When to Stop Using a 7805 — Linear vs Switching, and the Line Nobody Draws

The answer, in one sentence: use a linear regulator when the wasted watts fit the package, use a switching regulator when they don't — and the breakpoint is usually somewhere around one to three watts of dissipation.

This is not a "linear is dead" post. In the boards we support (2025–2026), the 78xx family still ships in huge volumes.

Half the applications genuinely don't need a switcher — and the other half are burning a board and a heatsink trying to make a linear part do a switcher's job.

The line is drawable. Here's how.

Linear Efficiency Is a Voltage Ratio, Not a Design Decision

A linear regulator's efficiency is set by physics: V_out / V_in. A 7805-class part dropping 12V to 5V is 41.7% efficient, no matter what the datasheet promises.

The other 58.3% becomes heat, and heat is the only thing you get to design.

The same 12V-to-5V job with a buck converter runs 85–92% — the gap is the whole argument, and it grows with the input voltage. From 24V, the linear part is 20.8% efficient and burns four times the useful power as heat. From 48V, nobody sane tries.

The forum shorthand is blunt: "a 7805 from 12V at 500mA is a 3.5W heater with a free 5V output." It's harsh, and it's exactly right when the heat has nowhere to go.

Read the heat column. At 12V and above, the linear part burns more than the buck delivers — the heatsink problem is the efficiency problem with a different name.

The Package Is the Decision — 350mW, 10W, and the Copper in Between

Here's the part the efficiency math hides: a linear regulator's real spec is its dissipation budget, and that budget lives in the package.

The SOT-89 78L05G we stock is rated 350mW — from 12V in, that's about 50mA of load before the junction climbs toward 150°C.

The TO-252 78D05AL is rated 10W with a proper copper pour — from 12V in, roughly 700mA. Same family, twenty-eight times the thermal budget, and the difference is metal.

On the forums, the classic DPAK failure is instructive: a part dropping 7V at 350mA — about 2.5W — reached 100°C on a board with a poorly soldered tab. The tab is the heatsink. If the solder is wrong, the copper doesn't matter.

And if the copper doesn't matter, the linear part is a fuse with a 5V pin.

The rough rule we use at distribution: under 1W of dissipation, linear wins on simplicity; 1–3W is the gray zone where copper and airflow decide; above 3W, stop negotiating and switch.

The Case for Linear: Noise, Cost, and the Missing Inductor

None of this makes the switcher the default. A linear regulator's case is short and real:

• Clean output. The 78xx family delivers ~80dB PSRR at 120Hz and microvolt-class noise — no switching ripple, no EMI, no layout that has to be careful about the inductor.
• Cost. A 78L05G runs pennies at volume; the switching equivalent's BOM is a controller, a MOSFET, an inductor, a diode, and a feedback loop — dollars, not pennies.
• Simplicity. Two caps and a ground plane. No loop compensation, no slope compensation, no transformer design, no EMI filter that eats the efficiency you bought.
• Reliability. The failure modes are thermal and documented; a switching supply's failure modes include layout mistakes that only show up in production.

The 78xx family survives because half the applications live in this column — bias rails, references, small MCU supplies, analog sections that can't tolerate ripple.

The Case for Switching: Watts, Heat, and the 48% Clamp

The switching column has one argument, repeated: the watts don't become heat.

From 12V at 1A, a buck delivers the same 5V with roughly 7W less heat than the linear part — that's the difference between a passive board and a board with a fan, or no heatsink at all.

Off-line, the argument is existential: there is no linear regulator for rectified 230V. The current-mode controller family — the UC2844G we stock is a good example — exists because the switch, the transformer, and the control loop are the only way to convert mains to a useful rail.

The 48% duty clamp that defines the UC2844 is a telling detail: switching designs accumulate constraints (transformer saturation, current-mode instability, EMI) that linear designs never meet — and each one is a thing that can be done wrong.

The Decision Framework: Five Questions

Here's the framework we actually use when a customer asks "linear or switching?" — it's five questions, in order:

1. What's the input-to-output headroom? Below ~2V, it's an LDO conversation, not a 7805 conversation. Above ~15V, the efficiency math starts screaming for a switcher.
2. How much current? Under 100mA, a linear part is almost always right. Over 1A, a switcher is almost always right. Between them lives the gray zone.
3. How many watts become heat? Compute (V_in − V_out) × I. Under 1W, linear. Over 3W, switch. In between, ask the next question.
4. Can the board shed that heat? TO-252 on a 2oz pour with vias can take a surprising amount; a bare SOT-89 cannot. If the answer is "no", the decision is made.
5. Does the load care about noise? Analog front ends, references, and audio hate ripple — a linear stage after the buck (or instead of it) earns its pennies there.

That's the whole framework. It's not about brands or datasheets — it's about watts and where they go.

When the noise question lands on "yes", the answer is often both: a buck for the watts, then a linear stage for the quiet rail — the 78D05AL or 78L05G doing what it does best after the switcher did the heavy lifting.

The hybrid row is the one most designers forget. The buck does the watts, the linear stage does the quiet — the 78xx part earns its pennies doing what it's best at, after the switcher did the work it can't.

What We Actually See in Distribution

If there's one pattern in the boards that come back to us, it's this: the linear part wasn't the mistake — the load on it was.

A 78D05A family part asked for 800mA from a 24V rail is a 15W heater; the part fails, gets replaced, fails again, and the customer blames the regulator before anyone computes the watts.

The second pattern is the reverse: a buck added where a bias rail would do — an inductor, a feedback divider, and an EMI filter for a rail that draws 20mA. The switcher wasn't wrong, it was just expensive.

Both patterns are the same mistake: the decision was made by habit, not by the five questions.

Frequently Asked Questions

Q1: Is a linear regulator ever more efficient than a switching regulator?

A: Yes — when the headroom is tiny. From 5.5V to 5V, a linear part is 91% efficient and a buck's switching losses make the comparison close. That's why low-dropout parts exist; the efficiency argument flips below ~1V of headroom.

Q2: At what current should I stop using a 7805?

A: When the watts beat the package, not at a fixed current. At 12V in, 500mA means 3.5W — too much for a bare TO-252. At 5.5V in, the same 500mA is 250mW — fine. Compute (V_in − V_out) × I and compare it to the package budget.

Q3: What's the efficiency of a 7805 from 12V?

A: 41.7% — it's 5V divided by 12V, and that's the best case. The rest becomes heat. A buck converter doing the same job runs 85–92%. The gap is the entire linear-vs-switching argument.

Q4: Why do people still use 7805s then?

A: Noise, cost, simplicity, and honest load sizes. 80dB PSRR, pennies per part, two caps, no inductor, no EMI design. For bias rails and small MCU supplies the heat is trivial — and the forums' "oversized regulator" warning applies to the 7805 on a 20mA rail just as much as to the switcher that replaced it.

Q5: Can I add a linear regulator after a buck for cleaner output?

A: Yes — that's a standard and often the best architecture. The buck takes the watts, the linear stage takes the ripple. The input headroom for the linear stage is small, so its heat is small, and the output is quiet enough for analog.

Q6: When does the switching regulator's EMI outweigh its efficiency?

A: When the load is tiny and the layout is tight. At 100mA from 12V, the linear part burns 700mW — acceptable on most boards — and the switcher adds ripple, an inductor, and an EMI story. Below ~1W of dissipation, linear usually wins the total cost of design.

Q7: What input voltage kills the linear regulator idea?

A: Anything above ~15V of headroom at real current. From 24V to 5V at 500mA is 9.5W — a heatsink conversation. From 48V or rectified mains, linear is physically impossible at useful power; the UCx84x current-mode family or an integrated switcher is the only path.

Q8: Where do I buy genuine linear and switching parts?

A: ICMASS stocks both sides — the UTC 78xx family (78L05G, 78D05AL) for linear rails and the UC2844G current-mode controller for switching designs. Volume pricing typically runs $0.03–$0.18/unit depending on part and quantity. Contact us for a quote and same-day dispatch from Shenzhen.