Logic Pro Multicore Benchmarktest !

[Firechild]

Originally Posted by dreamteampro

Mac Pro,Early 2008,2.8 Octo w/10GB of RAM doing 70 tracks here with apogee duet @ 256 latency...

That is pretty good,you must have found a golden batch !
I don´t say it´s too good, but some users report 54 tracks with the same machine and your machine is performing like a 8 core 3.2, congrats!

[van Overhalen]

Originally Posted by Firechild

...your machine is performing like a 8 core 3.2, congrats!

There are 2 users who report 57 tracks on 8x3.2
One reports 71 tracks on 8x3.0
And there is the one reporting 70 tracks on 8x2.8

???

I don't see any value in this test.

[blinkfrog]

Originally Posted by van Overhalen

There are 2 users who report 57 tracks on 8x3.2
One reports 71 tracks on 8x3.0 And there is the one reporting 70 tracks on 8x2.8
??? I don't see any value in this test.

IMHO audio interface is important. E-Mu 0202 differ from RME HDSPe AIO in terms of audio latency, cpu-loading and stability. USB is crap, firewire is reasonably good, but only pci-express can provide truly fast, stable and low latency streaming. Just my 2 kopecks.

[space2012]

Originally Posted by Firechild

Index of /logicprobenchmark

Here is a very simple benchmark test. Download the zipfile located on the link above.
Open with Logic 8.02. Then un-mute tracks one by one and check how many tracks your computer is able to run.
Audiobuffer:256 samples will give the best results.

Feel free to email your results to me, evan@evan.se

The chart below is missing reports from the older dual 2x2.66ghz MacPro running 10.5.3 or higher when huge improvements in multicore handling was introduced.

We are also LONGING for reports from the new Nehalem MacPro´s !

Mac does not have i7? or 32x core?
WTF!

AMD Opteron 8xxx vs. Intel Xeon E54xx plugin tests.

[YUGA]

Originally Posted by van Overhalen

There are 2 users who report 57 tracks on 8x3.2

Maybe SpeedStep is enabled on their Macs?

[master2000]

hello,

i must say the new macpro is a real beast!!!
i got a macpro 2x2,26 octo/12 gb ram/ati ,i installed logic and run the benchmark test with my ergo krk soundcart.

and i can use 54 tracks ,no problems....and cpu was only the first 5 cores in red.
i think when logic 8.0.3 and mac osx 10.5.7 came out my mapro will do 100 tracks.....
there are so much cpu power......

greetz

[peacenik]

Just did the test on my new Nehalem MacPro 2.66 QuadCore...

[OSX .5.6 with 6 gigs ram.
I HAD to use my mBox2 (used for post sometimes and transfers, I prefer Logic by far)
because I now HAVE to sell my RME9652 PCI-X and get
some sort of PCI-E or Firewire interface (new expense, oh joy!).]

I played reliably 51 tracks with a 256 buffer at 44.1k.
(See attached image.)

But get this.. I can play 54 tracks with buffer at 32! (?!)

Both these #'s is where it played in a loop for a while... 1 more track un-muted and I'd get overload.

(BTW - Using the built in Digital out played fewer tracks..)

Now for a real oddity.. when I first opened the file (before I started playing with buffers and interfaces and such)
I played 62 tracks !
Go figure... maybe the tracks wern't all really playing...
though I playeed it in a loop and restarted numerous times so I think they were...
I can't seem to re-live this moment after fiddling around though.

Voila.. looking forward to OS X updates, Snow Leopard and Logic updates though I guess I'm not desperate.
Cheers
R

[Firechild]

Tests with Cubase and the new Nehalem CPU:s ( in the PC world named i7 ) in combination with the fast DDR3 memory show that working with very low latencies will not affect CPU load as much as earlier technologies ( Harpertown MacPros for example ) and we can already see that the new Nehalem MacPro users will report similar scores no matter what latency settings, 32 samples or 512!
And as soon Logic 8.0.3 will support 16 virtual threads we will see the new series totally outperform the older early 2008 series. I guess the dual 2.26 will score 85 tracks and the top end over 100 tracks. But of course the early 2008 models is still really good value for money if you can find them in shops. Owning a 8 core 3.2Ghz myself I can say I am very happy and can´t see a reason to upgrade in the 3 forthcoming years.

D.A.W Bench : D.A.W Performance Benchmarking
According to this test the Nehalem CPU`s are TWICE as fast as previous Harpertowns at the same CPU:speeds. A single quad i7 has equal power as two quad XEON Harpertowns...

[peacenik]

Gotcha... and thanks for the 'D.A.W. Performance' link...
it'll be interesting to see how much further things go with various OS & ap updates.

So there's lot's of variables and performance needs...
but VERY low latency (sure, with 'high' track counts too) is more important 'to me' then 'insane' track counts...
since I play drum pads on very complex multi instrument setups. I'm very happy.

[HockeyMike]

Originally Posted by peacenik

Now for a real oddity.. when I first opened the file (before I started playing with buffers and interfaces and such)
I played 62 tracks !

That's something in Logic that I'm sure someone who knows the program well under the hood could explain. I found that on my previous machine (a dual 1.8 G5) if I had a mix that was really pegging the cpus and it wouldn't play all the way through on a real-time bounce, if I quit Logic and reopened the song I could usually get through the whole thing without a hiccup. If I started messing around with the mix at all, it would start overloading again. Seems like the longer a song is open in Logic, the more resources that get taken up by it.

Not a problem these days with my 2008 MP 2.8 8-core (10GB RAM)

I'm able to play 57 tracks @ 1024 buffer, 52 tracks @256 with a Fireface 800.

As far as what Nehalems will do once Logic is updated, that really remains to be seen...lots of wishful speculation going on around here, some of it rather, er, optimistic.

I'm interested in seeing what a 4-core Nehalem maxed at 8GB RAM will do compared to an 8-core maxed at 32GB RAM once the 4GB per app limit is gone in Snow Leopard...should be interesting.

[blinkfrog]

Originally Posted by peacenik

but VERY low latency (sure, with 'high' track counts too) is more important 'to me' then 'insane' track counts...
since I play drum pads on very complex multi instrument setups. I'm very happy.

In this case you have to seriously consider purchasing pci-express interface. With equal buffer settings pci-express can to provide much lesser round trip latency than firewire (best firewire interfaces have rtl ~5.5 msec @ 44.1 kHz). Also take in account that many pci-express interfaces in fact are bridge-based pci-interfaces. AFAIK there are currently only several true pcie interfaces: from Lynx (was misprint "Lexicon") and from RME.

2All: Who knows, in that order Mac OS shows Nehalem's cores: 8 real and 8 virtual, or real-virtual-real-virtual etc? If first, then we have seen already almost full power of 8x2.26, because how I understand it use first several cores which are real cores. If second, then true power will doubled after software updates because Logic currently using only half of real cores.

[peacenik]

Yea... as much as I want to invest this time in a good & flexible Firewire futureproof interface (MetricHalo or RME)... I'll probably go with a 5+ ms faster rme pci-e. I 'plan' to either get the rme AIO (and sell all my older rme stuff) or the rme Raydat which has only adat ports (and connect it to my current rme ADIO8-DS).

As far as my new MacPro.. I'm super impressed with the memory & latency improvements! And I figure I'll get it 12 gigs ram after the 4 gig simms price drops some more... (4 gig simms just dropped 'again' at OWC to $699... once it gets under 500 I plan to buy in).

[Firechild]

Chart updated with 2009 MacPro´s. The single 2.66 Nehalem is performing great already ( for a quad ) and we can expect the dual Nehalems scoring about 60-80 percent better in the next Logic update due to the lack of support for 16 cores in the current Logic version 8.0.2. ( Virtual or real ).

[blinkfrog]

While Evan's benchmark is very reliable test showing theoretical max all-cores performance, in real practice such situation is not such frequent case.

Real project using virtual instruments can use 1 track with heavy synthesizer and effects while 2 track can use just simple sampler. And DAW can't efficiently combine such various tracks and evenly distribute it among all cores. And Snow Leo + GC can't to resolve this problem absolutly efficiently. It is the nature of DAW: every single track can't be splitted to several cores because of plugins inputs in stack are depend on output of previous plugin.

And in real work this case is very frequent: while several cores are maximized other cores can be <30%. Of course it is possible to distribute loading using auxes (like Apple reccomends), but it is almost impossible to achieve even load in real project.

Evan, I propose an addition to this benchmark. I think it is good idea to benchmark single core performance (like Cinebench do) and to add result to your graph as additional lines under multicore results. May be stack of impulse reverbs or other heavy plugins can do the job. Although heavy synth with high polyphony would be better, but it can't be stacked...

Such test can show different result to existing one.

P.S. Sorry for my ugly english, I'm not native speaker.

[Allen Rowand]

Originally Posted by peacenik

Yea... as much as I want to invest this time in a good & flexible Firewire futureproof interface (MetricHalo or RME)... I'll probably go with a 5+ ms faster rme pci-e. I 'plan' to either get the rme AIO (and sell all my older rme stuff) or the rme Raydat which has only adat ports (and connect it to my current rme ADIO8-DS).

Our FW interfaces are 4.5ms roundtrip at 44.1kHz, 2.6ms at 96kHz. If you're playing external hardware (analog or digital) you can monitor in our mixer, which has a latency of 1.5ms at 44.1kHz. Even if you're using VIs, your latency would be beter than 4.5ms since you're only going through the output section.

Best,
Allen

[peacenik]

I appreciate your notes Alan and will definitely consider the MetricHalo...

[Firechild]

Originally Posted by blinkfrog

While Evan's benchmark is very reliable test showing theoretical max all-cores performance, in real practice such situation is not such frequent case.

Real project using virtual instruments can use 1 track with heavy synthesizer and effects while 2 track can use just simple sampler. And DAW can't efficiently combine such various tracks and evenly distribute it among all cores. And Snow Leo + GC can't to resolve this problem absolutly efficiently. It is the nature of DAW: every single track can't be splitted to several cores because of plugins inputs in stack are depend on output of previous plugin.

And in real work this case is very frequent: while several cores are maximized other cores can be <30%. Of course it is possible to distribute loading using auxes (like Apple reccomends), but it is almost impossible to achieve even load in real project.

Evan, I propose an addition to this benchmark. I think it is good idea to benchmark single core performance (like Cinebench do) and to add result to your graph as additional lines under multicore results. May be stack of impulse reverbs or other heavy plugins can do the job. Although heavy synth with high polyphony would be better, but it can't be stacked...

Such test can show different result to existing one.

P.S. Sorry for my ugly english, I'm not native speaker.

Thanks for your input.
It is true that it may not be a reallife scenario in this test but I was interested to see how Logic use multiple cores and when I made the test it was very poor, but after the 10.5.3 update things changed and now it is pretty good.
I don´t think it is interesting for Logic users to see how much a single core can do, because it is very rare to have so many plugins on a single track that you max out a core with one track with a modern CPU, ( Logic can´t split plugins on a single track to multiple cores, then you need to use a aux to have even more plugins ). An additional test would of course be Audio trackcounts (HD performance ) and RAM speeds but it is not exactly depending on Logic software but more generally task that can be tested with other benchmarktests. So this Multicore benchmark is still very interesting due to its Logic specifik nature.

[blinkfrog]

Originally Posted by Firechild

Thanks for your input.
It is true that it may not be a reallife scenario in this test but I was interested to see how Logic use multiple cores and when I made the test it was very poor, but after the 10.5.3 update things changed and now it is pretty good.
I don´t think it is interesting for Logic users to see how much a single core can do, because it is very rare to have so many plugins on a single track that you max out a core with one track with a modern CPU, ( Logic can´t split plugins on a single track to multiple cores, then you need to use a aux to have even more plugins ). An additional test would of course be Audio trackcounts (HD performance ) and RAM speeds but it is not exactly depending on Logic software but more generally task that can be tested with other benchmarktests. So this Multicore benchmark is still very interesting due to its Logic specifik nature.

You're right.

Althought single core performance results along with multicore performance would be handy for users who can't decide which mac pro to buy: 8x2.8, 4x2.66(2.93) or 8x2.26. How much less plugins single nehalem 2.26 core capable to play (in comparison with harpertown 2.8)?

As for me. My projects usually consist of 3...5 extremely cpu-heavy tracks (virtual instruments and effects), 4...6 with medium loading and 5...15 with light loading. As you can see, it is far from 50+ evenly loaded tracks. And which mac pro do I need to purchace? I prefer low latencies and new tech, so likely I need 2009 mac pro. 8x2.26 is more multicore-powerful then quads, but it is possible situation when my 3...5 HT-splitted low-GHz cores can't to support cpu-heavy tracks while rest of cores will have low loading.

BTW, splitting to auxes will bring additional latency.
In fact, it is a good idea to wait Snow Leo + Grand Central + Logic update, but I need new mac pro in month or two. Currently I am inclining to 8x2.26.

Well, this my post is mostly rhetorical.

[rcb4t2]

41 Tracks... I was expecting more, to be honest....

First gen dual-quad core mac pro @ 2.8 w/ 4gb ram, apogee ensemble

[Firechild]

Originally Posted by rcb4t2

41 Tracks... I was expecting more, to be honest....

First gen dual-quad core mac pro @ 2.8 w/ 4gb ram, apogee ensemble

Something is not right. You should at least score 54 tracks.
Latest OSX 10.5.6 ?
Latest Apogee drivers?

[rcb4t2]

Yes, latest drivers - this is actually the first time I've been able to really overload my machine (I guess I haven't been trying hard enough)... I have tweaked some logic settings towards low-latency virtual instruments, so maybe that f***ed me up. Not sure - also have chase and smooth cycle algorithm engaged...... any thoughts? Was there an embedded sample rate in that file? Must be, right? I usually run at 96k and the ensemble is configured as such (cut out most of the digital i/o). Made sure to close out everything else while I did the test... hmm, now I'm all worried haha

edit: I was being pretty cavalier with the way I was unmuting... just down-arrow, m, down-arrow, m pretty quickly... I'll run another right now to see if it was a fluke.

[Firechild]

Yes, it is a sample in the test taken from the Logic Library and an impulse reverb also from the library. the test should run @ 44.1 kHz.
Try the built in audio instead of Apogee. Some users report great results with Duet and some pretty mediocre but there have been issues with Duet in this test but I thought it was due to old drivers.

[rcb4t2]

Firechild - thanks for helping me chase this down, I solved the issue!

I had process buffer range set to small, because I figured with 8 cores I could afford to lean on the CPU a little. This let me have 41 tracks. When I set this to large, I got about 54. I was able to get 58 looping consistently(mostly), but get this: when I first unmuted the track, it would overload at the first midi note. But if I hit play again, it would keep trucking through it just fine, so I got a few extra tracks. I also disabled smooth cycle algorithm (just realized, I need to set that back before I forget ha) to squeeze an extra track or two, but it was clear to me that the process buffer range was the major issue. Anybody have any opinions on the process buffer range setting and what you want to set it at? Perhaps I've misunderstood its purpose....

[fermusic]

Hi to all,
I just add my report about my System:
56 tracks run quitely on Mac Pro Quad 2.66 2009 with 6GB RAM + Motu 828MK3 at 256 sample buffer and process buffer the "small".... (but sometime overload comes without using all 8 virtual cores)

Firefox opened at the same time...
(I have also 3XUAD-1 connected using -virtuavia ATX chassis PCIe to PCI expander)

Really nice test, but ... I think that Space Designer test is not the best to check the real CPU processing power with Logic because using 54 Space Designer in a pro Mixing is very rare...
I suggest to don't use Space Designer as channel strip insert plugin... or better... usually the best way is use AUX buss for Reverbs...

I would like to know if there is a particular technical reasons related to the use of FM Pulse Shadows patch?

Anyway, I also tested the same " song" on My Dual 2Ghz G5 using 1024 and large processes buffer... I can Run 16 Tracks on 2X2 G5 and 56 on Nehalem 2.66 Mac Pro.

Great test!!!

Thanks

G

[fabriciom]

Nelaham Dual 2,26 6Gb Ram and RME AIO at 256 52 tracks with no problems with 53 it would stop after a couple of tries. Sending the card back because it has hissing problem when i move the mose and do stuff. I noticed that 7 cores where at 100% yet 8th core was at 25% and the rest unused.

[Nu-tra]

Wow I guess it's not a good time to buy the new macs. How the heck are people recording at 96k if the track counts are crapping out at 56 tracks at 44.1
I am thinking I will only get 25 tracks at 96k. I need about 40 tracks at 96k

Dang

[YUGA]

Originally Posted by Nu-tra

Wow I guess it's not a good time to buy the new macs. How the heck are people recording at 96k if the track counts are crapping out at 56 tracks at 44.1
I am thinking I will only get 25 tracks at 96k. I need about 40 tracks at 96k

Dang

Don't worry.
This bench mark test inserts Space Designer on each track to measure CPU performance. You can record/playback 40 tracks at 96KHz if your HDD is up to par unless you don't insert an insanely huge number of plugins.

[Nu-tra]

Nah not a lot of plugs. I use what I need.. I don't go crazy.

[jazzymike]

Mac Pro 2009
8 Core Xeon "Nehalem"
8 x 2.26 GHz, 11GB Ram

42 Tracks @ 32
50+ @ 256

Interface: RME HDSP Multiface (mk1)

[kaboomer]

Hey Everyone:

Sorry to resurrect such an old thread, but I'm about to invest in a new mac, and (like all of us) I'll need to pull every cost-cutting trick in the book to make this work.

Mr. Chang, Ialsoeathummus - if you're out there, would you mind explaining a little more about the production tax exemption that you scored? That was in California, right? Was it just a form you filled out at the Apple store, or did you have to jump through some state bureaucratic hoops?

If you (or anyone else) had a second to fill in the blanks a little, I'd really appreciate it.

Thanks!