Selection guide
How to size an S7-400 CPU against a PLC-5/40 scan time
Short answer
No S7-400 CPU maps one-to-one to a PLC-5/40 by instruction timing, and none needs to: every CPU from the 414 upward is faster on logic execution. Select on memory, I/O and communications capacity, then accept the choice only when the translated program's measured worst-case cycle lands at or below two-thirds of the measured PLC-5/40 scan.
On this page
- What scan time do I compare when replacing a PLC-5/40?
- What are the PLC-5/40's documented timing numbers?
- Which S7-400 CPU is comparable to a PLC-5/40 on speed?
- How do I estimate the PLC-5/40 scan before I measure it?
- How do I measure the real PLC-5/40 scan time?
- How do I benchmark the translated program on the candidate CPU?
- When is the CPU selection accepted?
- What if the benchmark misses the target?
- Questions asked
Size the replacement on complete controller cycles, not on catalog fragments: the PLC-5/40's per-rack remote I/O figure and an S7-400's bit-operation time describe different work, and neither predicts the cycle time of your translated program. On raw logic speed every S7-400 CPU from the 414 upward clears a PLC-5/40, which executes typical ladder in two to three milliseconds per 1K — so choose the CPU on memory, I/O and communications capacity, then prove it with a benchmark. The sections below give the documented numbers, the estimate arithmetic, the measurement points and the acceptance rule that turns a hunch about the 414 into a decision.
What scan time do I compare when replacing a PLC-5/40? #
The number that decides the selection is the complete scan — all inputs refreshed into the image, all logic evaluated, communications serviced, outputs written — which is the cycle the process actually experiences, not the remote I/O scan and not an instruction benchmark.
The per-rack remote I/O figure is the scanner's time to service one rack on the remote I/O link. A bit-operation time is the execution cost of a single instruction on a named CPU under its documented test conditions. Your machine runs none of these in isolation: it runs the whole cycle, and the whole cycle is the only fair comparison between a PLC-5/40 and an S7-400. If what the process truly needs is an input-to-output response rather than a scan figure, measure that response on both machines, because the scan is only one component of it.
What are the PLC-5/40's documented timing numbers? #
The PLC-5/40's documented figures are 10 ms of remote I/O scan per rack at 57.6 kb/s, 7 ms at 115.2 kb/s and 3 ms at 230.4 kb/s, plus approximately two to three milliseconds of logic execution per 1K of typical ladder.
| Timing figure | What it measures | Documented value | Role in sizing |
|---|---|---|---|
| Remote I/O scan per rack | Scanner servicing one rack on the remote I/O link | 10 ms at 57.6 kb/s; 7 ms at 115.2 kb/s; proportionally less at 230.4 kb/s | One input to the baseline, never the whole scan |
| Logic execution | Typical ladder execution per 1K of program | Approx. two to three milliseconds per 1K | Pre-measurement estimate only |
| STI last and maximum scan | Timed interrupt task | Status words S:54 and S:55, in ms | Separate task budget |
| PII last and maximum scan | Process interrupt task | Status words S:56 and S:57, in ms | Event-path budget |
| S7-400 bit-operation time | One bit instruction on the named CPU | Published per CPU in the vendor's technical data | Family ranking, not cycle prediction |
The per-rack remote I/O numbers scale with the link's baud rate, so the baseline moves if the link speed changes. The per-1K logic value is an approximation for typical ladder, not a guarantee for a specific program. Treat every row above as an input to an estimate, then measure.
Which S7-400 CPU is comparable to a PLC-5/40 on speed? #
Benchmark a 414 first: every S7-400 CPU at that level and above outruns a PLC-5/40 on logic execution by a wide margin, so anything at 414 level or above is the right candidate list and the wrong final answer; the benchmark is the selection.
There is no equivalence table between the two families and you do not need one. Siemens ranks S7-400 CPUs against each other with per-CPU instruction execution times in the technical data, and every CPU at 414 level or above sits far ahead of the PLC-5's per-1K logic execution. That makes raw speed a weak discriminator and capacity the real one: program memory, I/O count, and the communications load the CPU must service. A CPU with ample memory and I/O can still miss a cycle target, so treat capacity as pass/fail checks and speed as a benchmarked result. Expect the shape of the workload to change in translation as well — the families organize tasks and I/O scans differently, the split covered in how ControlLogix and S7-400 PLC architectures differ.
How do I estimate the PLC-5/40 scan before I measure it? #
Multiply the program size by the per-1K logic range and add the per-rack remote I/O time at your link speed; a 12 K program with three racks at 57.6 kb/s lands at 54-66 ms.
Illustrative inputs, not measurements: a 12 K program of typical ladder and three remote racks on one 57.6 kb/s link.
- Logic: 12 K × the low and high ends of the per-1K range = 24 ms to 36 ms.
- Remote I/O: 3 racks × 10 ms = 30 ms.
- Complete scan estimate: 24 + 30 = 54 ms up to 36 + 30 = 66 ms.
- The same three racks at 230.4 kb/s: 9 ms total, so 33 ms to 45 ms.
The scanner services the link while the processor runs logic, so the straight sum is a conservative upper band, and the STI and PII tasks are separate budgets on top. Decision from the estimate: if your process tolerance is tighter than the band's top number, the PLC-5/40 was already running near its limit and the remote I/O architecture — not the CPU badge — is what to redesign during the migration.
How do I measure the real PLC-5/40 scan time? #
Read the controller's status file: it carries last and maximum scan-time words for the main scan, and S:54 through S:57 report the last and maximum scan of the STI and PII tasks in ms.
Log the main scan's last and maximum values over a full shift, covering normal production and the heaviest expected communications and logic load. Record the link baud rate and the operating conditions alongside the numbers. The gap between the last and maximum values tells you what messaging peaks cost. That measured worst case — not any catalog figure — is the baseline the S7-400 candidate has to beat. Do not compare one controller's best case against the other's worst.
How do I benchmark the translated program on the candidate CPU? #
Run the translated program with representative I/O and communications configured on the candidate and read its measured cycle time, first at representative load, then at the heaviest expected logic and communications load.
Configure the real I/O and messaging before measuring, because the replacement I/O network's update time is part of the cycle you are buying. Translate the actual logic, not a toy program — restructuring changes execution paths, the effect you can watch in porting a PLC-5 filter routine to ControlLogix structured text. Run the benchmark through the heaviest expected production condition, read the cycle time the engineering tool reports for the running CPU, and record the worst case under that load.
When is the CPU selection accepted? #
Accept when the candidate's measured worst-case complete cycle is at or below two-thirds of the measured PLC-5/40 worst-case scan and never above the process's response limit; a result above the baseline is a rejection, and one between the two is a workload problem.
Continuing the example: if the measured PLC-5/40 worst-case baseline is 60 ms, the acceptance target is two-thirds of that, 40 ms. A candidate measuring 35 ms under worst-case load passes with headroom for future messaging. A candidate at 55 ms has not endangered the process but has no headroom, so profile the workload before paying for a bigger CPU. Confirm memory and I/O capacity as separate pass/fail checks, the same benchmark-then-buy discipline used in sizing a CompactLogix to replace four SLC 500 processors. If the timing limit exists to prevent harm — a guard, an overtravel stop, an overflow interlock — that function belongs in a safety-rated system, not in a faster standard CPU.
What if the benchmark misses the target? #
Move up one CPU in the family and retest only after you have identified which part of the workload eats the time — I/O update, communications, or logic — because a bigger processor does not shorten a network's update period.
Profile the cycle first: how much of it is I/O update time, how much is communications servicing, how much is logic execution. Fixing the dominant slice is usually cheaper than the CPU step, since a faster processor leaves a message queue or an update period exactly where it was. If the profile shows logic execution dominates, move up one CPU within the family and retest once. If the second test still misses the target, stop selecting and escalate to the vendor's support channel or the responsible system integrator with both measurement sets in hand.
Questions asked
Can I keep the PLC-5's remote I/O racks with an S7-400?
No. The S7-400 family scans distributed I/O over its own networks, such as PROFIBUS DP or PROFINET, so the PLC-5's remote racks must be replaced or served through gateways. Budget the replacement network's update time into the benchmark, because on a migrated machine that number, not the CPU, often sets the floor.
How do I test input-to-output response instead of scan time?
Drive a spare input with a pulse, wire the output it triggers, and measure the delay with a scope or a high-speed counter. Budget the worst case as roughly one complete scan plus one I/O update period: an input that arrives just after the image refresh waits a scan before logic sees it, and the output then waits an update period before the field device moves.
How long should the benchmark run before I accept a CPU?
Run it through the heaviest expected production condition — the longest recipe, peak messaging, peak operator load — not just idle cycling. Match the exposure of the baseline: log the PLC-5/40 status-file maximum over a full shift so communications peaks are captured, then compare the two worst-case numbers.
Is the per-1K logic estimate valid for my program?
Only as a first estimate. The documented range approximates typical ladder, so a program heavy in math, file operations or messaging runs longer per 1K, while simple bit logic can run faster. Use it for the estimate, then let the measured status-file maximum set the real baseline.