Are there performance issues with Nuendo and Apple's M4 Max CPU?

Nuendo
16

We observed many interesting result. To briefly conclude what I and my friends have got these days:

  1. I have not come across any compatibility issue on M4Max (4e+12p) yet.

  2. Cubendo project scaling tests. M4Max is as slightly more than 2× powerful as M4 base(6e+4p) on my Uplifting Trance projects with tones of synths and plugins. Similar to Cinebench R23 scores.

  3. For x86 chips: We have no economy to test Threaddripper or any newer Xeon. So only ‘consumer brand chips’.

  4. For most of those popular synths eg. serum/serum2, pigments, Diva, M4 series is more efficient than any x86 we have tested (intel core 12th 13th ultra1 ultra2, AMD Ryzen 7000s, 9000s, x3d/non x3d). On average M-series chips handles the most polyphony in a single instance; only AMD 7950x(3d)/9950x(3d) sometimes handles more ‘total’ polyphony across multiple instances than M4Max 'coz of more cores and threads. Intel chips perform the worst.

  5. Acustica: M4 p-core ≈ Zen5 > Zen4 >= M2 p-core > Meteor Lake P-core,

  6. Fabfilter plugins Zen5&Zen4 chips a little better than M4 chips. Intel not yet tested.

  7. Kontakt 7 gets the worst efficiency on M series Macs. Maybe because macs limit 4k random read, even when you use internal ssd, or simply because NI stuff are not optimised for Apple silicon.

  8. Different plugins with QUITE DIFFERENT working conditions. Some plugins eats cache/RAM IO so AMD x3d(large L3 cache) and apple chips(large L2 and SLC cache) and lower RAM latencies and higher RAM bandwidth machines get benefits. Some might be sensitive to core to core latencies (for example SpecCraft when you use a lot of them. In fact 7950x3d handles less paralleled SpecCraft than 7800x3d (same on 9950x3d and 9800x3d)…what?). Some are only bottlenecked by float point calculation speed.

  9. Different buffer size also QUITE DIFFERENT results. In general when using low buffer size x3d and M-chips are more efficient. Importance for realtime work: Larger cache, Low RAM Latency > Higher single core performance > Higher RAM bandwidth >> others eg. higher core count. However higher core count benefits way more when you’re dealing with Large projects and you don’t require super low RTL.
    [7. & 8.] ==> If someone really wants the absolute ‘best value’ then he have to go and test by himself with his own workflow.

  10. Intel big.LITTLE cpus suffers for most DAWs or plugins. Problematic thread scheduling.

  11. win11 24h2, macOS 15.4.1. We have rule out DPC problems on Desktop Wins.

2 Likes