Home - Power Supply Knowledges - Can You Parallel Two LED Drivers? Usually No — Here’s What Works Instead
Can You Parallel Two LED Drivers? Usually No — Here’s What Works Instead
Every few weeks somebody asks whether they can put two 200 W drivers in parallel to get 400 W, because the 400 W model is out of stock, or because the enclosure has two small bays instead of one big one, or because they want redundancy and assume that paralleling is how you get it.
The answer for a constant voltage LED driver is almost always no, and the reason is not conservatism. It is the ±5% output tolerance printed on every one of these datasheets, including ours. Two units that are both “24 V” can be 22.8 V and 25.2 V, and when you tie their outputs together, physics does not average them. The higher one takes the whole load.
This article explains what actually happens when you try, what the legitimate ways to share load are, and the one scenario where redundancy genuinely pays for itself.
What the tolerance figure really means
A constant voltage driver regulates its output within a band. Ours is ±5%, which on a 24 V unit means anything from 22.8 V to 25.2 V at the terminals, at 25 °C, at nominal load, before you account for the temperature coefficient and the cable drop. That is not a defect; it is the specification, and it is the same ±5% that most of the industry publishes.
Now connect two of them to a common 24 V bus.
The unit sitting at 25.2 V sees a bus at 25.2 V and is happy. The unit at 22.8 V sees a bus 2.4 V above its set point, so its error amplifier winds the drive down to zero and it stops contributing. It sits there, powered, doing nothing, while the first unit carries 100% of the load.
When the load rises to the point where the 25.2 V unit hits its current limit, its output collapses toward the load line, the bus voltage drops, and the second unit wakes up. Now both are conducting, but not in a controlled ratio — the sharing ratio is determined by the shape of two slightly different fold-back curves, and it drifts with temperature. You have built an oscillator with a 24 V output.
The failure mode when one unit dies is worse. If it dies open-circuit, the other carries everything, which is fine until it also runs at 100% continuously and its capacitor life halves. If it dies with a shorted output — which happens with a failed output diode or a collapsed primary — the healthy unit drives current straight into the short and follows it down.
Why “but it works on the bench” is not evidence
People do parallel drivers and report that it works. It does, for a while, under three conditions: the two units happen to be close in output voltage, the load is well below the combined rating, and neither unit reaches current limit.
That combination is common in a lab test at 50% load for twenty minutes. It is not common in a signage installation running 18 hours a day through a summer, where the two units are at different temperatures because one is nearer the vent, and their output voltages have drifted apart by 300 mV.
The test that tells you the truth takes an hour: run the pair at 80% of combined rating for 45 minutes, then clamp a DC current probe around each output cable separately. If the two readings are within 10% of each other, you have a sharing mechanism. If they are 90/10, you have a hot spare that is slowly cooking.
The four options that actually work
| Option | Sharing accuracy | Redundancy | Relative cost | When it makes sense |
| Droop-sharing drivers (specified for parallel) | ±10% | 1+1 with ORing | 130% | OEM projects, 500+ units |
| ORing diodes / ideal diode modules | Determined by driver | 1+1 or N+1 | 115–125% | Kiosks, signage, remote sites |
| One larger single unit | n/a | None | 100% | Almost always, if a bigger model exists |
| Split the load across circuits | n/a | Partial | 105% | Two independent lighting zones |
Droop sharing
The proper way to parallel is to design the driver for it: add a deliberate output impedance so that the output voltage falls by, say, 0.5 V between zero and full load. Two droop-sharing units then find an equilibrium on the shared load line, and the sharing accuracy is set by the slope tolerance, typically ±10%.
This is a design change, not a wiring trick. It has to be built into the feedback loop, it usually needs a current-share bus between the units, and it costs money. We build it on OEM projects, usually at 400 W and above, for customers who need 800 W or 1,200 W in a single enclosure. It is not something you can retrofit to a standard catalogue model by joining two output terminals.
ORing diodes and ideal diode modules
For redundancy, the standard approach is one Schottky diode per supply, cathodes joined. Now a shorted supply is isolated by its own diode, and the healthy unit keeps the bus alive.
The cost is a forward drop: 0.4 to 0.6 V on a Schottky at rated current. On a 24 V bus that is 2% of your voltage, and 0.5 V × 16.6 A = 8.3 W of heat in the diode for a 400 W unit. You will spend more on the heatsink for the diode than on the diode.
Ideal diode controllers replace the Schottky with a MOSFET and a controller IC, dropping the loss to 20–30 mV. They cost €8–15 per channel, they need a bias supply, and they are the right answer for sealed enclosures where 8 W of extra heat is not acceptable.
Either way, the sizing rule is the one people get wrong: in a 1+1 redundant pair, each supply must be rated for the full load. Two 400 W units in 1+1 give you 400 W of protected capacity, not 800 W. If you need 800 W with redundancy you buy two 800 W units, and the cost curve stops being funny very quickly.
One larger unit
If a 600 W model exists and fits the enclosure, it is cheaper, more efficient and more reliable than two 300 W units plus diodes. One set of input components, one set of output capacitors, one control loop, one thermal path.
Our 600W IP20 indoor power supply exists for exactly this reason. It is 24 V at 25 A, in a single 250 × 110 × 50 mm case, and it replaces a parallel pair in most signage and kiosk designs at about 85% of the cost of the pair plus the ORing hardware.
The constraint is usually mechanical, not electrical: if the enclosure has two 120 mm bays and no room for a 250 mm case, the single unit is not an option, and that is a legitimate reason to look at sharing. Just be honest about which problem you are solving.
Split the load
The unglamorous answer that works: divide the load into two independent circuits, each with its own driver. Two runs of LED strip, two signage faces, two halves of a lightbox.
You lose single-point redundancy — a driver failure still takes out half the sign — but you gain fault isolation, simpler wiring, standard catalogue parts, and no diode losses. For a double-sided pylon sign, this is almost always the right architecture, and it is what most signage OEMs actually do once they have been through the parallel experiment once.
The redundancy question, priced properly
Redundancy is worth paying for when the cost of downtime exceeds the cost of the hardware by a wide margin. That is true for a roadside variable message sign that requires a lane closure to service, and it is true for a payment kiosk in a supermarket where a dark screen costs sales. It is not true for a warehouse high-bay where a failed fixture is one of two hundred and someone changes it next week.
A quick way to decide: multiply the callout cost by the expected failure rate. A driver with a 20,000-hour life running 24/7 fails roughly once every 2.3 years. If a callout plus lane closure costs €600 and the redundant pair costs €180 extra, redundancy pays back within the first prevented failure, with margin. If the callout costs €60 and the redundant pair costs €180, you are buying insurance at three times the expected loss.
Note that this calculation only works if the redundancy covers the dominant failure mode. In a constant voltage supply, the dominant field failure is the input electrolytic capacitor drying out — which is a wear-out mechanism that affects both units at almost the same time if they are in the same enclosure at the same temperature. A 1+1 pair of identical drivers in the same box does not protect you against that; it just means both fail in the same month instead of one failing twice.
That is the part the redundancy sales pitch leaves out. To get real protection against wear-out, the two units need to be thermally independent — different positions, different airflow, ideally different manufacture dates — and even then you are only buying the difference in their temperature histories.
The failure mode nobody designs for
There is one more thing worth saying, because it bit a customer of ours on a kiosk project.
When two supplies share an output bus, a slow drift in one unit’s regulation — not a failure, just a 400 mV rise over 18 months as a reference ages — pulls the whole bus up. The loads see 24.6 V instead of 24 V, which is inside their tolerance, so nothing trips. Then the other unit, which is now permanently backed off to zero, has its own output capacitors sitting at a voltage they are not regulating, and its control loop is in a state the designer never tested. When the first unit finally fails, the second one has to take over from a standing start, and about a third of the time it doesn’t.
The fix is a per-unit output voltage monitor and an alarm, or an ORing arrangement where each unit only ever sees its own output node. Both are design decisions, not installation decisions, and both are why “just parallel them” is a sentence that should make a reliability engineer wince.
What to ask before you parallel anything
Five questions, in order:
- Is the driver specified for parallel operation, with a stated sharing accuracy? If the answer is not a number, the answer is no.
- What is the output voltage tolerance and the temperature coefficient? ±5% over temperature is very different from ±1%.
- What happens at current limit — fold-back, hiccup, or constant current? Fold-back units are the worst to parallel because they retreat instead of sharing.
- In a redundant pair, is each unit rated for the full load? If not, you have load sharing with a bonus, not redundancy.
- Are both units in the same thermal environment? If yes, redundancy does not cover the dominant wear-out failure.
If you cannot get answers to 1 and 3, do not parallel. Buy the bigger model, or split the load, or come and ask us to build the sharing version.
FAQ
Can I connect two identical 24 V drivers in parallel to double the current? Not unless they are designed for it. With ±5% output tolerance, the unit with the higher set point carries the entire load until it limits, then the pair hunts. You get a 24 V output with an unpredictable sharing ratio and a shortened life on the harder-working unit.
What if I add a diode to each output? That gives you redundancy against a shorted supply, not load sharing. Each unit still carries what it carries. And a Schottky costs you 0.4–0.6 V and 8 W of heat on a 400 W unit at 24 V.
Does paralleling work better at higher voltage, like 48 V? Slightly, because the tolerance band is a smaller percentage of a larger absolute voltage, so the imbalance current is lower. It does not fix the control-loop problem, and 48 V drivers in this power class are less common.
Can I parallel two drivers of different brands or wattages? No. Different fold-back shapes, different transient response and different output capacitance will produce oscillation at load steps. Even two units of the same model from different production batches can misbehave.
Is N+1 redundancy worth it for a signage installation? Only if a callout costs several hundred euro and access requires equipment. For a standard wall-mounted sign, a spare driver on the shelf and a maintenance contract is cheaper than a redundant pair, and it covers the same failure.
Need 800 W or 1,200 W in one enclosure instead of two units fighting each other? The 600W IP20 indoor power supply covers most single-unit cases, and our OEM LED driver manufacturing line builds droop-sharing and 1+1 redundant versions with the sharing accuracy stated on the test report. Send the load profile and the enclosure drawing.