Reaktor 6: poor performance under Apple Silicon

SaschaFranck
SaschaFranck Member Posts: 163 Advisor
edited October 22 in Reaktor

Hi,

What could I possibly say? Reaktor 6's performance under macOS Sonoma on a brandnew Macbook Air M3 is downright abysmal. Worst performance of anything I used on that machine now. And fwiw: It performs quite a lot better (!) on my 2010 2x2.66GHz Mac Pro running Mojave.

In pretty much all typical DAW scenarios, the Macbook has around 3-4 times the singlethread juice of the Mac Pro, but with Reaktor I can not use Super 8, Form and some others at all, at least not at my audio settings (44.1kHz, 32 samples buffersize - and no, I'm not going to raise the latter at all, it's been one of the key reasons for me to buy the new Macbook).

Fwiw, Kontakt 7 Player performs lousily, too. Totally non fluid UI behaviour. Compared to Kontakt 6, it's pretty much shocking just how bad it is.

What gives?

Tagged:
«1345

Comments

  • Sunborn
    Sunborn NKS User Library Mod Posts: 3,099 mod

    32 samples buffer size is too low for some Reaktor beasts (and not only). Of course the performance also relies on CPU, the quality of your audio card and other things. Through the years i have found that for best results you should use a buffer size of 128. However, every setup is different so for your machine could be 64 or 256 and the best way to find out, is to experiment. But for most setups 32 is too low, at least for a heavy performance.

  • SaschaFranck
    SaschaFranck Member Posts: 163 Advisor

    No, 32 samples is not too low. Super 8 is working fine with that setting on my Mac Pro from 2010. The new Macbook has around 3-4 times more singlethread oomph. Pretty much all other plugins are taking advance of that, just some NI stuff (namely Kontakt 7 Player and Reaktor) are not.

    It's entirely NI's fault. And I won't sacrifice my super low latency performance (one of the reasons I got that new machine) just because NI can't get their silicon-Mac act straight. They need to work on this urgently.

  • Sunborn
    Sunborn NKS User Library Mod Posts: 3,099 mod
    edited May 1

    Sorry, but it is too low, at least partially. If you don't want to take my word (a professional musician for decades) for it, then, check on some audio websites.


    You see, nobody uses a 32 buffer size constantly (maybe some super-top studios, or some super expensive audio cards)… for the rest of us, things are a bit different… especially those who are doing audio engineering too (mastering etc.) they even go up to 1024 if necessary…

    You can use a 32 samples buffer while you recording or while you playing live with midi instruments, but for playback you will not avoid problems with that buffer size… it highly depends on your situation though, you may run easily a DAW project of 8 channels but you will not be able to run a project with 80 channels!

    so… buffer size is not something that we simply "set and forget" but something that we adjust if and when necessary.

  • SaschaFranck
    SaschaFranck Member Posts: 163 Advisor
    edited May 1

    Woohoo, argumentum at verecundiam - way to go!

    I was hoping we could leave that out and deal with some hard facts only, but ok, now that we're at it already:

    • Professional musician since decades
    • Studied jazz/rock/pop at the local music university (Hannover)
    • Establishing the medialab for a new course at that very university some years back
    • Betatesting for pretty big names (*cough*) since around 1999 or so
    • Building audio workstations from scratch ever since, too

    So, I damn sure know what buffersizes are about - but all that is absolutely irrelevant! And I absolutely don't care about what you or anyone else has to say about working at 32 samples buffersize - because you seem to miss the most important point: It works! It works with each and every other plugin I own but Reaktor and partially Kontakt 7.

    But as if that wasn't enough already, the main point you're missing is this:

    "Super 8 is working fine with that setting on my Mac Pro from 2010."

    But hey, keep trying to not read what I've been saying and try to educate me - seriously, the NI fanboyism 'round these parts is a pretty sad affair.

  • SaschaFranck
    SaschaFranck Member Posts: 163 Advisor
    edited May 1

    And as far as your links go, the second one is like troubleshooting 101, the last one is from 2004 and in the first one we find this here:

    "For recording, you should set the buffer to as low a value as your
    computer can handle. 64 samples is a good place to aim for; 127 samples
    can be adequate;
    32 samples - if you can manage it - is an excellent
    place to be.
    "

    So much about "32 samples is too low" and your attempt trying to back this up "scientifically".

  • SaschaFranck
    SaschaFranck Member Posts: 163 Advisor

    And as a last one on the "32 samples is too low" matter: I can run 36 instances of Raum (NI's reverb, which you will possibly be familiar with, default setting) on a stereo "live" track (record enabled, hence monitoring through it, hence using just one CPU core) at 32 samples buffersize.

    I can do any other test you like, but each test will show that this machine has singlecore power (exactly what you need for playing things such as Reaktor in realtime) up to the grand wazoo.

    And still: just one single instance of Reaktor and Super 8 loaded will cause crackles when playing a single voice.

    But hey, keep telling me it's the low buffer size and hasn't got anything to do with Reaktor. Right.

  • Maciej Repetowski
    Maciej Repetowski Member Posts: 674 Guru

    Of course 32 it’s not too low. Actually, at least Logic works better with smaller buffer on Apple Silicon Macs than with bigger one (which is counterintuitive, but still true). There was a thread about it at Vi-Central some time ago.

    Simple answer is Reaktor is old code made compatible with M series, but they didn’t optimise it otherwise.

  • SaschaFranck
    SaschaFranck Member Posts: 163 Advisor

    Thanks - finally a voice of reason.

    And well, it's a pretty sad affair. NI is abandoning product support as if there was no tomorrow since years already, trashing most excellent plugins - and in case they keep their products alive, it often ends up in a fiasco. Battery 4 is utter ****** compared to Battery 3 (which was damn excellent), Kontakt 7 and Reaktor seem to run horribly bad on silicon-Macs.

    Looking at Reaktor, I'm wondering whether they even have any Mac testers left. Because the thing is an entire red flag and releasing it that way would be considered a showstopper some years back.

  • Paul B
    Paul B Member Posts: 164 Advisor
    edited May 1

    You only give the CPU specs of the two machines, without regard for what other differences there might be except the OS version.

    Looking up the 2010 Mac Pro, that model is most likely the 12-core model (it has 2 6 core 2.66GHz CPUs as you described). Which you have replaced with a how many core M3?

    The highest spec MacBook Air M3 I can find has 8 cores. Wikipedia says that is a 4 performance, 4 efficiency core CPU. So you've dropped by 4 CPU cores and have 2/3 as many as before, and only 4 of the remaining 8 are performance cores.

    What DAW are you using? Some DAWs will only use the performance cores, so on my 12-core M2 Mac mini, Ableton Live will only use the 8 performance cores. In your case it would use only 4 cores. 1/3 as many as your Mac Pro had.

    You say "In pretty much all typical DAW scenarios, the Macbook has around 3-4 times the singlethread juice of the Mac Pro". This was measured how? When running the problem instruments, are you running them solo or with multiple other instruments and effects running?

    "Pretty much all other plugins are taking advance of that". Software does not generally get to decide how much CPU it can consume, beyond the settings in Reaktor and Kontakt to scale back certain things if their use goes above a certain percentage.

    What percentage CPU use do you see in Reaktor's CPU meter when used standalone on the new machine, vs. on the Mac Pro?

    Reaktor performs significantly better on my Mac mini than it did on my previous Intel Mac, and Kontakt also, and while running at a lower buffer size - 128 vs the 1024-2048 I used on my old Mac. I can go lower except when using certain Kontakt instruments, but since there is no need to (latency at that buffer size is already below the point of human perception) and I don't want to have to change it to handle those Kontakt instruments when I use them, I'm fine with 128.

    As an example, using Toybox Audio's Tangle Synth in Reaktor, two instances would make my Intel Mac fall over. I got up to around 15 on the Mac mini without a problem (don't recall the exact number, may have been one or two more). Live's CPU meter was close to the limit at that point. 15 is more than enough so I stopped there.

    In part it's that the CPU cores are vastly superior, and there are more available for Ableton to use (8 vs 4 on the Intel Mac), and also that the MacBook would regularly thermal throttle when under heavy load while the mini shows no signs of this. The MacBook Air has no active cooling, so thermal throttling is a definite possibility (even active cooled MacBooks may do this if the load is high enough).

    So your experience, while valid, isn't necessarily a problem with Reaktor or Kontakt. That I have the opposite experience means, at the least, that chances are good the problem is somewhere else, or that it is specific to certain computer and software configurations (which, once identified, can be investigated for a resolution). It certainly is not a universal "these apps are just bad" situation.

  • Sunborn
    Sunborn NKS User Library Mod Posts: 3,099 mod

    i am happy for your "titles" and may you get even more, to make a nice collection for your elder days…

    however, since you mentioned "hard facts", here are the hard facts… there are only 2 things you can do:

    First, you keep complaining and do nothing about your problem, while waiting for NI to do something about it (which it might take long time, if it happens at all, because it depends also on how many users reported the same problem… so, if its just you, they will probably do nothing, unless if the software gets an update).

    Second, you can choose to be less arrogant and more compromising, and try to work at 64 buffers, until your problem is solved.

    Either way, your "NI fanboy" wishes you good luck (seriously, i can not give you any other advice, in my Windows i work with 128 buffers and it is really OK for me, i can easily go down to 64 too, but i keep it at 128 because i also do a lot of audio engineering too… i never have any problems at studio, while i can do decent live sets without any noticeable latency, so i am happy with this! …and neither am i a NI employee to give you a more clear insight)… in fact, i am sorry that i tried to help you, how silly of me! …i won't do it again… 😊

  • SaschaFranck
    SaschaFranck Member Posts: 163 Advisor
    edited May 1

    "Second, you can choose to be less arrogant and more compromising, and try to work at 64 buffers, until your problem is solved. "

    Pardon? I bought this machine in order to NOT having to work at 64 samples.

    The only arrogant replies are yours as you try to defend NI at all cost when it's entirely their fault.

  • SaschaFranck
    SaschaFranck Member Posts: 163 Advisor
    edited May 1

    The cores of either machine are totally irrelevant as there's always only one in use when you're playing a virtual instrument.

    Before you start throwing around technical things, please be so kind and stick to the relevant things. In this case it's singlethread performance. Which my Macbook Air /w M3 CPU is extraordinarily great at. All possible tests prove this - as long as the software is coded properly. Which Reaktor very clearly isn't.

  • oobe79
    oobe79 Member Posts: 127 Advisor

    I put Sunborn on ignore a long time ago. I'd suggest you do the same. He's one of the most active apologists on the Forum, and you will get nowhere with him when it comes to troubleshooting something that is potentially NI's fault.

  • SaschaFranck
    SaschaFranck Member Posts: 163 Advisor
  • Kubrak
    Kubrak Member Posts: 3,067 Expert

    Number of cores is well relevant. Up to certain number. OS and programs run hundreds of threads. Or DAW may need few cores just to work, and few other for plugins.

    I am not on Mac, but what may happen is, that your DAW (or OS) allocates Reaktor threads to e-cores… And they are not able to handle it…

    If it is possible to switch e-cores off on Mac, try to switch them off and try if Reaktor runs better or not.

    Also speaking about Reaktor. Reaktor is just code compiler and the "program" may be from decently demanding to CPU cruncher, depending on what ensemble one loads to Reaktor.

    So one cannot say Reaktor kills my CPU, maybe certain ensemble does, others would not….

    If you want to have buffer 32 samples, you need to have good quality audio interface and strong computer. Four performance cores is not strong computer by any means in current days. You should know that, if you build workstations for audio.

    It may be that code compiler in Reaktor could make better code for AS. But NI does not develop compilers, they have used 3rd party one. If developer makes improvements it may reflect in Reaktor one day…

    You should understand that AS is still rather new thing, it will take several more years, till things will be ironed out. That time Apple will switch to other CPU type and the whole thing will repeat again….

This discussion has been closed.
Back To Top