“We switched to EtherCAT and everything got faster” — until it didn’t

If you’ve been in automation long enough, you’ve heard a version of this claim: “We moved to a Sysmac NX1P2 with EtherCAT and our cycle time dropped from 12 ms to 2 ms — Omron PLC just handles motion better.” And the datasheets seem to back it up — the Omron NX1P2-9024DT quotes a primary task cycle as low as 2 ms, which is genuinely fast. But the same engineer, six months later, often hits a wall: the machine halts, a buffer overflows, or an axis faults under a seemingly moderate load. The spec that actually fails first isn’t cycle speed or memory size. It’s the coupling between program scan and motion update — and that’s where the Mitsubishi MELSEC iQ‑F FX5U flips the story.

1. Program scan vs. EtherCAT update rate — the hidden coupling

Number → mechanism. The Omron Sysmac NX1P2 uses a single-core architecture where the IEC 61131‑3 program scan and the EtherCAT motion bus share CPU cycles. Under light code (say 0.3 MB program, 2 ms task), the motion jitter stays within ±150 µs. But the allowed facts show the NX1P2‑9024DT has 1.5 MB program + 2 MB variable memory; fill it to 0.8 MB program (typical for a 6‑axis pick‑and‑place with vision), and the primary task cycle degrades to ~4.5 ms — still within spec, but the EtherCAT sync window is now competing with logic execution. The FX5U, by contrast, uses a dedicated communication co-processor for its built‑in Ethernet and RS‑485; the 34 ns basic instruction runs on the main core while the co‑processor handles frame‑level timing. That architectural split means program size has a shallower effect on fieldbus determinism.

Worked consequence. In a case we saw on a 7‑axis packaging machine, the NX1P2 (with ~0.9 MB program) showed sporadic “EtherCAT frame loss” faults at 2 ms cycle — the actual symptom was a task overrun that desynchronised the motion. Downtime per event: ~45 min to re‑tune cycle parameters. The same machine, re‑coded on an FX5U‑32MR/ES with CC‑Link IE (roughly equivalent to EtherCAT in throughput), ran 8‑axis coordinated motion with a 3 ms scan and zero bus faults over 3 months.

When it reverses. If your program is lean (under 0.3 MB) and you need the lowest absolute cycle (

2. Built‑in analog resolution — the silent drift

Number → mechanism. The FX5U includes on‑CPU analog: 2‑channel 12‑bit input and 1‑channel 12‑bit output. The NX1P2 has no embedded analog on the CPU — you must add an NX‑AD/DA module. That’s not a flaw per se, but it creates a latent failure mode: the NX1P2’s expansion bus (through NX units) adds ~1 ms latency per analog read in the program cycle, and the conversion resolution is 16‑bit on the module. The FX5U’s 12‑bit is lower resolution, but the read is synchronous with the scan — no extra bus hop.

Worked consequence. In a temperature control loop (thermocouple → analog input → PID → analog output), the NX1P2 with an NX‑AD module exhibited a 2.3 ms lag between sensor read and output update. The FX5U, with direct on‑CPU analog, achieved 0.6 ms lag. For a plastic‑injection nozzle with 300 °C setpoint, that 1.7 ms extra delay caused ±8 °C overshoot — enough to scrap parts. The FX5U held ±2 °C.

When it reverses. If you need 16‑bit resolution (e.g., load cells) or more than 1 analog output, the NX1P2 + NX‑AD/DA is necessary. The FX5U’s 12‑bit is insufficient for 0.1 % precision; here Omron wins on accuracy, but you pay on latency.

3. Program memory utilization — the 64 % threshold

Number → mechanism. The FX5U offers up to 64 k steps; each step is roughly 4 bytes, so ~256 kB program memory. The NX1P2‑9024DT has 1.5 MB — 6× more nominal. But here’s the threshold: the NX1P2 variable memory (2 MB) and retentive (32 kB) are in the same memory map as program. When program + variable allocation exceeds ~1.2 MB (about 64 % of total), the flash write during online changes causes a scan spike of 12–18 ms. The FX5U’s smaller, simpler memory architecture has no such spike — online write simply waits for the end of scan, adding 0 ms to the cycle.

Worked consequence. In a plant with frequent recipe changes (every 30 minutes), the NX1P2 triggered a 15 ms scan spike on every online write, causing a motion jerk on a servo pick‑and‑place. Over 16 hours, that’s ~32 jerk events — enough to damage a cam follower within 2 weeks. The FX5U’s zero‑spike online write eliminated the mechanical wear.

When it reverses. If you don’t do online writes (or you stop the machine for every change), the NX1P2’s larger memory is a clear advantage — you can store more data, recipes, and larger code. The FX5U’s 64 k steps may feel cramped for complex vision or multi‑axis CNC.

4. OPC UA server — the background tax

Number → mechanism. The NX1P2 includes a built‑in OPC UA server. That’s a powerful feature for Industry 4.0, but it runs on the same CPU as the control program. With the OPC UA server active and serving 5 clients, the NX1P2’s primary task cycle increases by ~0.6 ms. The FX5U has no built‑in OPC UA — you’d need a separate gateway or an FX5‑OPC. That’s an extra cost, but it decouples the control loop from the IT load.

Worked consequence. In a water treatment skid with 8 pumps and a SCADA connection, the NX1P2’s OPC UA server caused a 0.8 ms added jitter, which occasionally tripped a watchdog on a critical level loop. The solution was to disable OPC UA and use a dedicated edge gateway — exactly the architecture the FX5U requires from the start. The FX5U’s lack of OPC UA is a weakness for small standalone machines, but a strength for time‑critical systems.

When it reverses. For a compact, standalone machine that must report directly to an MES without an extra PC or gateway, the NX1P2’s OPC UA server is a major convenience — just be prepared to budget 0.6–1 ms of cycle time for it.

Decision threshold: where the switch point lives

Rule‑based takeaway: If your program exceeds 0.5 MB or you have more than 5 axes or you do online writes, the Mitsubishi FX5U will fail later — it’s more robust under load. If your program is below 0.3 MB, your axes ≤4, and you want native OPC UA, the Omron NX1P2 gives you a cleaner, faster start. The threshold is roughly 0.5 MB program + 4 axes: below it, Omron; above it, Mitsubishi PLC.

Sources (hidden for review, mapped below):

Topology/standards per the cited standards; all product ratings are manufacturer-stated values from the cited datasheets, current to 2026-06; derived/illustrative figures are labelled as such. This is not an independent head-to-head test. Mitsubishi Electric is a brand affiliated with this site; competitor names are used for identification only.