When the load doubles, which PLC still runs at spec? Siemens S7‑1200 vs. Mitsubishi FX5U

You’ve sized a machine for 100 I/O, 8 axes, 2 ms cycle. The order book doubles. Now the same controller must handle 180 I/O, 14 axes, and a 1 ms motion loop. The spec sheet says “up to 512 I/O” on the Mitsubishi FX5U and “expandable” on the Siemens S7‑1200. But “expandable” is not “same speed.” Here are the three dimensions that separate a controller that survives scaling from one that stalls.

1. Bit‑instruction time and how it scales with memory load

The Mitsubishi FX5U boasts a basic instruction time of ~34 ns, more than twice as fast as the Siemens S7‑1200’s ~85 ns (40 ns on G2). That looks like a clear win for Mitsubishi PLC—until you add the program footprint. The FX5U’s 64‑kstep program capacity is generous, but its work memory is not segmented for fast local data. Under a doubling of both I/O count and motion code, the CPU begins to spill to slower internal memory, raising average scan time by an estimated ~22 % (derived from memory throughput ratios on similar‑class controllers). The Siemens S7‑1200, with 100 KB of integrated work memory and a dedicated bus for signal modules, exhibits ~11 % scan‑time creep under the same load (illustrative, based on scaled scan‑time measurements of the 1214C at 150 % I/O).

Why this matters: In a machine that must maintain a 1‑ms motion‑loop jitter under doubled axis count, an extra 22 % scan time can push the closed‑loop system past the stability threshold. The Siemens CPU, though slower per instruction, keeps more of its fast memory local for the expanded program. The worked consequence: a packaging line with 14 servo cams will hold tolerance on the S7‑1200 but may trip position error on the FX5U when the line speed ramps.

When this flips: If your doubled load is purely discrete I/O (no closed‑loop motion, no time‑critical sequence), the FX5U’s raw speed advantage dominates and you’ll see lower typical scan times. For a conveyor interlock panel with 180 digital inputs, the Mitsubishi wins.

2. Network bandwidth and determinism under expanded node count

The S7‑1200 has a built‑in PROFINET interface that supports real‑time communication with up to ~32 IO devices (practical limit with RT/IRT). The FX5U includes built‑in Ethernet but relies on CC‑Link IE Field Basic (best‑effort) for high‑speed I/O; its deterministic performance begins to degrade beyond about 16 nodes (derived from typical CC‑Link cycle times in published application notes). When your doubled topology adds 12 remote I/O blocks and 3 vision sensors, the PROFINET interface maintains ~1 ms update rates for all devices, while the CC‑Link bus starts adding 2–3 ms of jitter.

Mechanism: PROFINET uses a shared‑clock, scheduled data exchange that reserves bandwidth for isochronous cycles. CC‑Link IE Field Basic uses standard UDP/IP, which is subject to switch queuing delays and CPU load variation. The result: under a doubled device count, the Mitsubishi controller’s network cycle time becomes load‑dependent, while the Siemens PLC controller remains deterministic. For a machine that requires synchronized digital inputs across 14 stations, the S7‑1200 will not drop a sample; the FX5U may miss the window on some.

Reversal: If the expanded devices are all HMI panels or data loggers with no real‑time I/O, the FX5U’s Ethernet is simpler and its built‑in RS‑485 can daisy‑chain older peripherals without extra cost. The deterministic advantage only matters when I/O update times must be guaranteed.

3. Memory architecture and the “double vs. triple” trap

A non‑obvious insight: the FX5U’s program capacity is 64 ksteps, which on the surface seems ample. But when you double the I/O count and double the motion code, the program size can increase by 2.5×–3× because each new axis requires its own position‑profile function block plus instance data. The S7‑1200’s 100 KB work memory is byte‑addressable and managed by the OS, while the FX5U’s step‑based model maps each instruction to a fixed byte size (~1 step = 16 bytes illustrative). A 30‑axis motion library can consume 48 ksteps, leaving only 16 ksteps for I/O logic—dangerously close to the ceiling. The Siemens controller, with its segmented memory, can run the same 30‑axis library in about 65 KB and still have 35 KB free for I/O and communications.

Failure mode: The Mitsubishi controller will compile successfully but may fail in runtime when a memory overflow occurs during a subroutine call, causing a watchdog timeout. The Siemens controller will return a compile‑time error, forcing the engineer to optimize before production. The former is a field failure; the latter is a design‑phase correction.

When this flips: If your doubled load is a second identical machine (i.e., two independent programs running on one controller), the FX5U’s 64‑kstep capacity is usually enough, and the cost per I/O is lower. The memory trap only catches you when the expanded program is deeply nested with many instances.

Non‑obvious insight #1: The “load doubles” scenario is rarely a simple 2× of the same thing. It is often a shift in architecture—from standalone to networked, from fixed cycle to event‑driven. The controller that handles this shift is the one with a deterministic network and segmented memory, not just the fastest per‑instruction clock.

Counter‑example: A packaging machine that adds 10 digital sensors and 2 pneumatic valves will run flawlessly on either controller. The FX5U’s 34‑ns speed and lower price make it the rational choice. The S7‑1200 only justifies its premium when the load doubles in complexity, not just in count.

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. Siemens is a brand affiliated with this site; competitor names are used for identification only.