Single 12V rail or multiple 12V rails? The eternal question answered

[hunter315]

Many people are confused by whether a single 12V rail or a multiple 12V rail power supply is better. The confusion is justified as there are myths about regarding both floating around on the internet as well as people who swear by multi rail and others who swear by single rail. There are claims that one way supplies cleaner or more stable voltage than the other, but this has a lot more to do with the platform and line filtering installed than the rail setup. The answer to the question of which is best a single rail unit or a multi rail unit is that it does not make a difference, so lets explore why.

In older versions of the ATX specs Intel had required the CPU to have its own rail separate from the motherboard and other components. It had also specified that no rail have more than 240 VA available(20A @ 12V), this is what lead to multi rail power supplies in the beginning. Both of those requirements have now been lifted allowing for power supplies with a single 12V rail and 12V rails rated for more than 20A which is very useful.

All power supplies convert the AC(alternating current) power that comes out of the wall into DC(direct current), and most units only have a single 12V DC source, in a single rail unit all of the yellow 12V wires are tied directly to this source, in a multi rail unit they are split into bunches(rails) and tied to the source, there is a controller chip monitoring the current going into each rail and if it exceeds a set limit it will shut down the power supply. Some smaller cheaper units claim to have multiple 12V rails to comply with the old ATX specs, but actually just tie all the bunches together with no over current protection so while the label makes it appear as a multi rail unit its actually a single rail unit, this helps cut costs.

One of the big myths about multi rail units is that because of the overcurrent protection you can end up with “trapped power”, XFX uses a power supply with a 200W 12V DC source as an example:
In a multi rail unit they show the 12V source split into two 100W rails, and when loaded up with 75W on each rail from the graphics card adding 50W for the CPU onto either of the rails would cause it to shut down even though it had 50W remaining for the source and a single 200W 12V rail would have been able to handle it all fine.

The lie in this scenario is that multi rail units have their over current protection(OCP) limits set such that you wont end up with trapped power, in this scenario the unit would likely have its OCP limit set at about 130W on each rail so that there would not be an issue of “trapped power”, this is why for most units with multiple 12V rails you cannot simply add the rails together to figure out the total power available from the 12V source, you must look at the total power listed for the 12V rails listed below them. On the units where the sum of the listed max currents for the 12V rails does add up to the stated limit of the 12V source the true current limits are usually set 3-5A(36-60W) higher than the label shows. With a modern power supply you are unlikely to overload any rail without using a large number of adapters and splitters to get more connectors, the manufacturer already spread the connectors across the rails so that it would be very unlikely for you to overload any single rail without trying pretty hard.

In short, its the total capacity of the 12V source that matters, how the rails are laid out does not affect you much.

 

[Proximon]

Good stuff!

Last year some may remember that Antec held a contest on newegg. In order to enter you had to read a little tutorial written by Antec and answer a question about it... one of the things they said in that tutorial is that very large power supplies with single rails were dangerous because of the potential amperage involved.

Really though I think as long as you are getting a quality PSU with proper protections and don't try to use it for a purpose it was not designed for, you should be good.

 

[tecmo34]

Hunter315... I added this information to the Power Supply Guides and UPS info master sticky, as I find the information useful that others should be able to access.

 

[hunter315]

Proximon, thats actually a good point to bring up, i spent yesterday reading up on the matter a bit more, and it true, a single rail unit has more potential to cause damage to something in the system in the event of a partial short that doesn't trip the short circuit protection on the unit.

There is a caveat to go with that though, any rail over above 40A in a unit with a single or multiple 12V rails, regardless of if it claims over current protection, is not going to trip fast enough to save the component in the partial short. OCP is similar to a circuit breaker in a house, the larger the breaker the longer it takes to trip, with the OCP set point set at 40A you are talking about almost half a second before it trips, anything at the end of that rail will be toasted by the time the protection kicks in, so units like the Antec High Current Pro 850 with its 4 12V rails with stated limits of 40A will cook a component just as well as a large single railed unit would in the event of a partial short.

The units with multiple large rails are no safer for the attached components than a unit with a single large rail is, but the units with multiple rails can rely on their OCP to trip eventually to shut them down, the single railed unit will cook it until it burns out to an open circuit, a short circuit that trips the SCP, or draws enough and trips the over power protections of the unit and shuts it down.

The units with their rails limited to about 30A or less will have it kick in fast enough that it has a chance to save the attached component, so many smaller rails can help keep things safer but the odds of getting a partial short and very low so it will not have an effect on 99.99% of users.

I learned quite a bit looking into that, thanks!


@Tecmo, thanks for the addition to the sticky!

 

[Proximon]

Hey great, I learned something too!

 

[jamie_1318]

Good read, I have read this before, but you have summed all the issues up quite nicely. Thanks to other people posting about safety issued because those should be addressed as well.

 

[ko888]

Some smaller cheaper units claim to have multiple 12V rails to comply with the old ATX specs, but actually just tie all the bunches together with no over current protection so while the label makes it appear as a multi rail unit its actually a single rail unit, this helps cut costs.

This doesn't just apply to cheaper units.

You just have to look at the label affixed to the Sea Sonic S12II Bronze Series 520W (SS-520GB) as an example of this. You may be misled into thinking that it has two 20 Amp overcurrent protected rails. The unit does not have two 12V rails at all - it's a single 12V design. There is no multirail overcurrent protection in there. Sea Sonic should fix their label to match reality.

 

[hunter315]

Interesting, i had never noticed that before, the newegg description lists it as a single 40A rail, but the label shows it as having 2 20A rails. The seasonic site lists the 12V combined current available so it shows like it is a single rail unit on their site, it would appear that only the label doesnt think its a single rail unit. Very interesting, good catch!

 

[delluser1]

Some smaller cheaper units claim to have multiple 12V rails to comply with the old ATX specs, but actually just tie all the bunches together with no over current protection so while the label makes it appear as a multi rail unit its actually a single rail unit, this helps cut costs.

This doesn't just apply to cheaper units.

You just have to look at the label affixed to the Sea Sonic S12II Bronze Series 520W (SS-520GB) as an example of this. You may be misled into thinking that it has two 20 Amp overcurrent protected rails. The unit does not have two 12V rails at all - it's a single 12V design. There is no multirail overcurrent protection in there. Sea Sonic should fix their label to match reality.

Thats something that's been going on for quite while.
6 years or so ago ,when Corsair introduced thier HX line-up to the world ( HX520/620 ), it was discovered that there weren't 3x 18a rails as the label stated, just a whole lotta power that could be drawn from any single "virtual" rail ( there were only 2 ).
It's part of the reason for the love affair with early, Seasonic built, Antec EA units ( 2x 17a rails, 27/30/34a combined, again no OCP on the virtual rails )
As long as the quality remains, it doesn't matter to me if the label lies a little, the units being more capable than what the label states isn't a bad thing.

 

[4DC_Resin_Smoker]

The current drawn from a P/S will be based upon it's load, so 1 rail or 2 rails it shouldnt really matter. Anything that would cause a 40 amp P/S to fry your MB is just as likely to fry your MB at 20 amps...

Past 5 amps you could literally use the P/S to perform welding, hence most P/S use multible rails to split the current making it "somewhat" safer for the end user.

 

[win189]

Ok so in the end are we saying its better to go with a single Rail or multiple or that it dose not matter

im only asking as i use liquid cooling with anti frease and a Home made pelter Chiller

my system runs at full cpu and gpu load while 60% overclocked and runs at 15C with chiller on 50% power load (controled manualy)

 

[hunter315]

I'm saying it does not matter, a well laid out PSU with multiple 12 V rails will distribute the connectors in such a way that you are unlikely to overload any individual rail unless you use adapters or intentionally attempt to load an individual rail heavily.

As long as the PSU you are using to power that system has sufficient capacity I would not worry much about how the power is distributed, especially since there are now PSUs that claim to have a single 12 V rail but really have multiple 12 V rails. The most ironic example of this is the XFX 1250W who is always FUDding about multi rail designs being bad in their marketing.
http://www.jonnyguru.com/modules.php?name=NDReviews&op=Story3&reid=273

As long as you have sufficient capacity you will be fine.

 

[hellfire24]

absolutely worth my time.
thanks a ton.

 

[InvalidError]

OCP is similar to a circuit breaker in a house, the larger the breaker the longer it takes to trip, with the OCP set point set at 40A you are talking about almost half a second before it trips

How fast OCP/OVP/UVP/etc. can respond to shorts/OV/UV is only limited by the current/voltage sensor's input and filter circuitry bandwidth. Since the sense circuit has to be filtered to avoid tripping on normal operating noise, the sense bandwidth is somewhat limited but should still be fast enough to respond within a millisecond. Once the PWM shuts down, you are looking at microseconds for output caps on shorted rail(s) to discharge.

Since all PSU/VRM designs need sufficient voltage feedback bandwidth to meet increasingly fast load changes, there is plenty of expertise out there that can be applied to fast OCP sense loops as well so I doubt OCP (at least on good PSUs) is much sloppier than the voltage regulator sense loops.

The real problem is that on a 40A rail (which may have a 50A actual trip threshold), it is quite possible that a "weak" short (say 20A) will simply not draw enough current to trip OCP at all until it causes a lot more damage and a harder short elsewhere. Imagine a HDD failure causing a 20A short to ground on a 60A rail previously loaded to 30A. The total 12V load is only 50A which is within the PSU's spec, the #16 (maybe #18 on cheap PSUs) gauge 12V wire is going to burn with the computer still running "normally" until more wires start shorting out from the overheating wire melting/burning off their insulation. By the time the OCP trips, you may end up having to replace a whole lot more than the HDD you would have had to replace had individual rails been limited to 20A.

Having #16 wires connected to a 40+A source is a potential fire hazard. That would be my greatest concern about high-current rails.

 

[nix327]

Thanks for all the info!

 

[Anas Bashar]

On the units where the sum of the listed max currents for the 12V rails does add up to the stated limit of the 12V source the true current limits are usually set 3-5A(36-60W) higher than the label shows.

actually that image above is what made me believe that a single rail is more flexible, because this graph implies that the current limits on an individual rail is very tight.

OCP is similar to a circuit breaker in a house, the larger the breaker the longer it takes to trip, with the OCP set point set at 40A you are talking about almost half a second before it trips.

half a second? well i do not know, am an electrical engineer, and half a second as response time is way too much, such a response time is used in 10 folds of magnitudes of power...
anyhow, the OCP in these PSs might be different or specialized to this specific application. can you recommend me a RELIABLE reading about these issues in PSs?

at the end, i really appreciate what you did in this sum up. thanks.

 

[hunter315]

I'm glad you stopped by to read this, a lot of this was information was gathered through a large number of reviews/dissections of PSUs as well as several different PSU forums. JonnyGuru has a lot of good PSU faqs, he used to work for one of the larger PSU companies, but now primarily does PSU reviews, Jonnyguru and hardwareSecrets do the best PSU reviews in my opinion, both sites are quite knowledgeable about how they work and what gives each one advantages and disadvantages.


Sooo strangely enough, i am an electrical engineer who spent a fair amount of time earlier this week looking at a variety of current limiting and circuit protecting devices, I was quite disappointed by how long it takes things to clamp unless you greatly exceed their rated current. Here is an example for you, this is a datasheet for a Square D 15A 115/240V circuit breaker(different form factor than your house ones but same principles)
http://www.alliedelec.com/images/products/datasheets/bm/SQUARE_D/70060602.pdf

On page 20 you will find a chart that tells you the time to trip based on multiples of rated current, if you want that breaker to trip in under a second you need to exceed 6x its rated current, that means you have to have 90A of current flowing for 1 second to trip that breaker, now this is throwing a much larger switch than the one in your PSU which runs on slightly different principles, but the concept of trip time versus multiples of current remains the same for almost all circuit protection devices because you don't want to trip everytime something draws a ridiculous amount of current for 100ms on start up. PSUs tend to have a more active trip method with the OCP chip watching the voltage across a finely tuned resistor, so their curve is much tighter than the circuit breaker, but you still are affected by multiples of current, and in the event of a short there is SCP which works on a different principle. OCP is unlikely to do anything but keep you from hurting the PSU, any near short likely won't cause your OCP to trip, its more likely to cook, finish the job, and either do a full short that trips SCP or blow open and stop the issue. PSUs are intricate beasts and now i have to go learn more about SCP because i realized i don't yet know how that works.... I shall return!

 

[Anas Bashar]

Here is an example for you, this is a datasheet for a Square D 15A 115/240V circuit breaker(different form factor than your house ones but same principles)
http://www.alliedelec.com/images/products/datasheets/bm/SQUARE_D/70060602.pdf

On page 20 you will find a chart that tells you the time to trip based on multiples of rated current, if you want that breaker to trip in under a second you need to exceed 6x its rated current, that means you have to have 90A of current flowing for 1 second to trip that breaker, now this is throwing a much larger switch than the one in your PSU which runs on slightly different principles, but the concept of trip time versus multiples of current remains the same for almost all circuit protection devices because you don't want to trip everytime something draws a ridiculous amount of current for 100ms on start up.

this reminds me about the characteristics of circuit breakers of which we deal with in power plants...
the thing is, a circuit breaker (or an OCP) is designed to accommodate the load needs, so when you do not have a load that briefly draws high current (like a motor stating up), there is no need to allow the load to draw a high current even for few milliseconds (like the OCP we use in control systems or communication cabinets), because such delicate loads were not designed to withstand a relatively high current, they will just get crispy.
furthermore, designing and manufacturing an OCP that has multiple trip times depending on the current value is much more complicated and expensive, compared to designing an OCP that trips immediately (so to speak).
what is the point of allowing a PC to draw a high current for few hundred of milliseconds, if no load inside the PC is supposed to do so?
i will start my own reading now, this matter intrigues my.

 

[Anas Bashar]

i found a datasheet about a PC PSU protection chip (https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=10&cad=rja&ved=0CHQQFjAJ&url=http%3A%2F%2Fmedia.digikey.com%2Fpdf%2FData%2520Sheets%2FFairchild%2520PDFs%2FFAN768x.pdf&ei=2bmAUoD4Mq2S7AbDgIGYCQ&usg=AFQjCNFhwo6G8kc2lhKau8pDItPPLFta6w&sig2=U9BUSqCNOdij1FRHk_UJZg&bvm=bv.56146854,d.ZGU).
i do not know to what standards this chip complies, issued 2004.
in page 11, it states "When an overcurrent condition appears at the OCP comparator input pins for more than 73us*, the FPO** output goes high and PGO*** goes low"
* this time is "Noise Debounce Time"
** FPO: Fault Protection Output, Open Drain Output (there is no transition time delay in this output signal)
*** PGO: Power Good Output (there is no transition time delay in this output signal)

from what i see here, the total delay time in OCP trip is 73us, unless there is something am missing here.

I've been reading the ATX12V v12.01 standard. (https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=3&cad=rja&ved=0CDwQFjAC&url=http%3A%2F%2Fwww.formfactors.org%2Fdeveloper%2Fspecs%2FATX12V%2520PSDG2.01.pdf&ei=JrWAUsXZHea57AaLq4GoBg&usg=AFQjCNGyaiGxTJEv9ctY5MBszk75q4-unQ&sig2=Zttdc7IGeMqLfFaD45W_sw&bvm=bv.56146854,d.ZGU)
it says nothing about OCP tripping time limits. the only obligatory protection is the OVP and OCP, any other protection is optional.

 

[NojjH]

So if I had 2 12v rails at 30A each, but a GPU required 42A, would that mean it'll balance it out itself? Or would that need to be done by myself?

 

[ko888]

So if I had 2 12v rails at 30A each, but a GPU required 42A, would that mean it'll balance it out itself? Or would that need to be done by myself?

You would have to specify the exact brand and model of PSU and graphics card to get an accurate answer.

Dual +12V rail PSUs are not all wired exactly the same way. The OCP set point on each +12V rail is not the same for all PSUs.

 

[NojjH]

So if I had 2 12v rails at 30A each, but a GPU required 42A, would that mean it'll balance it out itself? Or would that need to be done by myself?

You would have to specify the exact brand and model of PSU and graphics card to get an accurate answer.

Dual +12V rail PSUs are not all wired exactly the same way. The OCP set point on each +12V rail is not the same for all PSUs.

I'm looking at the XFX Radeon HD 7970 http://www.scan.co.uk/products/3gb-xfx-radeon-hd-7970-dd-edition-with-ghost-thermal-5500mhz-gddr5-28nm-gpu-925mhz-2048-cores-hdmi-m
and I have a 700w Storm Super Silent PSU: http://www.scan.co.uk/products/700w-storm-silent-lpk19-35-quiet-fan-eps-12v-atx-v22-psu

 

[ko888]

I'm looking at the XFX Radeon HD 7970 http://www.scan.co.uk/products/3gb-xfx-radeon-hd-7970-dd-edition-with-ghost-thermal-5500mhz-gddr5-28nm-gpu-925mhz-2048-cores-hdmi-m
and I have a 700w Storm Super Silent PSU: http://www.scan.co.uk/products/700w-storm-silent-lpk19-35-quiet-fan-eps-12v-atx-v22-psu

The Storm Silent LPK19-35 700W is generic trash.

A 700 Watt PSU that only has one 75W 6-pin PCI Express connector is a dead giveaway that the PSU is unable to deliver the power specified on its label.

If the 700 Watt PSU doesn't have a maximum combined +12 Volt continuous current rating of 58 Amps or greater has at least four 8-pin PCI Express supplementary power connectors then it should be avoided.

For a system using a single Radeon HD 7970 graphics card AMD specifies a minimum of a 500 Watt or greater system power supply. The power supply should also have a maximum combined +12 Volt continuous current rating of 38 Amps or greater and have at least one 6-pin and one 8-pin PCI Express supplementary power connectors.

There's no evidence that the Storm Silent LPK19-35 700W is able to do the job.

 

[NojjH]

I'm looking at the XFX Radeon HD 7970 http://www.scan.co.uk/products/3gb-xfx-radeon-hd-7970-dd-edition-with-ghost-thermal-5500mhz-gddr5-28nm-gpu-925mhz-2048-cores-hdmi-m
and I have a 700w Storm Super Silent PSU: http://www.scan.co.uk/products/700w-storm-silent-lpk19-35-quiet-fan-eps-12v-atx-v22-psu

The Storm Silent LPK19-35 700W is generic trash.

A 700 Watt PSU that only has one 75W 6-pin PCI Express connector is a dead giveaway that the PSU is unable to deliver the power specified on its label.

If the 700 Watt PSU doesn't have a maximum combined +12 Volt continuous current rating of 58 Amps or greater has at least four 8-pin PCI Express supplementary power connectors then it should be avoided.

For a system using a single Radeon HD 7970 graphics card AMD specifies a minimum of a 500 Watt or greater system power supply. The power supply should also have a maximum combined +12 Volt continuous current rating of 38 Amps or greater and have at least one 6-pin and one 8-pin PCI Express supplementary power connectors.

There's no evidence that the Storm Silent LPK19-35 700W is able to do the job.

So would you recommend me going for a cheaper GPU and buying a better PSU then?

 

[hunter315]

The most important part of this is that the GPU does NOT require 42A, the system may need close to that.

A 7970 has a power draw of ~250W(20.83A), either of your two rails could feed it the 20A by itself, but it is also important to note the power connectors and realize that the GPU can pull up to 75W of power from the PCI-e slot so not all 20A are coming from the rail that has the PCI-e connectors, it is already distributed! Generally you will find the CPU and a PCI-e or two on one rail, the motherboard, peripherals and other bits on the second with another couple of PCI-e connectors. On 3 and 4 rail units the PCI-e connectors are usually on their own on rails 3 and 4.

The storm silent does only have 1 6 pin PCI-e connector which is suspicious for a "700W" unit, my EA-500 had 2 PCI-e connectors....