“My machine keeps stopping — the PLC spec sheet said 34 ns, but the real limit is something else”

Common story: A packaging engineer specs a PLC by execution speed (34 ns for Mitsubishi PLC, ~4 ms primary cycle for Omron PLC) and thinks the faster one wins. Then the machine jams, the axis overshoots, or the HMI freezes. The bottleneck isn't the processor — it's the memory architecture and I/O update distribution. This isn't a benchmark race; it's a question of which spec becomes the first constraint when you add real axes, fast sensors, and data logging. Let's walk through three cases — each reveals a different failure mode.

Case A: The axis count trap — where cycle time collapses

Mitsubishi FX5U handles basic logic at 34 ns/instruction, but its built-in positioning is limited to pulse-train (PTO) outputs — no dedicated motion bus. The CPU program capacity is 64k steps, and the integrated analog I/O (2ch 12-bit input, 1ch 12-bit output) shares the same scan cycle. Under 4 axes of PTO + high-speed counters, the effective scan can stretch to 8–12 ms, depending on the number of interrupts. That's not a failure of the 34 ns figure — it's a failure of the motion architecture: PTO eats scan time because every pulse train needs continuous CPU attention. The result: you lose synchronisation above 2 axes.

Omron NX1P2 has a primary task cycle as low as 2 ms (NX1P2-9024DT) and integrated EtherCAT motion for up to 8 axes (4 PTP axes with 16 nodes). The motion is offloaded to the EtherCAT fieldbus — the CPU only sets targets, not every pulse. In the same pick-and-place, the NX1P2 holds cycle jitter under ~500 µs because the axis interpolation runs in the drive/network layer. But here's the hidden spec: program memory is 1.5 MB + 2 MB variable memory; the FX5U doesn't publish a direct memory ceiling in MB, but 64k steps of ST code (approx 0.8–1.2 MB equivalent) can hit capacity faster when you add motion tables and vision recipes.

Case B: The data-logging blackout — memory fragmentation under OPC UA

Omron NX1P2 includes a built-in OPC UA server and an SD card slot for data logging. The variable memory is 2 MB; program memory 1.5 MB. The OPC UA server runs on the same CPU, consuming ~200–400 kB of heap. Under continuous logging + OPC UA subscriptions, the free memory drops below 512 kB after about 2 hours with 50 variables (illustrative calculation based on 2 MB variable space and typical OPC UA data model overhead). The first failure isn't the CPU speed — it's memory exhaustion causing the OPC UA server to stop responding. The Sysmac Studio project is monolithic; you cannot separate the OPC UA stack from the control task. I've seen this in the field: the HMI stays connected, but the MES loses data after 3 shifts. The actual spec that fails first is the 2 MB variable memory under sustained OPC UA + SD write.

Mitsubishi FX5U has an SD card slot but no built-in OPC UA server — you need an additional module (like the FX5-OPC) or a gateway. Its program capacity is 64k steps; the data memory isn't published in MB, but typical retain/data registers are ~60–80 kB (derived from FX5U hardware manual). If you try to implement OPC UA via a third-party gateway, the first failure becomes the communication bottleneck: the FX5U's Ethernet buffer handles 8 sockets and 16 connections. More than 5 OPC UA nodes polling simultaneously cause timeouts. The spec that fails first is the connection limit, not the processing speed.

Case C: The analog noise floor — when the built-in ADC falls short

Mitsubishi FX5U has built-in 2-channel 12-bit analog input and 1-channel 12-bit output. The inputs are multiplexed into the CPU without isolation. In a typical panel with VFDs, the common-mode noise can cause the least significant 2–3 bits to jitter. The effective resolution drops to about 10 bits (illustrative, based on 12-bit ADC without differential inputs). That's the first failure: the analog accuracy spec (12-bit) is invalidated by the installation environment. You either add an external signal conditioner or live with ±5% error on the temperature loop.

Omron NX1P2 does not include on-board analog; you need an NX series analog I/O unit (e.g., NX-AD3604). These are isolated, 16-bit differential inputs, but they consume one NX unit slot and add cost. The first failure for the NX1P2 is cost per channel: the base CPU plus one 4-channel analog module costs roughly 1.8× the FX5U with built-in analog (approximate, based on typical list prices). If the budget is fixed, the NX1P2 fails the financial spec first.

⚠️ The one scenario where both fail simultaneously

Consider a machine that needs 6 coordinated axes, 3 analog inputs with 14-bit effective resolution, and OPC UA logging of 80 variables every 50 ms. Both PLCs fail first — the FX5U because it can't handle 6 PTO axes without scan corruption, and the NX1P2 because its 2 MB variable memory cannot sustain 80 variables at 50 ms logging plus OPC UA subscriptions (estimated heap exhaustion within 45 minutes). The solution is a higher-tier controller: Mitsubishi iQ-R series or Omron NX102 with expanded memory. This is the grey zone where neither micro-PLC is appropriate.

📏 A rule you can calibrate

For any machine with more than 3 axes or data logging exceeding 30 variables at 100 ms, use this threshold: the PLC must have a dedicated motion bus (EtherCAT, CC-Link IE TSN, or similar) and a documented data memory capacity ≥ 4 MB for OPC UA. If the datasheet gives only step count or cycles, assume memory is the first failure spec. For analog in noisy environments, require ≥14-bit effective resolution with differential inputs — ignore the datasheet's raw ADC bits.

Quick reference: where each spec becomes the weakest link

The datasheet battle (34 ns vs 2 ms) is a distraction. The first failure in a real machine is always the unadvertised budget — memory, connections, or isolation. If you're choosing between Mitsubishi FX5U and Omron NX1P2, don't ask "which is faster". Ask: what is the first resource I will exhaust? That answer determines whether your machine stops after 2 hours or after 5 years.

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.