Fixed: Serious WTF... I’m getting same DSP % in AUM with same preset on new iPad Air3 as on my Air2

Sigh... Got new iPad air3 setup. Air dropped an AUM present from old Air 2. It’s a pretty intense setup in AUM with lotsa fx and synths...

I’m seeing about the same damn DSP/cpu usage on this new “better iPad as on the ancient Air 2!

What gives?!?

On paper this Air3 should be like 2x at least faster... What am I missing here?

«1

Comments

  • If you want to compare horsepower then push your instruments, polyphony and fx amount to the max until you hear crackles/distortion/interrupted audio and see how many the Air 2 and how many the Air 3 can handle.
    CPU load is a bad indicator for judging how much a cpu can handle.

  • CPU monitors in iOS aren't accurate representations and shouldn’t be relied on. For one thing, the device curtails processor power when not needed to prolong battery life. You could well see CPU utilization go down by loading it more heavily.

    What counts is how much you can load it down without getting crackles.

    Btw, are the audio settings (latency, sampling rate, and audio resolution) the same on both? Those all affect processor utilization too.

  • Try this. I use it to get general idea of cpu/ram load between my air 2 and pro 10.5. Pinch of salt recommended though. Some apps are cpu, others Ram hungry. If your apps eat up your ram and cpu is twice as fast and ram is the same there will not be a massive improvement in performance. Just an example I don’t know your devices’ specs.

    System Status Lite by Kenichi Yoshikawa
    https://itunes.apple.com/gb/app/system-status-lite/id992911130?mt=8

  • @wim 👏🤝😅

  • Just goes to show how good the Air2 was :smile:

    Seems like every other iPad is the one to go for.

  • @BiancaNeve said:
    Just goes to show how good the Air2 was :smile:

    Seems like every other iPad is the one to go for.

    Still chugging away with mine here. I have very few problems compared to so many frustrated posts with “2018” and “Pro” in them, I wonder if I’ll ever upgrade.

  • I never pay attention at DSP % in AUM.

    I don't care. If it start to crackle, just shutdown some channels. Fixed.

  • edited May 13

    I have the biggest concert I’ve performed in a while coming up next week and my Air2+mini4 combo is doing the job exceptionally. Rock solid and smooth running loooooads of shit, including AB/AUM with model D, half dosen other synths and a bunch of sound processing fx and utility stuff on air2 and a fully loaded BM3 session on mini4.

    No update coming soon.

  • I recently bought the 2018 iPad ,mostly because my Air2 battery is getting old.
    It feels like a downgrade, it’s heavier , thicker and I can feel a gap in the screen, plus swipe doesn’t feel as smooth as my Air2.
    For audio performance ,the 2018 gives me some performance boost but will further test on small buffer size .

  • When I load up instances of Model D or Factory in AUM on my Air 3, the CPU hovers around 50% until I load a 3rd, when the CPU actually goes down. As @wim said, it’s the way the software manages the tasks. Push it a bit harder and more seems to become available.

  • We need a good article in the wiki about this topic. It comes up so often. Bram or Jonatan wrote a great post that I'd love to re-find about it.

  • @MonkeyDrummer My Air2 is smooth as is my ipad pro 1st Gen. Apple have really messed up with latests ipads it seems. Hope you get it sorted.

  • @espiegel123 said:
    We need a good article in the wiki about this topic. It comes up so often. Bram or Jonatan wrote a great post that I'd love to re-find about it.

    I think what wim said below sums it up quite nicely.

    @wim said:
    CPU monitors in iOS aren't accurate representations and shouldn’t be relied on. For one thing, the device curtails processor power when not needed to prolong battery life. You could well see CPU utilization go down by loading it more heavily.

    What counts is how much you can load it down without getting crackles.

    However, the DSP% meter in AUM is very accurate, but the value is not meaningful in the ways you might think, since it shows the utilization of the CPU-time available at the current CPU speed (which is unknown, controlled by iOS, and varies all the time to handle the current load).

  • I find the AUM meter is very good — if I aim for a range of 55-65 total DSP then I’m all safe for live performance — it can be pushed a bit but this amount gives a little headroom and keeps things running reliably.

  • edited May 13

    Yeah it seems pretty accurate to me. When I get DSP to high 90s % then crackling starts to creep in.
    I think the main things causing cpu issues in AUM is the iOS graphics displaying GUIs etc

    Might be an idea if AUM had a custom macro controller view where you can have the most optimal performance without needing to open guis and scroll around etc :)

  • edited May 13

    @MonkeyDrummer

    Air 2 cpu (A8x) single core geekbench is 1828, Air 3 (A12) is 4700 .. so there is approx 2.5x CPU performance gain .. try for example same project which eats lets say 90% of CPU on Air 2 run on Air 3 and you will be for sure under 40% of load :)

  • @rs2000 said:
    If you want to compare horsepower then push your instruments, polyphony and fx amount to the max until you hear crackles/distortion/interrupted audio and see how many the Air 2 and how many the Air 3 can handle.
    CPU load is a bad indicator for judging how much a cpu can handle.

    Pretty much this ^

    The only sort-of-accurate measurement you can make is comparing the point where things break. Active CPU core throttling makes all other comparisons virtually meaningless. :)

  • I feel your pain, I had this happen to me when I got a new but old iPad 4. I thought I could be ripping audulus and multi midi audio things but no way my cpu meter crashes audulus in standalone with an admitedly complex patch. I was like almost crying ha, but you will find workarounds and such. I am sorry but I bet it will rock here soon!

  • @dendy said:
    @MonkeyDrummer

    Air 2 cpu (A8x) single core geekbench is 1828, Air 3 (A12) is 4700 .. so there is approx 2.5x CPU performance gain .. try for example same project which eats lets say 90% of CPU on Air 2 run on Air 3 and you will be for sure under 40% of load :)

    Unfortunately it's rarely that simple... :) CPUs/processor architectures handle different task loads with greater/lesser efficiency. All a geekscore really tells you is how well that processor handled the tasks that they decided were representative. Are those tasks representative for audio? Probably not...

  • @wim said:

    @BiancaNeve said:
    Just goes to show how good the Air2 was :smile:

    Seems like every other iPad is the one to go for.

    Still chugging away with mine here. I have very few problems compared to so many frustrated posts with “2018” and “Pro” in them, I wonder if I’ll ever upgrade.

    I’d like to be able to run a few more instances of the newer synths, so eyeing up the Air 3, but considering my Air 2 has been going for over 4 years I’m pretty chuffed with it.

  • wimwim
    edited May 13

    A suggestion for meaningful comparisons...

    • Make a session with one instance of the sequencer of your choice (Rozeta, Atom, StepBud, Photon ... whatever) sending out a typical pattern.
    • Add a channel with one app you use often, and one FX that you use often. Try to pick apps that generally seem well-behaved.
    • Set the channel to receive from the sequencer.
    • Duplicate that channel as many times as it takes to get crackles.
    • Note the sample rate, buffer size, and bit depth settings if available in the host.
    • Duplicate the setup in the other hardware
    • Repeat the above test on the other hardware and see how many channels can be handled.

    CPU metering can be done differently from host to host, so it’s not valid to compare CPU readings between hosts. It is valid to test the instances different hosts can handle, though.

    IMO, CPU meters are helpful as a safety margin check as @OscarSouth mentions, but can be misleading for other purposes.

  • @wim said:
    A suggestion for meaningful comparisons...

    • Make a session with one instance of the sequencer of your choice (Rozeta, Atom, StepBud, Photon ... whatever) sending out a typical pattern.
    • Add a channel with one app you use often, and one FX that you use often. Try to pick apps that generally seem well-behaved.
    • Set the channel to receive from the sequencer.
    • Duplicate that channel as many times as it takes to get crackles.
    • Note the sample rate, buffer size, and bit depth settings if available in the host.
    • Duplicate the setup in the other hardware
    • Repeat the above test on the other hardware and see how many channels can be handled.

    CPU metering can be done differently from host to host, so it’s not valid to compare CPU readings between hosts. It is valid to test the instances different hosts can handle, though.

    IMO, CPU meters are helpful as a safety margin check as @OscarSouth mentions, but can be misleading for other purposes.

    Yea, that would be a good test protocol. The new iPad seems to be a lot snappier, and I can see how the dsp meter may not really be like a rev counter... I’m thinking AUM and iOS act more like a car with a CVT... I was thinking the upgrade would be more like adding a blower and a Hurst. 😏

  • @MonkeyDrummer said:
    Yea, that would be a good test protocol. The new iPad seems to be a lot snappier, and I can see how the dsp meter may not really be like a rev counter... I’m thinking AUM and iOS act more like a car with a CVT... I was thinking the upgrade would be more like adding a blower and a Hurst. 😏

    😂😂😂😂

  • @MonkeyDrummer said:

    @wim said:
    A suggestion for meaningful comparisons...

    • Make a session with one instance of the sequencer of your choice (Rozeta, Atom, StepBud, Photon ... whatever) sending out a typical pattern.
    • Add a channel with one app you use often, and one FX that you use often. Try to pick apps that generally seem well-behaved.
    • Set the channel to receive from the sequencer.
    • Duplicate that channel as many times as it takes to get crackles.
    • Note the sample rate, buffer size, and bit depth settings if available in the host.
    • Duplicate the setup in the other hardware
    • Repeat the above test on the other hardware and see how many channels can be handled.

    CPU metering can be done differently from host to host, so it’s not valid to compare CPU readings between hosts. It is valid to test the instances different hosts can handle, though.

    IMO, CPU meters are helpful as a safety margin check as @OscarSouth mentions, but can be misleading for other purposes.

    Yea, that would be a good test protocol. The new iPad seems to be a lot snappier, and I can see how the dsp meter may not really be like a rev counter... I’m thinking AUM and iOS act more like a car with a CVT... I was thinking the upgrade would be more like adding a blower and a Hurst. 😏

    Be interested to hear how you get on with it after a few weeks use, since this would be my upgrade path too.

  • @dendy and others did some interesting benchmark comparisons using Zeeon. Can someone point me to that thread? (I'll add it to the wiki and then we build some nice articles from the discussion)

  • @supadom said:
    Try this. I use it to get general idea of cpu/ram load between my air 2 and pro 10.5. Pinch of salt recommended though. Some apps are cpu, others Ram hungry. If your apps eat up your ram and cpu is twice as fast and ram is the same there will not be a massive improvement in performance. Just an example I don’t know your devices’ specs.

    System Status Lite by Kenichi Yoshikawa
    https://itunes.apple.com/gb/app/system-status-lite/id992911130?mt=8

    Great point

  • @espiegel123 said:
    @dendy and others did some interesting benchmark comparisons using Zeeon. Can someone point me to that thread? (I'll add it to the wiki and then we build some nice articles from the discussion)

    https://forum.audiob.us/discussion/32132/new-ipad-mini-performance-test

    I did some more testing comparing a 11" 3rd Gen Pro with 10.5" 2nd gen Pro and 9.7" 1st Gen Pro to compare which one I wanted to keep. At some point I'll get around to posting the results....

    Spoiler: I decided to go with the 10.5" 2nd Gen Pro.

  • @Carnbot said:
    Yeah it seems pretty accurate to me. When I get DSP to high 90s % then crackling starts to creep in.
    I think the main things causing cpu issues in AUM is the iOS graphics displaying GUIs etc

    Might be an idea if AUM had a custom macro controller view where you can have the most optimal performance without needing to open guis and scroll around etc :)

    Hi @j_liljedahl :)
    I know we discussed a bit in the past, but there seems to be a significant difference in the graphics performance in AUM to other hosts, testing new Sugar bytes Factory and the menu and refresh rate is slow in AUM but much faster in apeMatrix and Audiobus. And this could explain why the cpu spikes are more significant in AUM when opening guis etc. At least this is what I'm noticing on Pro 2017 12.9 ipad. Have you noticed anything like that in your setup?

  • @MonkeyDrummer said:

    @wim said:
    A suggestion for meaningful comparisons...

    • Make a session with one instance of the sequencer of your choice (Rozeta, Atom, StepBud, Photon ... whatever) sending out a typical pattern.
    • Add a channel with one app you use often, and one FX that you use often. Try to pick apps that generally seem well-behaved.
    • Set the channel to receive from the sequencer.
    • Duplicate that channel as many times as it takes to get crackles.
    • Note the sample rate, buffer size, and bit depth settings if available in the host.
    • Duplicate the setup in the other hardware
    • Repeat the above test on the other hardware and see how many channels can be handled.

    CPU metering can be done differently from host to host, so it’s not valid to compare CPU readings between hosts. It is valid to test the instances different hosts can handle, though.

    IMO, CPU meters are helpful as a safety margin check as @OscarSouth mentions, but can be misleading for other purposes.

    Yea, that would be a good test protocol. The new iPad seems to be a lot snappier, and I can see how the dsp meter may not really be like a rev counter... I’m thinking AUM and iOS act more like a car with a CVT... I was thinking the upgrade would be more like adding a blower and a Hurst. 😏

    Ok, I tried this out. Used Viking which is kind of a cpu pig running into AudioEffX all driven by Particles. Got up to 8 channels at 256.

    Airdropped the setup to my Air 2 and it had a fucking aneurysm!

    So yea, the DSP meter is a good gauge for headroom but seems pointless for inter-device benchmarking.

Sign In or Register to comment.