From Brian Willoughby Sent Tue, Jan 23rd 2018, 22:49
From Brian Willoughby Sent Tue, Jan 23rd 2018, 22:49
My experience with other DSP systems is that %CPU measurements are not = precise. The Catch-22 is that increasing the precision increases the = amount of CPU used by the measurement itself. Therefore, most systems = opt for the simplest measurement algorithm to maximize the availability = of CPU for more important calculations, at the expense of accuracy. Basically, the only perfectly accurate way to know that you=E2=80=99ve = used 100% of the CPU is when audio goes south. One potential resource other than CPU is memory. Delays obviously use = the most, but some other algorithms might use memory. Your clever reuse = of modules may end up requiring some amount of additional buffering. Unfortunately, I=E2=80=99m not familiar with Nord Modular, although I = wish I had one. I=E2=80=99m already suspicious that Note Detector is 0%, = because I don=E2=80=99t see how it=E2=80=99s possible to implement any = feature with no CPU (except, perhaps, routing of existing data to = existing remote streams, or something that doesn=E2=80=99t actually = require CPU). Brian On Jan 23, 2018, at 12:00 PM, Mike Perkowitz <xxxx@xxxxxxxxx.xxx> wrote: > Anyone here a Nord Modular (g1) expert? I'm working on a patch, and it = says resources are used up and mutes audio even though I'm not at 100% = of DSP. Aside from a vague reference in the manual to "other resources" = being consumed, I can't find anything about what other limits there = might be. Anyone know? I'd like to figure out what it is I'm actually = using up. >=20 > For context, the patch is an attempt to fit an entire 16-sound drumkit = into a single slot. I'm up to 14 sounds, about 87% DSP, but I get "full" = when I add even 0% modules like Note Detector. There's a fair bit of = routing and logic to make it work, but nothing that seems extreme. >=20 > If you're curious, I'm doing things like using the same oscillator for = multiple tom sounds with the trigger determining a pitch offset; having = two snares and rim shot share oscillators; and using the same filtered = noise for closed hat, open hat, and cymbal. It's... challengingly fun. >=20 > mike >=20