Electrical – Voltage control relays, Should I use separate power supply circuits

This is becoming a quite long answer, but I added lots of pretty pictures, which should keep you from falling asleep ;-)

I'm aware of bistable relays, and they're the big savers, but here I'll discuss different solutions all for the same non-latching relay, in case you don't want to use a latching relay. That could be for feedback, or more complicated drive reasons, for instance. (One way to get feedback is by using one contact of a dual pole relay, but then you reduce it to a single pole relay. Three pole relays exist, but are expensive.)
Anyway, this is about your common, low-cost astable relay. I'll be using this relay for reference.

Series resistor
A cheap and simple way to reduce power, and applicable to most relays. Look out for must operate voltage in the datasheet, sometimes called "pull-in voltage". For the 12 V standard version of the above relay that's 8.4 V. That means the 12 V relay will also work if you apply minimum 8.4 V to it. The reason for this wide margin is that the 12 V for relays is often not regulated, and may vary, for instance with mains voltage tolerances. Check the margins on the 12 V before doing this.
Let's keep some margin and go for 9 V. The relay has a coil resistance of 360 Ω, then a 120 Ω series resistor will cause a 3 V drop, and 9 V remaining for the relay. Power dissipation is 300 mW instead of 400 mW, a 25% power saving, with just a series resistor.

In this and the other graphs the common solution's power is shown in blue, normalized for 12 V input, and our improved solution in purple. The x-axis shows the input voltage.

LDO regulator
With the series resistor the power savings are a constant 25 %, the ratio of our resistors. If the voltage rises the power will rise quadratically. But if we can keep the relay voltage constant, independant of our power supply voltage, power will only rise linearly with rising input voltage. We can do this by using a 9 V LDO to power the relay. Note that compared to the series resistor this saves more power at higher input voltages, but less if the input voltage drops below 12 V.
Power saving: 25 %.

Sensitive relay
This is the most simple way to drastically reduce power: use the sensitive version of the relay. Our relay is available in a standard version which needs 400 mW, and a sensitive version which is happy with half of that.
So why not always use sensitive relays? First, not all relays come in sensitive types, and when they do they often have restrictions, like no change-over (CO) contacts, or a limited switching current. They're more expensive as well. But if you can find one that fits your application I would certainly consider it.
Power saving: 50 %.

12 V relay at 5 V
Here we get to the Real Savings™. First we'll have to explain the 5 V operation. We already saw that we can operate the relay at 9 V, since the "must operate voltage" was 8.4 V. But 5 V is considerably lower than that, so it won't activate the relay. It appears, however, that the "must operate voltage" is only needed to activate the relay; once it's activated it will remain active even at much lower voltages. You can easily try this. Open the relay and place 5 V across the coil, and you'll see it doesn't activate. Now close the contact with the tip of a pencil and you'll see that it remains closed. Great.

There's one catch: how do we know this will work for our relay? It doesn't mention the 5 V anywhere. What we need is the relay's "hold voltage", which gives the minimum voltage to stay activated, and unfortunately that's often omitted in datasheets. So we'll have to use another parameter: "must release voltage". That's the maximum voltage at which the relay will guaranteed switch off. For our 12 V relay that's 0.6 V, which is really low. The "hold voltage" is usually only a bit higher, like 1.5 V or 2 V. In many cases the 5 V is worth the risk. Not if you want to run a 10k/year production of the device without consulting the relay's manufacturer; you may have a lot of returns. But for a hobby project with a one-time production you can see for yourself if it works.

So we only need the high voltage for a very short time, and then we can settle for the 5 V. This can easily be achieved with a parallel RC circuit in series with the relay. When the relay is switched on the capacitor is discharged and therefore short-circuits the parallel resistor, so that the full 12 V are across the coil and it can activate. The capacitor then gets charged and there will be a voltage drop across the resistor which reduces the current.

This is like in our first example, only then we went for a 9 V coil voltage, now we want 5 V. Calculator! 5 V across the coil's 360 Ω is 13.9 mA, then the resistor should be (12 V - 5 V)/13.9 mA = 500 Ω. Before we can find the value for the capacitor we have to consult the datasheet once more: maximum operate time is 10 ms maximum. That means the capacitor should charge slow enough to still have 8.4 V across the coil after 10 ms. This is what the coil's voltage over time should look like:

The R value for the RC time constant is the 500 Ω parallel to the coil's 360 Ω, due to Thévenin. That's 209 Ω. The graph's equation is

\$ V_{COIL} = 5 V + 7 V \cdot e^{\dfrac{-t}{RC}} \$

With \$V_{COIL}\$ = 8.4 V, \$t\$ = 10 ms and \$R\$ = 209 Ω we can solve for \$C\$ and we find 66 µF minimum. Let's take 100 µF.

So in steady state we have a 860 Ω resistance instead of 360 Ω. We're saving 58 %.

12 V relay at 5 V, reprise
The following solution gives us the same savings at 12 V, but with a voltage regulator we'll keep the voltage at 5 V, even if the input voltage would increase.

What happens when we close the switch? C1 gets quickly charged to 4.3 V via D1 and R1. At the same time C2 gets charged through R2. When the analog switch's threshold is reached the switch in IC1 will toggle, and C1's negative pole will be connected to +5 V, so that the positive pole goes to 9.3 V. That's enough for the relay to activate, and after C1 is discharged the relay is powered by the 5 V through D1.

So what our gain? We have 5 V / 360 Ω = 14 mA through the relay, and coming from a 12 V via an LM7805 or similar that's 167 mW instead of 400 mW.
Power saving: 58 %.

12 V relay at 5 V, reprise 2
We can do even better by using a SMPS to get our 5 V from our 12 V power supply. We'll use the same circuit with the analog switch, but we'll save a lot more. At a 90 % efficient SMPS we have an 80 %(!) power saving.

(graphs made with Mathematica)

The environment in a car is quite different from that experienced by electronic components in a box on a table in your office. There are several characteristics in particular that stress electronics components more in a car than a office environment:

1. Temperature. A car still has to start on a cold winter morning in Nome or after baking in the hot July sun until mid afternoon in Yuma. Not only do these extremes pose a challenge to operating ability, but the daily thermal cycling causes mechanical stresses.

2. Vibration.

3. Noisy power. "12V" car power is nowhere near as nice as what a 12 V regulated power supply provides. It can drop significantly during starting, goes to 14 V when the engine is running, and can have 10s of volts of nasty spikes on it. For example, the normal 7805 regulator can be damaged by car power, even though it is rated to 35 V. There is a special regulator series that is rated for car power, which of course costs more.

4. Humidity. The relative humidity is unpredictable and can be at one extreme or the other for extended periods of time.

In addition to this, the end users generally don't tolerate failures as well in cars as they do in ordinary office appliances.

I can't say how much of this is relevant to your situation since you haven't told us what you are really trying to do. However, keep in mind that there are millions of cars sold every year, so the volumes are high and there is considerable competition. If manufacturers didn't have to pay for the extra robust characteristics of the components, they wouldn't. The fact that a car company is willing to pay the extra $.50 for a relay that goes into a million cars, should tell you that it matters. That's $500k they could save if they didn't. Put another way, that's about 3-4 engineers for a year to come up with a cheaper alternative.