thanx for the useful infos..
infact the most cpu consuming things from my tests are properly the oscillators...i hope something will be done to improve a them...
thanx
lalo
Urs wrote:Hey lalo,
there are some things that might take up more cpu than they should. I.e. some modules don't go into standby mode if there's nothing to do and thus might even suffer from denormalization. Candidates are RingModulator, Shapers and even some filter types.
Regrading SSE/Altivec, there's not much that could benefit. For instance, the filters are faster now *without* SSE/Altivec than their counterparts in Zebra1.5 which use Altivec. It's generally easier to parallelize software with a static layout (i.e. if one can calculate multiple identical modules at once), but this can hardly be done in Zebra where everything has to work in arbitrary order. Another obstacle is, the power of SIMD can only be unleashed on 32-bit audio data (single precision floating point). Now, most modules in Z2 require 64-bit precision...
The one thing that benefits from Altivec big time on the Mac are the oscillators, due to MacOS X' very fast FFT implementation. I'm gonna check out if buying the corresponding Intel libraries will give us a speed bump, but I want to wait and see if there's a chance to buy them for both Win and the new Intel Macs.
However, here are some tips that might help to make individual patches faster:
- mute unused Grid channels (red square between Grid and corresponding mixer)
- lower "Resolution" for oscillators in corresponding "more OSCx" tabs if possible
- avoid "LP Vintage" and "LP Vintage2" filter modes if possible, these are expensive analog simulations
- use one Osc in "dual" or "quad" mode rather than 2 or more Oscs in "single", if possible
- use shorter Release times on VCA envelopes (but don't set them to 0 or you might get clicks)
- use "few" voices or even switch to "legato" or "retrigger" voicing modes
Cheers,
Urs