I’m assuming we are talking about my experience with a Capybara-320 w/ 16 processors?
The challenges that I remember (it was a long time ago) generally where memory related. When loading up large samples, wavetables with many entries, or whatnot I’d end up trying to do something which didn’t fit in whatever amount of SRAM was attached to a DSP. On the compute end of the spectrum I typically hit “OUT OF REALTIME” when using/modifying the polyphonic sounds Ed Eagan and others put together for use with a Continuum Fingerboard.
In the hands of a skilled Kyma user (not something I would classify myself as) I’m sure one could have rung out more from the hardware I had before running into limits. It was obvious to me that when I ran into problems it was often a case where the Sound (DSP network) in question couldn’t be split up sufficiently to allow work to be scheduled across more/all of the DSPs. I like the experience with the Pacarana hardware much more because I don’t feel like I have to pay anywhere near as much attention to the cost (it terms of compute and/or memory) for the signal processing chains that I’m laying down.
