What units take the most CPU?

Hi folks,
I’m very new to the 301. I’m working with a little performance skiff where I have only a Klee sequencer and the 301. My goal is to create a Pad, Drums, Bass, and Melody all on the 301 and each in it’s own Mixer unit.

The plan is to basically have a Dub recorder and effects on each “mixer channel” and I’ll record the incoming CV from the Klee for each channel and then switch the Dub recorder to “0” Dub so that I can move on to the next track and reuse the Klee to record a new sequence without changing the previous sequence.

I got it working, but I’ve already gotten massive slow down and about 80% CPU usage and starting to get clicks and really slow GPU.

So my question is - what are the Units that use the most CPU? I’m using a HPF and LPF on each voice and tried to do VCOx3 on one of the voices. For Drums I’m scanning the “start” position of another sequencer to build a drum pattern. I have a couple reverbs and delays-I’m guessing those are probably the biggest culprits?

I’m hoping to refactor this so it’s easily manageable with my 301. Thanks for any tips!

Is there any way to set up a delay and reverb as a “send” that I can send other mixer units to so that I don’t have to create more than one of each?

Reverbs are a likely culprit. I made a video about setting up a send/return setup. Perhaps it will help.

@a773 has graciously allowed me to publish the results of his experiments here:



Thanks for sharing that video Joe! I would have never figured this out without that! Got reverb and delay auxes set up now. So fun!


This is brilliant!! Thanks to @a773 and @odevices for sharing - perfect timing as I’m putting together a new liveset and the ER301 is the hub/mixer for everything so I need to optimise the patches to manage the CPU load.

1 Like

Glad you found it useful!


@a773 Thanks mate!! :blush:

Were you able to reduce the overall CPU of your patch with this technique?

I find the two different findings, 30 units vs 100 units, for the Linear Bipolar VCA to be interesting.

Yeah, when it’s complete I’ll dig into it. For now @odevices mentioned that he would consider the CPU measurements more accurate when close to 80% than at lower stress levels (paraphrasing here, correct me if I’m wrong).

Working through this I found that the 301 is not linear in performance. Units take up more time towards 80% where things start to break up. Also all operations (copy, insert new units, clearing chains) is slower when the CPU is stressed…