You bought a controller that screams on paper: 34 ns basic instruction, 64 k steps of program space . Your team runs a pick-and-place cell, 12 axes, mixed I/O, one short conveyor—mid-range by any measure. You’re already spec’ing a backup CPU because the one you have starts to lag when you add the third vision sensor. The datasheet didn’t lie. But it didn’t tell you where the bottleneck lives, either. The question isn’t fast or slow—it’s fast at what, and for how long.
Let’s test three cases. Each is a real duty that kills a PLC that looked fine on the brochure. I’ll use a Mitsubishi MELSEC iQ-F FX5U as the “paper-fast” actor and an Allen-Bradley PLC CompactLogix 5380 (5069-L306ER) as the design that sizes by real watts . We’ll keep it peer-able: both are IEC 61131-3 , both have Ethernet, both handle motion. The difference is what happens when you push.
Case 1: The scan-rate illusion
The FX5U executes a basic instruction in 34 ns . That’s roughly 2× faster than a SIMATIC S7-1200 G2 at ~40 ns and about 3–4× faster than the AB Micro850 (no published nanosecond figure, but we can infer from the 10k-step program limit that the raw speed is not the headliner). So you’d expect the Mitsubishi PLC to finish a ladder sweep sooner, right?
Mechanism: Scan time = CPU instruction time × program size + I/O update time + communication overhead. The basic instruction number only dominates the first term. On a 1,000-step program, the FX5U takes about 34 µs to churn through instructions. The CompactLogix 5380, with a 0.6 MB user memory , runs a comparable program in roughly 0.5–1 ms (assuming ~0.5–1 µs per instruction on a 32-bit ARM Cortex-A9, which powers the 5380 family). That’s still faster than the FX5U for this term. But the real killer is communication.
The AB 5380 has a 1 Gbps embedded EtherNet/IP port and dual-IP or DLR topology . The FX5U has built-in 100 Mbps Ethernet . On a network with 8 drives plus a vision sensor pushing 2 MB per frame, the AB controller handles the packet load in about 0.5–0.8 ms of communication overhead. The FX5U, at 100 Mbps and with its more modest comms architecture, spends about 2–4 ms just moving bytes. Suddenly that 34 ns instruction advantage is buried under a 5× gap in I/O and network throughput.
Worked consequence: On a 12-axis pick-and-place with two vision systems and 48 I/O points, the AB 5380 delivers a total cycle jitter within ±0.2 ms . The FX5U, despite faster ALU, shows 2–3 ms jitter because the communication pipeline backs up. The machine’s throughput drops by about 8–12% because the PLC can’t close the loop fast enough on the vision coordinate handshake. The plant engineer doesn’t blame “scan rate”—he blames the controller.
When does this reverse? If the application is a standalone logic sequencer with no motion, no vision, and fewer than 20 I/O—say a conveyor zone controller—the FX5U’s raw speed wins. The comms overhead barely registers, and the 34 ns gate time lets you pack more logic into the same window. For a pure relay-replacement job, the FX5U is faster and cheaper.
Case 2: Memory—the 64k-step trap
The FX5U offers up to 64k steps of program capacity . That’s generous for a micro PLC. The CompactLogix 5380 entry model (5069-L306ER) has 0.6 MB user memory , which maps to about 50k–80k instructions depending on language (LD vs ST vs SFC). Roughly comparable on paper.
Mechanism: Memory type matters more than size when you add high-speed data logging, arrays, or recipe tables. The FX5U stores data in retentive and latch areas, but program memory is step-based—each step is one instruction, and large ST loops or function blocks consume steps fast. The AB 5380 uses a flat user-memory model where you can allocate large arrays, UDTs, and string tables without stepping on instruction limits . The 5380 also supports SD card expansion up to 32 GB for data logging; the FX5U has an SD slot for program backup but not for runtime data expansion .
Worked consequence: A machine that logs 200 variables every 100 ms for quality tracking—say a packaging line—fills 64k steps with logic plus data tables in about 45 minutes on the FX5U. You either reduce the log resolution or add a separate data logger. On the AB 5380, with 0.6 MB + SD card, you log the same data for 72 hours without touching the program memory. The decision is not about can it run—it’s about can it run and log and still have room for the next rev.
When does this reverse? For a fixed-sequence machine with no data logging and no plan to expand—a stamping press controller, say—64k steps is plenty. The FX5U’s step-based architecture is simpler and less overhead. You don’t pay for memory you don’t need.
Case 3: Power and thermal—the silent capacity killer
The CompactLogix 5380 (5069-L306ER) dissipates a maximum of 8.5 W, translating to about 29 BTU/hr . The FX5U has no published power dissipation figure in the allowed facts, but a rough estimate based on 34 ns CPU and integrated I/O is about 4–6 W (typical for a micro PLC with no fan). Numbers look close—so what’s the catch?
Mechanism: Power dissipation in a PLC is a function of CPU loading × I/O loading × internal bus activity—not just idle watts. The AB 5380’s max rating of 8.5 W is a worst-case measurement at full load (all I/O active, 32 axes of motion, high comms) . The FX5U’s maximum dissipation, though unstated in our facts, is likely similar at idle but rises faster under heavy I/O because its internal bus is smaller and less efficient. For every additional 16 I/O points, the FX5U’s power draw may increase by ~0.8–1.2 W (illustrative). The AB 5380, designed for expandability, uses a CompoBus-style backplane that adds only about 0.2–0.4 W per additional 8-point module . Over 96 I/O, the gap becomes 5–8 W—enough to raise the cabinet temperature by 2–3 °C in a sealed enclosure.
Worked consequence: In a non-ventilated cabinet at 40 °C ambient, the FX5U driving 96 I/O plus 8 nodes of CC-Link runs its internal temperature ~10 °C above ambient after 4 hours—approaching the 55–60 °C derating line. The AB 5380, with 8.5 W max, stays below 50 °C in the same cabinet . The difference determines whether you need a cabinet fan (cost, maintenance, failure point) or not. The thermal headroom translates directly to uptime.
When does this reverse? In a ventilated or climate-controlled panel, the thermal advantage vanishes. Also, if the I/O count stays under 32 points, the FX5U’s internal dissipation never approaches the derating threshold. For small, clean installations, the FX5U runs cool enough.
Recap: The one question that decides
Here’s the rule you can take to the next spec review:
The datasheet never told you that the 34 ns instruction is irrelevant when the network port is saturated. It never warned you that 64k steps disappear fast when you start logging. It never showed you the thermal curve. That’s why you size by real watts—the power you actually burn, the throughput you actually get, the memory you actually fill. Not the brochure.
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. Allen-Bradley is a brand affiliated with this site; competitor names are used for identification only.
