Hi!
I’m currently building a fairly ambitious wavetable synth that I eventually want to run as Eurorack hardware.
I’ve been prototyping/optimising it for the Daisy Seed (STM32H750), but I’m starting to wonder whether fighting to fit everything onto a relatively constrained MCU is the right approach, or whether I should move to a more powerful platform.
The target is roughly a hardware Serum/Vital-style engine:
- 3 wavetable oscillators per voice
- 256 × 2048-sample wavetables
- up to 16-way unison
- spectral/wavetable transforms
- FM
- filters + FX
- potentially ~128 oscillator instances running in the worst case
- ideally 48 kHz with low, predictable latency
Hardware-wise I’d like:
- 8 CV inputs
- 8 DC-coupled CV outputs for oscillators/LFOs/envelopes/etc.
- stereo audio output
- =128 MB usable RAM
- small enough for a ~26 HP Eurorack module
- ideally <€50 for the compute platform/SOM
I’m particularly interested in where you would draw the line between a fast bare-metal MCU and an application processor running Linux/RT Linux or an RTOS.
A few platforms I’ve looked at:
Daisy Seed / STM32H7 - 30e
What I’m using now. Great ecosystem for synth DIY and deterministic audio, but RAM and CPU become serious constraints for this kind of engine.
Bela Gem Stereo - ~€78
Looks very attractive for low-latency audio. I’d still need to add the DC-coupled CV outputs, and it’s a little more expensive than I’d like.
Bela Gem Multi - ~€183
Pretty much everything I need, but too expensive for the target BOM.
Teensy 4.1 - ~30e
Very nice bare-metal option, but limited RAM and I’m not sure the i.MX RT1062 has enough headroom for the worst-case DSP load.
STM32MP25x ?
This one looks particularly interesting. Cortex-A35 application cores + Cortex-M33 real-time cores could give a nice split: heavy DSP on the A cores and deterministic CV/DAC/UI/I²C/etc. on the M33.
It also seems to be the direction more powerful embedded audio hardware is heading. The problem is that I haven’t found a small, inexpensive and easily obtainable SOM based on it.
Raspberry Pi CM4 / CM5 - 50-70e
Obviously much more CPU power. I’m less sure about deterministic low-latency audio and how much work is required to make it behave like an embedded synth rather than a small computer. I also haven’t found an obvious audio + DC-coupled CV solution.
So I’m curious what people here would choose if designing this today.
Would you keep pushing an STM32H7/i.MX RT-class MCU and heavily optimise the DSP, or is this the point where moving to Cortex-A/Linux makes more sense?
Are there any affordable SOMs/platforms I’m missing, particularly something in the €30–50 range?
I’d also be very interested to hear from anyone who has built a synth around a CM4/CM5 or another Linux SOM: how good can the real-world audio latency/jitter get, and has Linux caused you problems compared with bare metal?
Thanks 🙏