PSU trouble with SLI / Crossfire article at extremetech

PSUs: The source of DC power for all components in the PC & often a big noise source.

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee, Devonavar

Post Reply
Tzupy
*Lifetime Patron*
Posts: 1561
Joined: Wed Jan 12, 2005 10:47 am
Location: Bucharest, Romania

PSU trouble with SLI / Crossfire article at extremetech

Post by Tzupy » Sun Mar 05, 2006 11:10 am

At www.extremetech.com there is an article about the trouble PSUs get into when powering high-end SLI or Crossfire setups. I think it's worth reading it. IIRC, the point is that the maximum power on the 12V lines is only reacheable if other lines are getting more load. So don't just get 2 Radeon X1900XTX, but also 4 Raptors 150GBs. Hmm, if only I had the money to do that...

MikeC
Site Admin
Posts: 12285
Joined: Sun Aug 11, 2002 3:26 pm
Location: Vancouver, BC, Canada
Contact:

Post by MikeC » Sun Mar 05, 2006 12:20 pm

I noticed this article a few days ago -- it's a good piece, helps you understand why people are having trouble with 600W PSUs.

Aris
Posts: 2299
Joined: Mon Dec 15, 2003 10:29 am
Location: Bellevue, Nebraska
Contact:

Post by Aris » Sun Mar 05, 2006 6:42 pm

another reason why the PSU industry needs to just go to a single 12v ouput design, along with the fact that it would increase effeciency. let the mobo manufacturers put dc/dc power conversion circitry in to create the other voltages.

Mr Evil
Posts: 566
Joined: Fri Jan 20, 2006 10:12 am
Location: UK
Contact:

Post by Mr Evil » Sun Mar 05, 2006 7:51 pm

Aris wrote:another reason why the PSU industry needs to just go to a single 12v ouput design, along with the fact that it would increase effeciency. let the mobo manufacturers put dc/dc power conversion circitry in to create the other voltages.
I was thinking about that the other day. Many components already have their own DC-DC converters and it wouldn't cost much to add them wherever else they are needed, but the benefits to PSU and wiring simplicity would be significant; maximum power ratings might even become meaningful. A change to decent connectors would be nice too.

I don't know if 12V would be the best choice though: with very high currents being drawn by some components, a higher voltage might be more efficient and allow cheaper wires.

TomZ
Posts: 386
Joined: Tue Feb 07, 2006 2:59 pm

Post by TomZ » Sun Mar 05, 2006 8:22 pm

Interesting ideas. One improvement on the idea of a single power supply rail would be to raise the voltage so that the current is decreased more. Running high current through wires, connectors, and on PCBs is a real problem. Other industries, e.g., telecom, use 48V a lot, and so there is a good deal of experience with DC/DC converters from that power supply range.

Anyway, it will probably never happen, but who knows?

Devonavar
SPCR Reviewer
Posts: 1850
Joined: Sun Sep 21, 2003 11:23 am
Location: Vancouver, BC, Canada

Post by Devonavar » Sun Mar 05, 2006 8:58 pm

There is already a large section of the computer market that uses this system: Most laptops run around ~17V. Now we just need the desktop market to move to a similar design...

SnooP
Posts: 106
Joined: Mon Nov 28, 2005 1:01 am

Post by SnooP » Sun Mar 05, 2006 11:31 pm

Good article, though theres either a mistake in method or explenation, in particular the mention of 5V minmum loads on zeus 650w and then right after their fix of using pcie-molex adapter will confuse some readers. The cross loading requirements has nothing at all to do with the problem of exceeding 18-20A on one of the 12V rails (virtual). In particular

"In other words, if you balance the loads on the different rails, then you won't have this kind of problem."

The cross loading requirements are not altered by moving the graphics card to a different 12V rail.

"As we noted earlier, you can prevent these shutdowns by using a molex-to-PCI Express graphics card adapter and sharing the load with another 12V rail. Alternatively, just load up the other rails with more hard drives or other power-drawing peripherals."

This advice will only work for the zeus 650w, as most supposed 'SLI' psu's are dual rail, so no way to power one of the graphics cards off 12V1 unless you mod the psu internals (and if your doing that you might as well bridge the 12V1/12V2 rails than using adapters anyways). Any other 3/4 rail psu will have the graphics cards on seperate rails (afaik). Also it is fairly obvious cross loading requirements and 12V 20A current limits are being confused.

autoboy
Posts: 1008
Joined: Fri Dec 10, 2004 8:10 pm
Location: San Jose, California

Post by autoboy » Mon Mar 06, 2006 8:59 am

So let me get this straight. The 12V1 rail is dedicated to the cpu for the most part. The 12V2 rail is for the rest of the system. The power supply can supply as much amperage on either rail up to the limit of both rails added together. So if i had a power supply capable of 32 amps on both rails and if i had a P-M that uses only 2 amps on the 12V1 rail then I have 30 amps on the 12V2 rail available which should be enough for 2 graphics cards running 20 amps and motherboard and harddrives for a total of 25 amps on the 12V2 rail leaving me with 5 amps headroom. However, because of the 20 amp maximum per rail the power supply shuts down. How would changing to the molex instead of the PCI-e help this and why would you want to load the 12V2 with more harddrives? Is my thinking wrong here?

MikeC
Site Admin
Posts: 12285
Joined: Sun Aug 11, 2002 3:26 pm
Location: Vancouver, BC, Canada
Contact:

Post by MikeC » Mon Mar 06, 2006 9:19 am

autoboy wrote:So let me get this straight. The 12V1 rail is dedicated to the cpu for the most part. The 12V2 rail is for the rest of the system. The power supply can supply as much amperage on either rail up to the limit of both rails added together. So if i had a power supply capable of 32 amps on both rails and if i had a P-M that uses only 2 amps on the 12V1 rail then I have 30 amps on the 12V2 rail available which should be enough for 2 graphics cards running 20 amps and motherboard and harddrives for a total of 25 amps on the 12V2 rail leaving me with 5 amps headroom. However, because of the 20 amp maximum per rail the power supply shuts down. How would changing to the molex instead of the PCI-e help this and why would you want to load the 12V2 with more harddrives? Is my thinking wrong here?
Some clarifications --
  • It is usually 12V2 that supplies the 2x12V (AUX12V) connector, which feeds the CPU.
  • A huge number of so-called multiple independent 12V line PSUs actually have only ONE 12V line. (rail =line)
  • Multiple 12V lines are not required for nVidia SLI certification of PSUs. AFAIK, the certification is provided after nVidia actually tests the PSU in specific SLI configs under vafrious loads.
  • The 240VA max limit stipulated in Intel's ATX12V v2.xx document has actually been defunct for at least 6 months, depite the absence of any change in the spec doc itself. Intel has verbally informed PSU makers that this is no longer required.
  • The most interesting portion of the extremetech article is the cross-loading behavior of SUs, ie, need to load both the 5V and 3.3V loads at least a bit to ensure very high 12V output performance.

autoboy
Posts: 1008
Joined: Fri Dec 10, 2004 8:10 pm
Location: San Jose, California

Post by autoboy » Mon Mar 06, 2006 9:34 am

So if it the 240VA is not valid anymore, how do we know which power supplies still use this? I just wanna know as much about this situation because i am trying to expel the more power is better myth in power supplies among my friends and this kind of performance just makes things worse. We all know it as the "Oh, my 650W power supply is shutting down...let me go buy the 850W and i will be fine" myth. All they really need to to go buy a seasonic 500W and they will be fine. I'm tired of having to explain my Super Silencer 300 running my gaming rig.

MikeC
Site Admin
Posts: 12285
Joined: Sun Aug 11, 2002 3:26 pm
Location: Vancouver, BC, Canada
Contact:

Post by MikeC » Mon Mar 06, 2006 10:16 am

Which PSUs don't have the 240VA limiter on the 12V line.... this is not something that is easily determined. Very tough w/o complex test gear.

What I can say is that the PSUs on Intel's ATX 12V v2.xx approved list marked with two asterisks -- ** -- don't have the 240VA limiter on the 12V line. They could have a limiter that's at 250VA; it's more likely that they have a limiter that's set to kick in at the total 12V combined VA limit.

From the Intel page: "** Power supply did not meet 240VA requirement during OCP test..."
Last edited by MikeC on Mon Mar 06, 2006 5:00 pm, edited 1 time in total.

Mr. Perfect
Posts: 74
Joined: Sat Nov 27, 2004 2:53 pm
Location: United States

Post by Mr. Perfect » Mon Mar 06, 2006 3:46 pm

So it's looking like this is caused by the "shared power plane" between 5v and 12v. I can't remeber hearing that term before, is this usually refered to as something else? Seperate voltage regulators for each voltage maybe?

TomZ
Posts: 386
Joined: Tue Feb 07, 2006 2:59 pm

Post by TomZ » Mon Mar 06, 2006 5:49 pm

Sounds like sub-par power supply design to me. Especially when you consider that some PSUs have the problem while others don't. This means that it is not necessary to design the PSU with this limitation.

Also, I thought it was interesting that Silverstone said they were going to "redesign" their PSU.

SnooP
Posts: 106
Joined: Mon Nov 28, 2005 1:01 am

Post by SnooP » Mon Mar 06, 2006 6:06 pm

Mr. Perfect wrote:So it's looking like this is caused by the "shared power plane" between 5v and 12v. I can't remeber hearing that term before, is this usually refered to as something else? Seperate voltage regulators for each voltage maybe?
This is what i was getting at, the 'shared plane' is the total 12V capacity. the 5V rail and cross loading requirements have nothing to do with the problem of the zeus 650w loading two pcie graphics cards off 12V3, causing the current limiter to be tripped. Moving one of the cards to 12V4 via molex splitter fixes the issue.
MikeC wrote:The most interesting portion of the extremetech article is the cross-loading behavior of SUs, ie, need to load both the 5V and 3.3V loads at least a bit to ensure very high 12V output performance.
Yes, though there are psu's commonly for sale with far worse cross loading requirements than the zeus. cross loading graph of zeus 650w:
http://www.xbitlabs.com/articles/other/ ... up_15.html
From that graph its fairly apparant that a 5V+3.3V of at least 25W or more to ensure psu runs stable @ very high 12V loads, preferrably at least 50w to get good (within say 3%) regulation.

Now look at the popular ocz modstream 520w:
http://www.xbitlabs.com/articles/other/ ... up_15.html
To load 12V up to capacity (28A), a 5V load of over 100W! is needed. Obviously this makes the psu useless for anything but fairly light single videocard systems.
autoboy wrote:. All they really need to to go buy a seasonic 500W and they will be fine.
Assuming the seasonic has 240VA limiters set for 240VA, the same problem will arise as zeus 650w (a single x1900xt draws around 120w stock, more when overclocked, perhaps 95% on 12V rail, probably more). If their set higher, well then it will work fine. I'd suspect the former, going off the xtremeoc review of the unit (sep 20, 2005) (12V1 and 12V2 are seperate on the pcb, so 240VA limiters are likely).

http://www.extremeoverclocking.com/revi ... 00W_5.html

MikeC
Site Admin
Posts: 12285
Joined: Sun Aug 11, 2002 3:26 pm
Location: Vancouver, BC, Canada
Contact:

Post by MikeC » Mon Mar 06, 2006 6:28 pm

SnooP wrote:Assuming the seasonic has 240VA limiters set for 240VA, the same problem will arise as zeus 650w (a single x1900xt draws around 120w stock, more when overclocked, perhaps 95% on 12V rail, probably more). If their set higher, well then it will work fine. I'd suspect the former, going off the xtremeoc review of the unit (sep 20, 2005) (12V1 and 12V2 are seperate on the pcb, so 240VA limiters are likely).

http://www.extremeoverclocking.com/revi ... 00W_5.html
The Seasonic does not have such limiters. (see the Inte list I mentioned above) It's also one of the PSUs that did NOT experience any SLI problems in extremetech's testing.

SnooP
Posts: 106
Joined: Mon Nov 28, 2005 1:01 am

Post by SnooP » Mon Mar 06, 2006 7:03 pm

Ahh yes, or as you said early the limiters are set to 250VA or maybe even 260VA (in which case once videocards are overclocked, psu 21A limit may trip. Otherwise would seem a bit strange having the two seperate 12V wirings on the pcb if theres no current limters, as the extreme o/c review has a picture of. Or maybe its Seasonic's way of trippping up 'younger players' :D, or that it used to have the 20A current limiters, now removed from pcb, but otherwise design unchanged. Anyways no big deal, if it works great!

ATWindsor
Posts: 285
Joined: Mon Apr 18, 2005 11:53 pm

Post by ATWindsor » Thu Mar 09, 2006 10:39 am

MikeC wrote:
autoboy wrote:So let me get this straight. The 12V1 rail is dedicated to the cpu for the most part. The 12V2 rail is for the rest of the system. The power supply can supply as much amperage on either rail up to the limit of both rails added together. So if i had a power supply capable of 32 amps on both rails and if i had a P-M that uses only 2 amps on the 12V1 rail then I have 30 amps on the 12V2 rail available which should be enough for 2 graphics cards running 20 amps and motherboard and harddrives for a total of 25 amps on the 12V2 rail leaving me with 5 amps headroom. However, because of the 20 amp maximum per rail the power supply shuts down. How would changing to the molex instead of the PCI-e help this and why would you want to load the 12V2 with more harddrives? Is my thinking wrong here?
Some clarifications --
  • It is usually 12V2 that supplies the 2x12V (AUX12V) connector, which feeds the CPU.
  • A huge number of so-called multiple independent 12V line PSUs actually have only ONE 12V line. (rail =line)
  • Multiple 12V lines are not required for nVidia SLI certification of PSUs. AFAIK, the certification is provided after nVidia actually tests the PSU in specific SLI configs under vafrious loads.
  • The 240VA max limit stipulated in Intel's ATX12V v2.xx document has actually been defunct for at least 6 months, depite the absence of any change in the spec doc itself. Intel has verbally informed PSU makers that this is no longer required.
  • The most interesting portion of the extremetech article is the cross-loading behavior of SUs, ie, need to load both the 5V and 3.3V loads at least a bit to ensure very high 12V output performance.
Ok, so lets say I have a filserver, and a PSU with 12v1 at 20 A, 12v2 at 20A and max total at 35A, you are saying that changes are good that if mye system (Mobo and CPU), use 5A, I can use a whole 30A for harddisks (eventhough neiter 12v1 og 12v2 is specified for that high current), becuase in reality its only one line?

AtW

MikeC
Site Admin
Posts: 12285
Joined: Sun Aug 11, 2002 3:26 pm
Location: Vancouver, BC, Canada
Contact:

Post by MikeC » Thu Mar 09, 2006 10:47 am

ATWindsor -- yes

TomZ
Posts: 386
Joined: Tue Feb 07, 2006 2:59 pm

Post by TomZ » Thu Mar 09, 2006 10:48 am

ATWindsor wrote:Ok, so lets say I have a filserver, and a PSU with 12v1 at 20 A, 12v2 at 20A and max total at 35A, you are saying that changes are good that if mye system (Mobo and CPU), use 5A, I can use a whole 30A for harddisks (eventhough neiter 12v1 og 12v2 is specified for that high current), becuase in reality its only one line?
Your question highlights the concern I have with Mike's advice that many PSUs have just one 12V rail instead of two. He said many, not all. You cannot make assumptions about the behavior of your particular PSU based on this generalization. Your PSU may or may not have separate rails.

The information Mike reported is interesting from a general perspective, but it cannot be used for a particular setup unless you know whether your PSU is designed this way or not. If someone gives you specific advice without knowing this detail, they are going out on a limb.

ATWindsor
Posts: 285
Joined: Mon Apr 18, 2005 11:53 pm

Post by ATWindsor » Thu Mar 09, 2006 10:50 am

MikeC wrote:ATWindsor -- yes
Ok, thanks for the info.

AtW

ATWindsor
Posts: 285
Joined: Mon Apr 18, 2005 11:53 pm

Post by ATWindsor » Thu Mar 09, 2006 10:51 am

TomZ wrote:
ATWindsor wrote:Ok, so lets say I have a filserver, and a PSU with 12v1 at 20 A, 12v2 at 20A and max total at 35A, you are saying that changes are good that if mye system (Mobo and CPU), use 5A, I can use a whole 30A for harddisks (eventhough neiter 12v1 og 12v2 is specified for that high current), becuase in reality its only one line?
Your question highlights the concern I have with Mike's advice that many PSUs have just one 12V rail instead of two. He said many, not all. You cannot make assumptions about the behavior of your particular PSU based on this generalization. Your PSU may or may not have separate rails.

The information Mike reported is interesting from a general perspective, but it cannot be used for a particular setup unless you know whether your PSU is designed this way or not. If someone gives you specific advice without knowing this detail, they are going out on a limb.
Hmm, interesting, I have to keep an eye on that, wehn building fileservers a gret deal of the draw will be from HDs, so a single rail would be much better suited for that use.

AtW

MikeC
Site Admin
Posts: 12285
Joined: Sun Aug 11, 2002 3:26 pm
Location: Vancouver, BC, Canada
Contact:

Post by MikeC » Thu Mar 09, 2006 11:07 am

ATWindsor wrote:
TomZ wrote:
ATWindsor wrote:Ok, so lets say I have a filserver, and a PSU with 12v1 at 20 A, 12v2 at 20A and max total at 35A, you are saying that changes are good that if mye system (Mobo and CPU), use 5A, I can use a whole 30A for harddisks (eventhough neiter 12v1 og 12v2 is specified for that high current), becuase in reality its only one line?
Your question highlights the concern I have with Mike's advice that many PSUs have just one 12V rail instead of two. He said many, not all. You cannot make assumptions about the behavior of your particular PSU based on this generalization. Your PSU may or may not have separate rails.

The information Mike reported is interesting from a general perspective, but it cannot be used for a particular setup unless you know whether your PSU is designed this way or not. If someone gives you specific advice without knowing this detail, they are going out on a limb.
Hmm, interesting, I have to keep an eye on that, wehn building fileservers a gret deal of the draw will be from HDs, so a single rail would be much better suited for that use.

AtW
TomZ is right. I am generalizing, but I will also say that there are more PSUs with a single 12V rail. PSUs with truly independent 12V lines are few and far between.

The other thing to look at is the total combined power output, rather than just the 12V combined. To use your example, even tho it may be rated for 35A on the 12V lines, the total max power might still be just 450W. If there is more than 30W drawn from the 5V and 3.3V lines, naturally the total power available on the 12V lines is less than 35A (420W).

sthayashi
*Lifetime Patron*
Posts: 3214
Joined: Wed Nov 12, 2003 10:06 am
Location: Pittsburgh, PA

Post by sthayashi » Thu Mar 09, 2006 9:47 pm

Ugh, reading that article made me cringe. A graphics subsystem drawing over 240W in SLI configuration has me wondering how they're getting rid of the heat. And that's when I shudder.

and worse, the "solution" to the power problems that people have been having is to ironically draw more power.

I'd try and sell power resistors to help draw more power passively, but if you're drawing that much power to begin with, you probably don't care about a few extra hard drives or whatnot.

Post Reply