Hardware platforms and profiles
The Pi 5, Pico 2, Daisy and ESP32-S3 target lanes, their delivery paths and the evidence still needed on real boards.
The pedal workbench currently registers four target producers. They are experimental build and evaluation lanes, not a promise that every module fits or that a device is qualified. This scope was reviewed on October 2, 2026.
| Target ID | Platform | Delivery path | Current evidence and limits |
|---|---|---|---|
linux-pi5 | Raspberry Pi 5, Linux ARM64 | Effect .so and resident audio/MIDI service; SSH install/rollback tooling | Cross-build and host-native audio/MIDI checks exist. ALSA, install, timing and recovery have not been exercised on a Pi. Static realtime audit failures remain. |
rp2350-pico2 | Raspberry Pi Pico 2, Cortex-M33 | Linked firmware, UF2 through USB BOOTSEL; CDC telemetry | Firmware builds and QEMU checks exist, including USB-MIDI effects. Audio needs the profile's external codec wiring. Memory and realtime fit are per module. No physical-board qualification. |
daisy-seed-h750 | Daisy Seed, Cortex-M7 | Linked firmware through USB DFU; CDC telemetry | Firmware builds and QEMU audio checks exist. SRAM/SDRAM placement is recorded in the build. MIDI effect mode is absent. No physical-board qualification. |
esp32-s3 | ESP32-S3-DevKitC-1 N8R8, Xtensa LX7 | ESP-IDF firmware/ELF, serial flashing and telemetry | Experimental canonical Processor export and Espressif QEMU checks. Proposed ADC/switch, isolated UART MIDI and external duplex I2S wiring. PSRAM fit does not establish realtime capacity; physical qualification pending. |
A Raspberry Pi 5 package and a Pico 2 UF2 are different artifacts. Selecting a board profile determines the build and transport; it does not grant compatibility by itself.
Fit is a property of a module and a board
Guitarabin illustrates why this matters: its current MCU instance needs about 874 KB of state, exceeding Pico 2 SRAM. It builds for Daisy with external memory, but emulated processing exceeds its indicative CPU budget. Compiler lowering of 64-bit arithmetic is one active performance issue.
Small MIDI transforms can be useful earlier: MIDI Transpose and Chorder have Pi and Pico 2 build/emulation paths. They still require device MIDI, timing, stuck-note and recovery tests before a supported release.
A multieffect can be expensive even when most slots are disabled. The current Multifx implementation prepares state for many algorithms; an October 1 source-state scan found 12,186,204 bytes of typed-array backing storage at 48 kHz, before objects, generated adapters and runtime overhead. Choosing phaser and distortion on its face does not reduce that allocation. A smaller hardware configuration needs a change in the canonical module/shared engine, then fresh state and sound verification.
Platforms under investigation
Elk Audio Stomp/Jackpot and Roland/Lydia have integration studies and earlier packaging experiments. They are not target producers in the current pedal workbench. Their plugin/runtime and vendor integration paths need separate qualification. Other boards require an explicit profile and adapter; a similar CPU is insufficient evidence.
What a profile means
A profile describes the intended architecture, audio layout, rate and block constraints, logical controls, display and deployment requirements. Development profiles can describe a virtual control surface. They do not prove that pins are wired, a codec works or a timing budget is met.
On the device, record the exact board revision, codec, firmware/OS, clocks, memory placement, controls and artifact hash. Then measure processing cycles, xruns, round-trip latency, noise/clipping, MIDI behavior and recovery. The target is qualified only for the tested configuration and module release.
See Hardware architecture for the host relationship and From module to hardware to start an evaluation.