TMC2209 every time. Reasons:
1. Silent at any speed (StealthChop)
2. Sensorless homing if you ever want it
3. UART config — set current in software, no jumpers
4. Built-in thermal protection
A4988 / DRV8825 work but are noisier, hotter, and don't do sensorless homing. Cost difference is ~€2/driver. No reason to pick the older ones for a new build.
Short version for a hot-wire cutter: all three will move your gantry without breaking a sweat — a foam cutter asks almost nothing of the motors. So the choice comes down to noise, smoothness, and setup convenience, and on those the TMC2209 wins clearly. A4988/DRV8825 are the cheap, loud, proven option.
A4988 — the classic "StepStick"
Allegro A4988. Cheapest, most common, what every RAMPS/CNC-shield was built around.
~8–35 V motor supply; realistically ~1 A/phase bare, up to ~2 A with a heatsink + airflow.
Microstepping up to 1/16 (set with the MS1/MS2/MS3 jumpers).
Fixed mixed-decay → noticeable mid-range resonance and whine, and no current reduction at standstill (motors hum and stay warm even when parked).
Current limit (standalone): I = Vref ÷ (8 × Rsense) — Rsense varies by board (0.05–0.1 Ω), so check your sense resistors before trusting any "Vref = X" chart.
DRV8825 — the beefier drop-in
TI DRV8825. Same footprint as the A4988 but the pot is on the opposite end and the current formula is different — don't blind-swap and reuse the same Vref.
Higher headroom: 8.2–45 V, ~1.5 A bare, up to ~2.2 A cooled. Better high-speed torque.
Microstepping up to 1/32, slightly smoother decay than the A4988, has a FAULT pin + over-current/thermal protection.
Current limit: I = Vref ÷ (5 × Rsense); on the usual 0.1 Ω boards that's the famous Vref ≈ amps ÷ 2.
Still a step/dir driver with no standstill current drop — quieter than an A4988 but not quiet.
TMC2209 — the silent, smart one (recommended)
Trinamic. 4.75–29 V, ~1.4 A RMS continuous (≈2 A peak), ~2 A RMS with cooling.
StealthChop2 makes it near-silent at the speeds a foam cutter runs — genuinely the difference between "is it even on?" and a shed full of whine. SpreadCycle kicks in for torque at higher speeds.
Accepts up to 1/16 step input but interpolates internally to 1/256 (MicroPlyer) → glassy-smooth motion regardless of what you feed it.
UART mode: set run/hold current, microstepping and modes in software — no Vref pot, no jumpers. Also runs noticeably cooler than the other two.
Bonus features that matter on a clean build: StallGuard (sensorless homing — skip the limit switches) and CoolStep (drops current under light load, which a foam cutter almost always is).
Downsides: a bit pricier, and UART wiring/addresses add a little setup effort.
Why most of the spec sheet doesn't matter here
A hot-wire cutter moves a thin wire through foam at a few mm/s — the load is trivial. So the DRV8825's 45 V and high-speed torque are wasted, and you'll run low current on any of them (set the motors gently; they barely work). What you actually feel day-to-day is noise during long cuts and setup hassle — both of which favour the TMC2209.
On ESP32 + FluidNC specifically
TMC2209 has native FluidNC support — wire the UART pin, set run_amps, microsteps, stealthchop, even sensorless homing right in the YAML. No pots, no jumper guessing. On a shared UART you give each driver an address via its MS1/MS2 pins.
A4988/DRV8825 are "standard" step/dir to FluidNC — set microstepping with the board jumpers, set steps_per_mm in YAML, and dial Vref by hand. Simpler wiring, louder result.
Verdict for this machine: TMC2209 in UART mode — quiet, cool, software-configured, optional sensorless homing. If you already have A4988s/DRV8825s in a drawer, they'll cut foam perfectly well; just set a low Vref and expect some singing.
One safety note: whichever you pick, set the current low for a foam cutter and check motor temperature — overshooting Vref cooks both driver and motor, and these machines need almost no torque.