Diva with Rosetta2 on Apple M1

Official support for: u-he.com
Banned
254 posts since 14 Oct, 2020

Post Sun Jan 24, 2021 12:14 am

Ploki wrote:
Sat Jan 23, 2021 11:57 pm
how many notes should each diva instance be playing?
6 as written in the 1st page

Banned
254 posts since 14 Oct, 2020

Post Sun Jan 24, 2021 12:14 am

Ploki wrote:
Sun Jan 24, 2021 12:05 am
so, Logic Pro X
I did a chord, so 3 notes per instance, quality great, BS Deep Space Diva
no multicore option, 44.1k.
@256 and "4 cores (high performance), 33 tracks (32 if logic is in native mode)
@1024 and "8 cores (4 high + 4 efficiency), 52 tracks (51 if logic is in native mode)

seems like if you're running rosetta plugs, you get better efficiency by running logic in rosetta mode.
which is too bad, because logic itself is much snappier in native mode, load time is halved.
6 notes at 48khz and 512 sample buffer size (written in the 1st page)

KVRAF
3631 posts since 17 Dec, 2009

Post Sun Jan 24, 2021 12:19 am

i'm stupid, thanks. thankfully i made aliases so it's an easy fix

KVRAF
3631 posts since 17 Dec, 2009

Post Sun Jan 24, 2021 12:31 am

ok so:
BS Deep Space Diva, 48k, 6-note chord.
M1 MBP, Logic Pro
Logic Rosetta mode, buffer 1024, 8 core setting: 23 instances. 24th plays, but drops out after one bar.
Logic Native mode, buffer 1024, 8 core setting: 23. 24th does not play at all.


i think buffer size in logic doesn't matter if its not a live track. but switching one track to "live input mode (i.e. i can play it) at 128 buffer size still plays all 23 tracks...

edit:
i can also play the 23th instance live at 64 buffer size - 8 voices of it, while 22 other instances are playing 6 voices, without crackles at all. and fans don't start.

Banned
254 posts since 14 Oct, 2020

Post Sun Jan 24, 2021 12:42 am

But why you select different buffer size but none of them are 512? :-D i mean your results are incomparable with others since the size does play important role aswel. The longer the buffer the less CPU power is required.

KVRAF
3631 posts since 17 Dec, 2009

Post Sun Jan 24, 2021 1:08 am

accessdune3 wrote:
Sun Jan 24, 2021 12:42 am
But why you select different buffer size but none of them are 512? :-D i mean your results are incomparable with others since the size does play important role aswel. The longer the buffer the less CPU power is required.
logic afaik plays all tracks in 1024 unless they're "live tracks".
Switching to 512 changed nothing.

however by using a summing stack instead of separate midi instances, i got 25 tracks instead of 23 for some reason.

live-armed tracks are a-whole-nother matter tho since logic switches them to single core by default i think.?

So i did another test.
Diva Multicore, Logic multicore setting to "playback tracks only"
LIVE TRACK, 10 divas in a track stack. (played live, 6 voices per instance)
10 tracks at 512, or at 128, buffer 64 starts glitching, 11 tracks starts glitching.
pointless, doesnt work well.

However!
Diva Single core, Logic multicore setting set to "playback and live tracks:
LIVE TRACK, 21 divas in a track stack. (played live, 6 voices per instance)
21 tracks at 512,
20 at 128,
17 at 64
15 at 32

i guess multicore distribution works much better if you leave it for Logic to do it.

i'll try pulling the same test on the Mini i7 until the guy comes to buy it in the afternoon :D

User avatar
Urs
u-he
25635 posts since 8 Aug, 2002 from Berlin

Post Sun Jan 24, 2021 3:27 am

Regarding MultiCore we have observed that Apple's new multithreading feature for Audio Units is extremely helpful. This is Big Sur or later, AU only.

It won't be long until we put out native versions with this feature enabled; we expect this to improve multicore performance a lot. As of now we have only tested this on Intel CPUs though.

KVRAF
3631 posts since 17 Dec, 2009

Post Sun Jan 24, 2021 4:02 am

I did a more structured test for Live track stack with my two macs.

BS Deep Space Diva (great), 6 notes midi region on a live track stack in logic, 4minute sustained, 48000Hz2

i7 8700B mini 32GB RAM (running on a 2k screen)
13” Pro M1 16GB RAM (rosetta, running on a 4k screen):
Screenshot 2021-01-24 at 13.03.05.png
Note that i was typing my notes on the M1 while it played in the background, Mini overloaded if i did anything at all while playing.
You do not have the required permissions to view the files attached to this post.

KVRAF
3631 posts since 17 Dec, 2009

Post Sun Jan 24, 2021 4:06 am

However when i tested spire, the finding were baffling.
I got 30% more tracks while running it in rosetta VS native.

SPIRE: 16-note sustained chord
Buffer 512
i7: 28 tracks (99 degrees, 4500RPM)
M1 Rosetta: 28 tracks (75 degrees, fans off)
M1 Native: 20 tracks (80 dregrees) (wtf?)

They must have done something wrong with the Native build. FabFilter Pro-L2 @32x oversampling test was more inline with expectations.

my conclusion, spire is broken.
You do not have the required permissions to view the files attached to this post.

User avatar
Urs
u-he
25635 posts since 8 Aug, 2002 from Berlin

Post Sun Jan 24, 2021 4:13 am

Hmmm, weird. I can assure you, our native builds are much more CPU efficient than Intel builds run with Rosetta.

KVRAF
3631 posts since 17 Dec, 2009

Post Sun Jan 24, 2021 4:15 am

Urs wrote:
Sun Jan 24, 2021 4:13 am
Hmmm, weird. I can assure you, our native builds are much more CPU efficient than Intel builds run with Rosetta.
good to hear, can't wait to repeat the test with native builds :)

And, so are FabFilter.
I couldn't get more tracks running Pro-L2 @32x oversampling, but CPU was 15degrees cooler and i had headroom for additional stuff, while Rosetta was already spinning fans, and i7 was literally screaming and begging me to stop

I'm still amazed how much butter M1 handles lower buffer sizes - even when running Rosetta.
You do not have the required permissions to view the files attached to this post.

KVRist
441 posts since 18 Nov, 2002

Post Sun Jan 24, 2021 5:08 am

Ploki wrote:
Sun Jan 24, 2021 4:02 am
I did a more structured test for Live track stack with my two macs.

BS Deep Space Diva (great), 6 notes midi region on a live track stack in logic, 4minute sustained, 48000Hz2

i7 8700B mini 32GB RAM (running on a 2k screen)
13” Pro M1 16GB RAM (rosetta, running on a 4k screen):

Screenshot 2021-01-24 at 13.03.05.png
Note that i was typing my notes on the M1 while it played in the background, Mini overloaded if i did anything at all while playing.
Thanks for the test. The most interesting thing to me is the behaviour at low latencies, its really good.

KVRAF
2048 posts since 25 Jun, 2008 from Montreal, Canada

Post Sun Jan 24, 2021 9:46 am

krraqk wrote:
Wed Jan 20, 2021 4:44 pm
accessdune3: in which DAW have you done the test?
Because I installed Bitwig demo and with the same Diva test I get 44 tracks/instances of Diva x 6 voices = 264 voices.
So the problem here maybe its not the pc but the DAW where you run the test.
Or the performance difference between DAWs is brutal or Bitwig is doing some workaround when running copied instances of the same plugin and same preset and chord notes...
xx JPRacer xx: could you try the test in Reaper? I think we will see very different results.
I just did the test in Reaper and I got the same result. 12 instances can play without problem. With 13 instances it's playing but crackling heavily.

KVRist
441 posts since 18 Nov, 2002

Post Sun Jan 24, 2021 6:15 pm

Thanks for your time.
Strange you get the same results. Its not my case, dont know what happens with Bitwig in my pc but I can run almost 40 instances, tried another time to see how many I can play in realtime even loading different presets in every instance and was able to play 6 note chords with almost 20 instances (stoped at 20 not tried more), thats insane compared to other hosts (4/5 instances in Cantabile for example is the maximum).

KVRAF
3631 posts since 17 Dec, 2009

Post Mon Jan 25, 2021 5:06 am

I got a new build of Spire, it's a bit unstable but I get 42 tracks on native build vs 28 on Intel (intel i7 or Rosetta M1 - got 28 with both).
If that's any indication of things to come, things are crazy.

Return to “u-he”