN'T using all available CPU power - in fact, just as in the PV tests i ran, Logic seemed to be assigning ALL of the DSP processing to one processor (TOP still shows Logic maxing out on DSP load at about 120% - out of 200%).
and to confirm this, i then used the Logic 'trick' and opened an ES2 on an AI channel, loaded up the 15 plugs as listed, and then added another 10 PVs to the output channel (remember, so long as the AI channel is selected, ALL of its DSP and ALL of the DSP of its assigned output channel are processed by the 'unused' processor). and when i did this, TOP showed Logic using around 195% of CPU. so clearly, Logic on X still ISN'T using both processors for DSP. fine. just what i expected to see.
but then i opened the same song on 5.3 in OS 9 and, sure enough, i couldn't get the song to play. so i started to remove plug-ins from the session - and kept removing them, and kept removing them. until i could finally get playback. at 3 TRACKS! less than HALF the performance under X.
i could only get THREE audio tracks to playback fully loaded with the plugs listed above. of course, playback was unaffected by the AI channel - i could only get 3 loaded audio tracks to playback regardless of whether the AI channel was empty or full, in the session or not.
and then i repeated this testing on Logic 5.3 under X and got the same improved results as 5.4 on X - so obviously this improvement occurred for Logic when it went to X.
but what is it due to? DSP performance for Logic on X has been optimized SOMEHOW - not for duals, of course, but they've done something. my question is - is this improved Logic coding for plug ins in X or is this due to improved audio handling with Core Audio (meaning, it is easier for Core Audio to streamline delivery between plugs)? or is it something else?
my gut instinct says it is the former, because there is NO improvement in PlatinumVerb performance between 9 and X.
but then i think - if it really was improved plug-in code, why wouldn't these improvements apply to 9 too? ok, so maybe they want to keep the improved code for X only to make it LOOK like Logic is using both processors on X (they did remove the second CPU graphic from the Logic System Performance window - giving the impression that Logic sees both CPUs as one processor. this, of course, isn't true - but it certainly looks like it is when plug-in performance doubles as well).
anyone have any ideas why this improved plug in performance is seen in Logic on X even though Logic X isn't using any more processing power? and why the improvement isn't seen in ALL plug-ins?
don't get me wrong, i'm thrilled to see ANY improvement for X - hopefully suggests more improvements for X might arrive soon. but is this REAL improvement for X or just a Logic 5.3 code improvement which hasn't been included in Logic 5.3 for 9? knowing nothing of coding, i have no idea how easy or hard it would be NOT to include some improvements.
any explanations would be most appreciated.
cheers
peaceful billy (aka stupid8track)
<strong>what is it due to? DSP performance for Logic on X has been optimized SOMEHOW - not for duals, of course, but they've done something. my question is - is this improved Logic coding for plug ins in X or is this due to improved audio handling with Core Audio (meaning, it is easier for Core Audio to streamline delivery between plugs)? or is it something else?
my gut instinct says it is the former, because there is NO improvement in PlatinumVerb performance between 9 and X.
but then i think - if it really was improved plug-in code, why wouldn't these improvements apply to 9 too? ok, so maybe they want to keep the improved code for X only to make it LOOK like Logic is using both processors on X (they did remove the second CPU graphic from the Logic System Performance window - giving the impression that Logic sees both CPUs as one processor. this, of course, isn't true - but it certainly looks like it is when plug-in performance doubles as well).
peaceful billy (aka stupid8track)</strong><hr></blockquote>
My guess is it has something (possibly alot) to do with Audio Units, which is OSX's native format for plug-ins. It is of course nowhere to be seen in OS9.
Edit: One day I might learn some grammar.
[ 10-21-2002: Message edited by: LowB-ing ]</p>
not a bad idea - perhaps this is it?
still, i'm just not sure why some plugs would show DSP performance improvement and others wouldn't.
and i'm not sure that the Logic plugs are actually audio units - they certainly aren't available system wide as audio units are. i believe that the logic plugs are considered part of Logic - don't know if this eliminates them from being AU plugs or not.
is it possible to actually restrict AU plugs to one program?
.......
Markus Fritze, one of Emagics head developers, has posted an important comment in the forums on the location of AudioUnit plug-ins. Unlike all previous plug-in formats, AudioUnits are not application specific. They are system wide and thus they are placed into a centralised location.
As with many aspects of Mac OS X, there is a choice to install applications and extensions in either your own personal account or in a common area. This way these objects are available for all users on your machine or just your own account.
The two different locations for the AudioUnit plug-ins are:
Your account:
~/Library/Audio/Plug-Ins/Components/
[the tilde represents your account]
or
Available to all:
/Library/Audio/Plug-Ins/Components/
.......
Perhaps the plugs would be available to the rest of the system if it was possible to simply move them to one of the locations mentioned above (which it probably isn't). This is just complete, utter speculation, however.
i don't see any indication of any Logic plug-in in either location. that's why i'm not sure they are actually AU plugs. or if they are, they are definitely restricted to logic only. i can't even find them anywhere to try and move them to another library location.
yeah, i have no idea if an AU plug HAS to be distributed by the OS, or if it could be restricted by the software to one processor or the other - like the logic plugs are now.
any ideas?
cool. thanks.
so any idea what is responsible for the improved performance of some logic plugs on OS X? cause the improvement can be dramatic, but it ain't from improved dual support.
i did a little testing with AU plugs (logic 5.4 ships with some AU plugs - or at least they appear in the logic plug-in menu under Audio Units).
there is no improvement in CPU distribution using the AU plugs in Logic. effectively still max out the system at 60% in top, and can still 'trick' the system to max out at 195% using the AI channel trick.
so i don't think any improved logic plug performance on X can be attributed to new AU plugs. logic seems to assign their processing as it does any of its other plugs.
i can't actually confirm these are true AU plugs because i can't find them in either library location that LowB referenced above.
so while logic calls them AU plugs, i'm not sure what they are.
yes, PCs dominate Macs on the Logic PlatinumVerb test. but this is actually not due to inferior hardware so much as it is due to almost non-existent dual processor optimization in Logic.
on my dual 1.25, with some trickery to make Logic use BOTH processors, i am able to get 45 PlatinumVerbs to play back at once. doesn't put the 1.25 at the top of the list, but certainly is better than the 27 i get without the trick.
of course, this trickery IS just trickery until Emagic can optimize for duals. and yet, Emagic have clearly done SOMETHING with some of their included Logic plug-ins in OS X to really improve performance. i'm trying to figure out what that could be.
i still think Logic is more than powerful enough on one processor to do any hip hop/rnb beat creation you might want - though be sure to buy a couple of their soft synths and their EXS24 sampler too. if that is more than you are looking to spend right now, then a program called "Reason" would be enough to get you started for around $250. the biggest drawback of Reason is you can't record vocals with it.