• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Dual Rail 12V and PSUs

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

AlabamaCajun

Member
Joined
Apr 22, 2005
Location
A Labyrinth
*Update, see notes at end and messages throughout this discussion.
Why do I care about Dual Rail and which PSUs are the best. I've seen a lot of questions concerning multi-rail PSUs, drooping and what's nest to buy. I decided to write this technical brief on PSUs based 2005 systems from years of knowledge in DC and AC electronics. My hope is that although more on the techinical side than most care about, you will be able to choose a great power supply for the Dream system and get better OC from it.

The magazines state look for the heavy PSUs and expect to pay a little more for a good one. This is true and I'll explain the inner reason in a minute. While you may see a lot of resemblance between different units they can vary significantly. There's only so many semiconductor, inductors and fans available and you may see the same parts on cheap PSUs as you see on expensive ones. That is why you have to research a little to find what you really need. If you are using old P3s, Slot or Socket A equipment then you can probably just get a decent 350Watter and run till the insulation falls of the wiring. However if you are running P4 prescotts, A64s and/or SLI then what you put in as a power source is going to mean the differece between Tinker and Overclocker.

The Nature of PSUs.
No all PSUs are not created equal! Naturally, most expensive PSUs uses more components, heavier transformers/coils and bigger heatsinks. Quality does also play in as well which will also drive up cost. Unlike mobos and slot cards, PSUs reguire more manual labor to place and solder components. It's not a task that robots can pick, place and wave solder to a board. Wires have to be run and large components have to be connected. This all drives up the cost we have to pay. Cheaper ones reduce the component sizes and filters circuits used to regulate the current arriving at you V-modds and overclocked CPUs. To do it right, it takes heavy transformers, chokes, capacitors and MOSFETs to deliver/

Transfomers, Coils and Chokes
These are the voltage converters and energy storage. Starting from house current at 120 or 240 volts, a isolation/stepdown transformer usually converts the voltage down a lower voltage from 25 to 50 volts. The I/S transformer also provides current isolation should something short. Without it, a direct short could send house current of 20 or more amps at 120volts blowing your chips all over the inside of the case. Most of the newer PSUs are replacing the primary transformer with a choke which provides the current isolation but eliminates the voltage step down. A Choke being a coil of wire on an iron or ferite core will only allow the amount of current defined by the wire gauge to flow through. Just like comparing a fire hose to your garden hose, which one do you want for putting out fires and which one to water your flowers with. Coils and chokes also store energy and lessen the effect of power surges and spikes as well as assist in voltage regulation. The larger the transform or coil the more energy it stores and this stored energy will determine if you OCed processor will run or crash when pushed. This is one of the differences I've seen in the better power supplies; they all have large coils and transformers in them. By far PC-Power&Cooling has such a largest transformers and coils in their units. So large that it looks like the airflow is blocked in reality, the air forced to pass around coils and closer to the heatsinks providing better cooling.

Phase Locked Loop (PLL)
These circuits, most are on chip, are the voltage watch dogs that monitor the voltage and bark louder if the voltage drops to increase current. This PLL is running as a voltage controlled pulse width modulator in PSUs. You would find one for each voltage supplied and each rail in multi-rail power supplies. What these low power but tight controlling voltage coupled oscillators do is respond to current demands in your PSU width differnt pulse widths. These PLLs are what drive the MOSFETs by switching on long enough to allow enough current through to maintain a predetermined voltage. When demand is low, the on times will be short. As current demand increases, the on-time will widden to allow more current through to keep up the voltage. This is where the term Phase Locked Loop comes from. The PLL is locked in step with current demand staying in tight control of voltage. The better quality supplies are very responsive to these fluctuations and will hold tight voltages.

MOSFETs
Theses are the providers of current. Their purpose is take the incomming voltage and provide for the current demand. This is where the term "switching power supply" comes from. The PLLs are sinaling the MOSFETS to turn on and off at the right intervals to supply the right amount of current to maintain tight voltages. Earlier I talked about transformers and chokes and how they transform voltage and current depending on circuit design. Due to the huge power demands it's important that these MOSFETs get the power they need to be able to deliver to the demands of ever more power hungry CPUs and GPUs. How MOSFETs are so effective in power regulation is that they use a low power signal from the PLLs and respond by switching large volumes of current at high speeds. You can search "Metal-Oxide Semiconductor/Field-Effect Transistor" for what these little beasts are and how they work. In our PSUs they act like a large dam holding back water with a valve that is easy to open and close, and it can be done quickly. The height of the dam represents the voltage the amount that can flow through the valve is the current. This height represents potential, the higher the level the greater the potential and the ability to supply for demands of current. This ability to respond to high current demands is why 12 Volts is used to provide power to your CPU. At 5 Volts the dam is not high enough to create enough pressure to respond when current demand increases and voltage suffers by drooping. The side effect to these higher voltages and increased current is heat generation. Good fans and headsinks are going to mean longer life and less of a possiblity of MOSFET failures. They are tough but subject to thermal breakdown at high temps.

Capacitors
Caps, short for capacitors store energy just like a battery but for shorter duration and faster cycles. With the MOSFETs being switched on and off by the PLL circuit, the power coming through is choppy. This is where the Caps come into play to filter or reduce this chop or ripple to a smooth even voltage. It's quite simple, they store energy during peaks and release it during valleys. The larger capacitors are the better they perform, skimping here will reduce the OC-ability of your system and stress components by allowing this ripple through.

Dual Rails
What a Dual Rail PSU provides is 2 seperate 12V power circuits. This really helps when dealing with A64 (X2 and FX ) and Prescott CPUs along with SLI video cards. Dual rail allows the CPU to run on one rail while the rest of the system uses the second rail. This isolates the heavy surge demands of the cpu from the load of drives and video cards. Drives do add to demand but the largest surge they create is during startup when the rest of the circuits are held asleep by the MOBO reset system. Video cards are starting to create heavier demands but they still don't come close to the power a cpu draws in bursts. These CPUs today can gobble up huge amounts of current flatline earlier PSUs with single rails and lower current. This current demand gets Ber with higher frontside busses, memory bandwidth and OCing the multiplier. This is due to the fact you are pushing circuits to switch at extemely high rates and this act of switching is causing these surges on the PSU that will affect any components in the same rail. A dual rail 450watt PSU should perform as good as a 500 to 550 watter since the it has the high demand seperated from the heavy load. The above average computer OCed is drawing in the 150 to 200 watts of continuos power it's the surges that are creating the need for hugher wattages. If you also note that the efficiency number is about 75% on good PSUs. This means that 125 watts of a 500 watter is lost to heat, filtering and regulation leaving you with 375 watts of power. But lets face it, the little used -12v line is wattage not available. The 3.3 and 5 volt rails are less used so thats close to half the total wattage not available to the important 12 volt supply. What you typically end up with is 225 to 350 watts on the 12volt rail(s). I've seen 50 watt swings on a P4C and 120watts running. With most of the demand on the 12volt rails, I might be looking at 100 to 125 watts to spare. This spare is important because of the swings the power supply needs more headroom to absorb the swings and keep the voltages inside tolerance. Think of dual rails as garden hoses and spigots. Hook up 2 pulse sprinklers with 2 hoses and one of those Y adaptors with a 3rd hose and trigger head on the end. Hook this to a single spigot and turn the water on with the trigger head closed and the sprinklers should have good pressure and covering distance accross the yard. Next take the trigger head and start washing the car and watch what happens to the sprinklers as you squeeze and release the trigger. Now move the 3rd hose to a seperate spigot and repeat the test. Now using the 3rd hose will have little or no effect on the sprinklers. You have a choice bigger spigots or more spigots.

Closing words
PSUs are getting better and good ones are becoming more mainstream driving down prices. If you re OCing PSU you will need a good to best quality PSU with good fans to improve it's ability to perform. Just like you would'nt wear a parka cutting grass in the summer, don't smother your PSU with wires or poor airflow.

*Updated Notes on this origional topic. Due to a inaccurate number from the intel 800 group of dual processors and what I also learned from other people weeks after this thread was posted a number of Chips will shut down some of the dual-rail PSUs. My original calulations were from Extreme and 800 dual Intel CPUS was far below the actual numbers some OCer are seeing. My origional number for Intel Prescotts, was about 150 watts, but from later findings this number exceeded 200 watts and would flatline the 16 to 18 amp rails on most PSUs. This may also apply to other Prescott and newer CPUs that are being pushed up to 4 and 5GHZ, units on Phase Change cooling and all Duals. This will also mean Xeons, Itaniums and Optirons that hit these high wattages are included in the "Get A Bigger PSU" crowd. For these chips you will need more than 20 amps on the rail that the CPU is on. This will mean a high wattage PSU whether it is single or dual rail starting at 550 or better. Unfortunately it looks like CPU power is exceeding what the PSU makers are putting on the market faster than they can update the power throughput.
I didn't mention AMD Athlon CPUs for the fact most are well under 150 watts AMD did an excellent job of producing a fast and efficient CPU. If you manage to get more than 175watts through your Athlon 64/X2 then you will also have to do a little more looking to find a beefier PSU.

Based on knowledge and reading by Alabama Cajun June, 2005.
 
Last edited:
AlabamaCajun said:
MOSFETs
Theses are the providers of current. Their purpose is take the incomming voltage and provide for the current demand. This is where the term "switching power supply" comes from.

hehe i still like to think that the switch on the back makes it a "switching" ps lol very good read thanks
 
excellent work.

i actually have experienced i very simmilar plumbing situation this summer at my house:

"Think of dual rails as garden hoses and spigots. Hook up 2 pulse sprinklers with 2 hoses and one of those Y adaptors with a 3rd hose and trigger head on the end. Hook this to a single spigot and turn the water on with the trigger head closed and the sprinklers should have good pressure and covering distance accross the yard. Next take the trigger head and start washing the car and watch what happens to the sprinklers as you squeeze and release the trigger. Now move the 3rd hose to a seperate spigot and repeat the test. Now using the 3rd hose will have little or no effect on the sprinklers. You have a choice bigger spigots or more spigots."

this analogy i instantly identifyed with, due tohaving nearly the exact situation in watering the lawn day-in and day-out (reminds myself that the plastic Y needs to be replaced)

however:

with 2 sprinklers and a hose running, all 3 will have reduced pressure, because of the limits of the household water main.
if if sister turns on the washing machine, and mom gets in the shower, while you attempt to wash your hands- not only do you have chaos, but the lawn dosnt get watered. the sprinklers stop pulsing pathetically, and "whiz" a mudpuddle into your lawn.

the main just cant keep up.

in much the same way, you are limited by the total DC output of the PSU.
DUH -
but wait, if the total output is your limit, then why does it matter if it is split up or not?
you either have enough power, or you dont.

deviding the output only works if the pressure is indreased proportionally - in sprinklerspeak.
in other words, why have 2 rails? no matter how you slice it you cant draw more power than you produce?
 
Last edited:
Thanks to all for the praise!
orionlion82 said:
excellent work.

i actually have experienced i very simmilar plumbing situation this summer at my house:

"Think of dual rails as garden hoses and spigots. ~clip from lead article~"

this analogy i instantly identifyed with, due tohaving nearly the exact situation in watering the lawn day-in and day-out (reminds myself that the plastic Y needs to be replaced)

however:

with 2 sprinklers and a hose running, all 3 will have reduced pressure, because of the limits of the household water main.
if if sister turns on the washing machine, and mom gets in the shower, while you attempt to wash your hands- not only do you have chaos, but the lawn dosnt get watered. the sprinklers stop pulsing pathetically, and "whiz" a mudpuddle into your lawn.

the main just cant keep up.

in much the same way, you are limited by the total DC output of the PSU.
DUH -
but wait, if the total output is your limit, then why does it matter if it is split up or not?
you either have enough power, or you dont.

deviding the output only works if the pressure is indreased proportionally - in sprinklerspeak.
in other words, why have 2 rails? no matter how you slice it you cant draw more power than you produce?

Yes, that is true, the sum of output is limited by the available supply. You might need to upgrade your street connection to the main from 3/4" to 1-1/4". Much like the chokes and transformers that limit the current to the regulator circuits, the street connection limits water volume and thus the number of taps on the spigot. Think of the sprinlers running on 12v, shower and the dish washer as 5V and the icemaker/fridgewater as 3.3V. This total usage would equate to the total PSU rating.
In the dual power supply rail situation, you still need to consider total usage and the total 12v supply. They both have to exceed the needs of your system to allow for surges from heavy use while maintaining the voltages at the correct level. Assuming the chokes and trasformers are supplying enough "head" wattage to the regulator circuits then it's up to the regs to do the rest.
While going to dual rails dose not increase the total wattage, it does seperate the noisy processor from the constant load by the drives and other 12V taps. The regulators have the toughest jobs of converting 50 to 120 volts down to 12,5 or 3.3 Volts. By using 2 independent circuits the load is split and the regulators are can keep tighter regulation on both 12 volt rails. What we have to watch is that PSU makers don't split the rails after regulation before filtering. This would once again create a bottle neck at the regulator, reduce the benefits and would no longer be a true dual rail system.
 
Last edited:
Have you been inspecting my pipes? :bday:

Now im really impressed! and i have a pretty clear understanding of the need for separation now as well. as a folder it diddnt quite click, because i allways run maxed out (though by no means allways at "peak" ) but a normal system bouncing around as it does - that really makes sense for decreasing stress on components- you just weaseled out "excellent" rating from me for this.
 
orionlion82 said:
Have you been inspecting my pipes? :bday:

Now im really impressed! and i have a pretty clear understanding of the need for separation now as well. as a folder it diddnt quite click, because i allways run maxed out (though by no means allways at "peak" ) but a normal system bouncing around as it does - that really makes sense for decreasing stress on components- you just weaseled out "excellent" rating from me for this.
Thanks, an happy birthday!

"No that must have been my brother, digging up your yard looking for the main he-he":D
 
While going to dual rails dose not increase the total wattage, it does seperate the noisy processor from the constant load by the drives and other 12V taps. The regulators have the toughest jobs of converting 50 to 120 volts down to 12,5 or 3.3 Volts. By using 2 independent circuits the load is split and the regulators are can keep tighter regulation on both 12 volt rails. What we have to watch is that PSU makers don't split the rails after regulation before filtering. This would once again create a bottle neck at the regulator, reduce the benefits and would no longer be a true dual rail system.
I once had a computer technician ask me about why hard drives were failing in a dual Britney server, particularly in the morning when the users are going online and putting a lot of load on the server. The drives will often log something about bad power before failing. There is a UPS and the PSU is relatively high quality (he told me it was a Fortron), so I asked him if he had an oscilloscope to check the 12v power. The problem was exactly what I suspected: the Britney CPUs transitioning from idle to full load and back put spikes on the 12v that could damage other circuits. The voltage will dip as low as 11v and spike as high as 13v when the load changes. I told him to put a noise filter on that line to the disk drives and so far, there have been no problems.
I'm not sure how much power a Britney CPU uses at full load (or even exactly what a Britney CPU is), but the best answer I've heard so far was, "About as much power as a small floodlight." (150w?)

I have seen a PSU that has a noise filter (seems to be just a large ferrite core) on the lines going to the drives. It's not as good as one that uses a noise filter made of a multiturn inductor and some low ESR/low ESL capacitors, but it's a step in the right direction.
Now we have three choices: a regular single 12v rail PSU, a single 12v rail PSU with a filtered 12v rail, and a true dual 12v rail PSU. While the filtered 12v rail probably won't improve much if the filter is just a ferrite core, it does do something. But a dual 12v rail PSU will offer a lot in terms of regulation. Of course, then it's back to the splitting problem again.
 
Dell_Axim said:
I once had a computer technician ask me about why hard drives were failing in a dual Britney server, particularly in the morning when the users are going online and putting a lot of load on the server. The drives will often log something about bad power before failing. There is a UPS and the PSU is relatively high quality (he told me it was a Fortron), so I asked him if he had an oscilloscope to check the 12v power. The problem was exactly what I suspected: the Britney CPUs transitioning from idle to full load and back put spikes on the 12v that could damage other circuits. The voltage will dip as low as 11v and spike as high as 13v when the load changes. I told him to put a noise filter on that line to the disk drives and so far, there have been no problems.
I'm not sure how much power a Britney CPU uses at full load (or even exactly what a Britney CPU is), but the best answer I've heard so far was, "About as much power as a small floodlight." (150w?)

I have seen a PSU that has a noise filter (seems to be just a large ferrite core) on the lines going to the drives. It's not as good as one that uses a noise filter made of a multiturn inductor and some low ESR/low ESL capacitors, but it's a step in the right direction.
Now we have three choices: a regular single 12v rail PSU, a single 12v rail PSU with a filtered 12v rail, and a true dual 12v rail PSU. While the filtered 12v rail probably won't improve much if the filter is just a ferrite core, it does do something. But a dual 12v rail PSU will offer a lot in terms of regulation. Of course, then it's back to the splitting problem again.

Thanks for sharing that experiance, servers are the most brutal on power supplies when it comes to cutting edge mutiprocs and large drive arrays. With this you can see why I chose to put the induction/reactance group first in the artical. You're absolutely right about using the right combination of filter, not just the cheap and easy parts. It should be an ATX/BTX requirement that the 12V feeds be seperated into groups and each group pass through filters on the way to our equipment. This would improve the stability and protect other equipment. Soon we will start to see the same effect with video cards. With SLI the wattage requirment is a lot higher than what is really required to run because of the demands needing the extra headroom. I think we will soon see 3 rails as GPUs are approaching CPU demands.
 
I just ordered a dual rail Enermax for my borthers new rig. Hey, at least now I know why. Thanks for education. I can always use it.
 
Gahhh i smell britney cpu's:

Dell_axim said:
I'm not sure how much power a Britney CPU uses at full load (or even exactly what a Britney CPU is), but the best answer I've heard so far was, "About as much power as a small floodlight." (150w?)

It's a hoax.
 
Nice

Excellent post, definitely worthy of the "glue stick" It takes a lot to hold my attention in a long thread...but this was worth it.
I actually came in here by mistake,(glad I did) was searching for any mods or upgrades I could make to my Antec PSU.
Thanks,
 
Bhugarov said:
Excellent post, definitely worthy of the "glue stick" It takes a lot to hold my attention in a long thread...but this was worth it.
I actually came in here by mistake,(glad I did) was searching for any mods or upgrades I could make to my Antec PSU.
Thanks,
Thanks, now if we can get the mods to find the glue stick. I thought this was really important when I decided to write it. I've made my last PSU purchase and will make my next PSU purchase based on this info. As I mentioned earlier in the post, I've been in and around electronics since the 70s before we knew what ICs were much less pump 35 amps into a chip the size of a finger nail. I've since learned that PC-P&C now has an SLI-PSU with 4 rails and it's not cheap. I'm not sure how the rails are designed but apparently they are seeing the light in PSU design. One point that I have to say is that as you get higher and higher amperages it gets harder to regulate to the level needed for these chips to run. Pushing 400 watts though one rail is a lot harder to control then 2 - 200 watt rails. I forsee 3 and 4 rail PSUs becoming mainstream by this time next year. I think these X2s along with SLI leave little choice if we want high stable OCs with these hungry beasts. -AC-
 
The problem is this, while well intentioned, leads people to the wrong supply. Current dual rail supplies don't focus as much power on the cpu line as can be consumed. Heavily OC'ed Prescotts and anything dual cpu or dual core can demand more 12V current than the cpu side of the largest dual rail units can povide. At present the cure is to use one of the large single rail units, namely the OCZ Powerstream 520, Sparkle FSP550-60PLG, or the PC Power&Cooling 510.

I believe that multiple rails are in concept superior, but until more powerful multi-rail units are developed their advantages are somewhat secondary. The AX500-A Fortron is a fine dual-rail unit, and serves well for about 90% of the machines here, but those with more extreme demands really have to utilize a single rail supply at this juncture to satisfy them.
 
To add to Larva's post, there is a 600w monster Zippy out there with a good dual 12v rail topology (20A+26A), but the problem is that unit is currently selling for big $$$. And it's a server unit, not really meant to be quiet. It is when units like this become more affordable that I'll begin thinking of them as viable solutions for really high draw 12v systems (ie SLI). Maybe Fortron will start turning them out now that they've expanded their operations ;)
 
And although the Zippy is the strongest of the dual-line units, if the cpu line is the 20A one it likely still isn't big enough. The cpu line on the OCZ PS600 is rated at 20A also, but has been overloaded by several members here. As far as I know all of them fixed the situation by going back to the single-rail PS520. A 24A cpu line and an 18A secondary 12V rail is what is really needed to allow the benefits of this topology to come to fruition.

*edit* Looking at the documentation on Zippy's website, it isn't totally clear which 12V rail is which. I think the 26A one is for the cpu, but I'm not positive. In any event the current over-protection kicks in at the same output level for either of the two 12V lines, and the unit is rated at a 12V total of 40A.
 
Last edited:
Back