Power issues on high-end SLI builds

G

grizzed

Limp Gawd
Joined
Jan 24, 2006
Messages
171
despite the 600-watt PSUs in each system, the power demands of all the parts, most particularly the pair of overclocked 512MB GeForce 7800 GTX 3D cards in each, is too much for the system during the most demanding gaming scenarios.
Click to expand...
We clocked both systems down to their stock speeds and experienced the same errors.
Click to expand...
Nvidia SLI Certified PSU fails

The systems tested were OEMs, but this could affect anyone. How much power are these systems going to need?
 
Well, before buying my PC Power & Cooling 850 SSI, I consulted with the company on either getting a 510 SLI or 850 SSI. I mentioned that I would enventually upgrade to a dual core with maybe a couple more HDD's and the rep said go with the 850 SSI for future proofing my system, so that is what I did. A lot of money, but well worth the investment for the next 3-5 years :D
 
I have a 510SLI with no power issues and I have every power connector in use. 1 Raptor, 3 PATA HDs, DVD-rom, DVD-R, rest of system spec's in my sig.
 
Both PSU indicated in the artice has multi-rails (4 for the Silverstone and 2 for the Enermax). I believe it's a matter of picking the wrong rail for putting the cards or it's probably a bad design causing issues.
 
510 SLI powering the following with no power-related issues:

fx-55
4 36GB raptors
2 74GB raptors
1 DVD-RW
1 CD-RW
1 Floppy
2 BFG 7800GTX OC
1 SB X-FI
1 Netgear wireless NIC
2 1024MB sticks of RAM
4 120mm fans
 
In a nutshell...
SLI "Certifcation" is bullshit.
Let me say that again for people hard of hearing. SLI "Certifcation" is bullshit.
You give nVidia money, they give you a sticker. That's it. None of those PSUs can support that kind of configuration. The PCP&C 510SLI can easily, yes. The rest? I am utterly not surprised they failed miserably. I'm also unsurprised by the complete bullshit from the vendors - so typical when you fail to test before ship. Especially since instead of adding drives, you can use $3 of resistors to achieve the same results.

On a related note, how the hell does FalconNW get off charging $6500 for a system that can't even hold a candle to one I charge $5000 for? Hell, I include custom paint, a proper soundcard, 4GB, and a PCP&C. :rolleyes:
 
Xilikon said:
Both PSU indicated in the artice has multi-rails (4 for the Silverstone and 2 for the Enermax). I believe it's a matter of picking the wrong rail for putting the cards or it's probably a bad design causing issues.
Click to expand...

dont get much choice
auxillary power to the cards are going to be on:

+12V1 in an ATX12V v1.3 hybred (since there is a single rail)
+12V2 in an ATX12V v2.0
+12V3 in a common plane EPS12V (3 x +12V)
+12V4 in a split plane EPS12V (4 x +12V)

of course neither the SilverStone Strider ST60F or the EnerMax 701 would be my first choice off the list
and nVidia has never guranteed that those supplies will run their SLi config plus an unlimited amount of other hardware, it does seem reasonable to expect however that they would run the configs sold, since somebody obviously spec'd them and sold em to the public :p

but dont make the mistake the author has and think in "Watts"
it aint the watts its the amps and where they are

as already mentioned a PCP&C 510 would kick both their asses
its far tighter regulated (1%), much cleaner (10mV p-p on all the mains), and since its rated at 50C delivers likely 600 to 650W compared to those supplies

and with 36A +12V that works out to 432W @ 50C regulated to 1% load, and thats continious
with a peak of 38A x 12V or 456W @ 50C (peak is a milisecond spec which unfortunately often gets applied to supplies instead of a continious rating)

a PCP&C 510 SLi or ASL would have no problem powering either of those configs
it also happens to be a single +12V rail, not a problem in a supply that tightly regulated

http://www.velocitymicro.com/wizard.php?iid=32
http://www.falcon-nw.com/config/build.asp
 
Athlon 64 FX-60 110W or 9.2A +12V
7800 GTX 100W per but lets up that to 125W per or 20A +12V
HDDs and opticals at 2A per for spinup and 0.5A once spinning lets say 2A
mobo 0.5A, some fans 2A (again thats alot of fans once they are spinning)

lets see I come up with 33.5A total or 402W on the +12V rail

if those supplies are having troubles Id suggest its because they where rated at too low a temperature,
or they have a load regulation issue when that highly loaded
or they dont have the amps where they need them,
or a combination of all the above
 
Ice Czar said:
if those supplies are having troubles Id suggest its because they where rated at too low a temperature,
or they have a load regulation issue when that highly loaded
or they dont have the amps where they need them,
or a combination of all the above
Click to expand...

Indeed - I've been recommending mostly known strong EPS single 12v designs for such rigs. Usually the OCZ Powerstream 520W, PC P&C 510W, Zippy HP2-6500P/PSL-6701P, Silverstone ST56ZF, and Sparkle FSP550PLG-SLI. That Silverstone is already getting fairly popular.
 
(Thanks to Xilikon for directing me to this thread.)

I was surprised that even a quality brand, such as Enermax, failed to power dual 7800GTX SLI video cards. :eek: But I guess any good 500-600 W PSU should have no problem running a single high-end video card, even if it's overclocked.

I have a 600 W Seasonic PSU; I wonder how it would've performed in either of those 2 PCs. :confused:
 
Seasonic S-12's exceed spec in a number of ways and are rumored to be rated at 40C
they are just fine especially with 18A on each of 2 x +12V rails w\ the 600W
 
Ice Czar said:
Seasonic S-12's exceed spec in a number of ways and are rumored to be rated at 40C
they are just fine especially with 18A on each of 2 x +12V rails w\ the 600W
Click to expand...
Thanks, that's good to know. I know I'm barely making it sweat by running a stock 3200+ Venice 2.0 GHz CPU and a single 6800GT video card. :eek:
 
Ice Czar said:
a PCP&C 510 SLi or ASL would have no problem powering either of those configs
it also happens to be a single +12V rail, not a problem in a supply that tightly regulated

http://www.velocitymicro.com/wizard.php?iid=32
http://www.falcon-nw.com/config/build.asp
Click to expand...

*sigh*
VelocityMicro goes for the stupid-overkill-overprice route with a 1KW.
Falcon only offers the Silverstone 600W.

Stupid thing is that I've dealt with these same issues in the past. Disk arrays put different loads on PSUs, and tend to make them unhappy. Just ask PCP&C; they offer an off the shelf part same as what I used to use. Stabilzing a rail only takes a $1.21 resistor. That's it. Not PSU overkill. Just a simple (admittedly expensive for what it is) resistor. Stick it on the rail and watch things stop being cranky.
http://www.pcpowercooling.com/products/viewproduct.php?show=VLR

What just keeps cracking me up is that if they'd even tested these systems at all before shipping, they would have seen the problem. Or maybe they chose to ignore it. Either way, it's below amateurish and utterly irresponsible from a management standpoint. And their responses.. well, if you asked me, only a fool would do business with anyone involved in that fiasco.
 
AreEss said:
Disk arrays put different loads on PSUs, and tend to make them unhappy. Just ask PCP&C; they offer an off the shelf part same as what I used to use. Stabilzing a rail only takes a $1.21 resistor. That's it.
Click to expand...

that might be simplifying it a bit too much
http://www.smpstech.com/undest.htm
http://www.smpstech.com/oven0000.htm
http://www.smpstech.com/filter00.htm
http://www.smpstech.com/junction.htm
http://www.smpstech.com/chaos000.htm

and Id point out those configs didnt have arrays :p
a couple of HDDs and an optical, in the case of one of then 2 opticals

not that Im sure that resistor wouldnt help

totally agree on the testing by the builder though
 
Ice Czar said:
that might be simplifying it a bit too much
and Id point out those configs didnt have arrays :p
a couple of HDDs and an optical, in the case of one of then 2 opticals
not that Im sure that resistor wouldnt help
Click to expand...

Oh, it absolutely is oversimplifying to the extreme. But I'm certainly not going to give away the designs to my load regulation rigs. ;)
Regardless; you have to take steps to regulate based on your load. Disk arrays were just an example; you won't see that sort of problem till you're past 10 drives or so. While railsplitting is part of the spec, I want a PSU that autocomps for uneven loads. Since nobody makes one, I had to figure out a load generation and regulation setup. Extra expense, but in the end, absolutely worth it. (Uneven loads do age multirail PSUs prematurely. That's why array-only PSUs don't have +3.3V, 5VSB, etcetera.)

Ice Czar said:
totally agree on the testing by the builder though
Click to expand...

Yeah. That's just below amateurish, as I said. I still can't believe the responses they gave. Quizzing them on Ohm's Law? Blaming source AC? Drivers?! (That had to be the most absurd of all.) An insult to the customer's intelligence. If these vendors are so good, why were they busy making excuses? Then begging forgiveness after it gets out and offering absurd solutions like spending yet MORE money on additional drives you just don't need.
I hope the folks here at [H] will remember this next time they're shopping or reviewing OEM, and take this behavior into account. Their conduct is not just unprofessional, I'd argue it borders on criminal. They sold a system that they reasonably should have known would not work. They sold known defective equipment. Under most consumer protection laws, that's illegal. Frankly, if I tried that with my customers, they'd nail me to the wall with burning hot rivets. Why should these two get a pass?
 
AreEss said:
That's why array-only PSUs don't have +3.3V, 5VSB, etcetera.)
Click to expand...

I run a few antique Dot Hill SANnets w\ N+1 moduals
not that I bother to keep them spinning most of the time they are now more like nearline "coldstorage" backup ( half a rackmount for 2TB :p ) not worth the heat or power.
(Its just me here on the LAN no great need)

Most of the time I just get by with a 6 drive 200GB RAID 5 and a few extra individual HDDs in my "NAS" for lack of a better word.
Tis a pity you wont spill your secret recipe Id find that load balancing quite interesting
(and Im trying to build a PSU \ Thermodynamics test lab currently)

As far as those builders, personally Id have stuck those boxes on a variable AC supply and done some testing before releasing it, and considering their market, overclocked a few till the magic smoke escaped as well, there is no excuse like you said for a system like those with the premium price tag they are demanding. They have certainly recieved a wrap across the knuckles with that C|Net coverage, something enthusiasts arent going to forget and let die.

Its unfortunate the author didn't know more about the "why" of it. Propagating the "watts" as the measuring stick "meme"
 
Ice Czar said:
I run a few antique Dot Hill SANnets w\ N+1 moduals
not that I bother to keep them spinning most of the time they are now more like nearline "coldstorage" backup ( half a rackmount for 2TB :p ) not worth the heat or power.
(Its just me here on the LAN no great need)
Click to expand...

Mmm, .Hill, good stuff. I built my own arrays here; used some old Mylex-based serial-managed boards I had laying around. They do good work. Have some Hitachi 9K parts I should either sell

Tis a pity you wont spill your secret recipe Id find that load balancing quite interesting
(and Im trying to build a PSU \ Thermodynamics test lab currently)
Click to expand...

I've thought about releasing it, but I need my edge over the competition like Sun and HP. Plus even if I did, most folks can't build it; it's a pair of 4 layer PCBs with a lot of surface mount components and fairly complex traces. Plus most folks really wouldn't need it. It's not for cleaning up unloaded rails so much as shifting load for high wattage devices like TECs and pumps, and providing easy 3.3V feed for things like SMBus and I2C.

As far as those builders, personally Id have stuck those boxes on a variable AC supply and done some testing before releasing it, and considering their market, overclocked a few till the magic smoke escaped as well, there is no excuse like you said for a system like those with the premium price tag they are demanding. They have certainly recieved a wrap across the knuckles with that C|Net coverage, something enthusiasts arent going to forget and let die.
Click to expand...

One can hope the enthusiasts don't. (I could use the business.) :D
In truth, even I don't do a variable AC test, but that's for lack of equipment more than anything else. In truth, I don't need to really. I just slap it on a crap powerstrip and plug it into one of the unfiltered outlets around here. 40 year old wiring, single wire breakers, and a really bad neighborhood feed. (Nothing like watching your multimeter go '100 115 105 115 95 115' all day long.) If it can put up with that crap, it'll take just about anything. If it's worse than that, well, then nothing's going to work - you'll just have to have your wiring fixed.
 
I will be waiting for the New PCPOWER AND COOLING 610 AND 750 SILENCER?

they will be out some time in March
 
AreEss said:
In a nutshell...
SLI "Certifcation" is bullshit.
Let me say that again for people hard of hearing. SLI "Certifcation" is bullshit.
You give nVidia money, they give you a sticker.
Click to expand...

Although I agree the Nvidia certification is primarily marketing driven, I think you're over-simplifying the process a bit. ;)

Initially, SLI certification was done by a third party. Only until recently (about six months ago,) this testing was done in house.

After blowing up a multitude of power supplies that would have passed testing at the third party facility, I was brought in as a consultant through a mutual aquaintence.

They use a load tester similar to mine. And how they were blowing up the power supplies was by putting a 15A load on each PCI-e connector. Needless to say, they were putting a 30A load on a single rail. Do this to a single rail PSU like an OCZ or PCP&C 510 and you'll pass with flying colors. Anything else blew up.

Referencing the Intel ATX12V spec, it was explained to them that they could not expect to be able to put a 30A load on a single rail of a multi-12V rail PSU because Intel limits the amperage capability of each rail to 20A. If your 12V rail is capable of >20A, then you're not ATX12V compliant.

So the test was adjusted. But by putting a 15A load on 12V1, you're not really accounting for how the load is distributed. For example, most dual 12V rail PSU's put the PCI-e on the same rail as the drives. So if you have a lot of drives, you're eating into the power allocated for the video cards. Some power supplies put the PCI-e's on the 12V2 with the CPU. But then what happens if whatever you're doing is not only GPU intensive, but CPU intensive?

Personally, I don't think 15A per rail is enough.

And also, I think they really need to examine how the rails are distributed on a case by case instance and actually break that load up per rail based on what that rail is powering. Unfortunately, my assistance was no longer required after I pointed out to them that a 30A load on a single rail of a multi-12V rail power supply.

So I guess I agree with you. SLI Certification is B.S. But I don't think their intention is really to just pass anyone with a willingness to pony up (I though Nvidia only charged to SLI certify RAM, but I could be wrong.) What I do think is that they need to get someone in there that really knows what they're doing.
 
jonnyGURU said:
Personally, I don't think 15A per rail is enough.
Click to expand...


as so obviously illustrated earlier this month
The Grand Clash for Watts: Power Consumption of Modern Graphics Cards @ Xbit
.......................single.........................dual
X1900 XTX........120.60W or 10A........20.1A
X1900 XT..........108.7W or 9.A..........18.1A
X1800 XT..........110.5W or 8.2A........18.4A
7800 GTX..........94.7W or 7.8A.........15.8A

of course nVidia SLi certified doesnt apply Crossfire configs,
but some people would equate anyway :p
but the 7800 GTX blows right out the top as well without any other components or overclocking

jonny I was under the impression +12V1 always designated the CPU rail
thats not the case?

ATX12V v2.2
http://www.formfactors.org/developer\specs\ATX12V_PSDG_2_2_public_br2.pdf

1.2.2 Main Power Connector
"pinout assignments are based on the SSI recommendation."


4.5.1 ATX main power connector
pin 10 & 11 +12V1DC

4.5.3 Peripheral Connectors
pin1 +12V1DC

4.5.2 +12V (aux mobo) Power Connector
Pin 3&4 +12V2DC
Click to expand...

interesting that they have dumped the peripherals onto to +12V1 with the CPU
considering auxillary vidcard power "should"? be consider a peripheral?
or do they typically pull the PCIe six pin connector off the +12V2DC?

it seems absolutely hilarious the 6 pin aux PCIe vidcard connector(s) isnt in the latest revision with its own pinout assignment
 
Fortunately, or perhaps not, the ATX12V specification is left open to a lot of intrepetation.

They state that 12V1 is for peripherals. But is PCI-e a peripheral? ;)

So if peripherals are on 12V1 and the CPU is on 12V2, you're ATX12V compliant.

It's like the -5V rail. Nobody said you had to REMOVE IT to be compliant. Intel merely states that to have it is not necessary to have it, so some PSU companies have removed it to save money. Others still have a -5V and list it, some have a -5V and don't even list it on the label! :)

But yes, I think PCI-e should be mentioned more specifically in the ATX12V reference. But remember that PCI-e is a development of the PCI-SIG (of which Intel is a member of.) ATX and AGP are specifications of Intel and with technologies like SLI being with-held from Intel (which may be why Intel chipsets are going to migrate to Crossfire technology) I don't think Intel is in any rush to put forth standards that apply to PCI-e specifically. At least not until Intel has fully implemented Crossfire technology leaving everyone else behind the 8-ball.

Remember that ATI tests simiarly to Nvidia except the ATI requirement is 38A on the combined 12V rails for X1800 and X1900 cards (what PSU's were tested for X1800 and X1900 and what PSU's were tested for older cards are not differentiated on ATI's site, but an educated consumer can look at that list and conclude that most of those PSU's DO NOT have at least 38A on the combined 12V's.) Whether that means an even split of 19A and 19A on each 12V rail or if they mix it up depending on the PSU, I can't say.
 
jonnyGURU said:
So if peripherals are on 12V1 and the CPU is on 12V2, you're ATX12V compliant.
Click to expand...

+12V1 is for the CPU for ATX12V compliance, says so right on the spec main connector pinout
all the specs designate the CPU on +12V1

in ATX12V they also put the periphreals on +12V1
the Auxillary mobo power 4pin employed by PCIe via the slot on +12V2


so my question would be based on the majority of supplies youve examined if the 6 pin PCIe video card connector(s)
are wired up to the same point as the 4 pin auxillary +12V2 source
or the +12V1 source the main connector is?


and further if the majority of peripheral connectors (4 pin molezx) are actually on +12V1
 
Keep in mind that I may be misuderstanding what you're saying because I thought you and I were on the same page concerning this....



Ice Czar said:
+12V1 is for the CPU for ATX12V compliance, says so right on the spec main connector pinout
all the specs designate the CPU on +12V1
Click to expand...

The 12V lead on pin 10 on the ATX connector (20 or 24-pin) is for fan headers and slot power only. Not for CPU. The 12V on pin 11 is for PCI express requirements where applicable (1.2.2 of V2.2 specs.)

The CPU is on the 12V2 via the 4-pin connector (hence the name "P4 connector.") Unfortunately, the only specific reference to this in the ATX12V specs is in the notes for each table: "12V2DC supports processor power requirements..." but I think it's pretty clear what they're trying to imply there. ;)



in ATX12V they also put the periphreals on +12V1....
Click to expand...

Correct.



the Auxillary mobo power 4pin employed by PCIe via the slot on +12V2
Click to expand...

Not that I know of. In the ATX12V specs they state that the extra 4-pins of the ATX connector is for PCI-e slot power and that this eliminates the need for an AUX connector, but they're not speaking of the 4-pin P4 AUX connector that's on 12V2 that's used for CPU power. They're talking about the old 6-pin AUX connector that server boards and boards with AGP Pro used to use (never saw a PCI-e board with an AUX connector, but maybe they're out there.)

If you refer to the diagram in 4.5.1 you'll see that both 12V leads on the ATX connector are on 12V1. Power for the board and power specifically for PCI-e via the slot.

Take a look at 4.5.2 and you'll see that 12V2 goes to the 4-pin connector, the P4 connector, and this is the only specification for 12V2 (this leads into the answer to the next quote.)



so my question would be based on the majority of supplies youve examined if the 6 pin PCIe video card connector(s)
are wired up to the same point as the 4 pin auxillary +12V2 source
or the +12V1 source the main connector is?
Click to expand...

Since there is no specification for power to the PCI Express in the ATX12V specifications other than power provided to the slot via the ATX connector via 12V1, many power supply manufacturers take liberty with how the power is divided up.

Mostly I've seen PCI-e 6-pin connectors on 12V1 and the 12V2 alone to supply power for the CPU.. which I think is kind of silly because a CPU doesn't need 18A of power! :D Sometimes I'll see the PCI-e's on 12V2 with the CPU and sometimes I'll see each PCI-e on a seperate rail (one on 12V1 and one on 12V2.)

Power supplies like the NeoHE make the most sense by putting the PCI-e's on a seperate rail altogether, but this would not be a dual-rail PSU, would it? ;)



and further if the majority of peripheral connectors (4 pin molezx) are actually on +12V1
Click to expand...

They all tend to be on 12V1 unless you're talking about a quad-rail PSU, which typically meets SSI requirements. But when you're talking about SSI your numbers are all jumbled up. 12V1 is for one CPU and 12V3 is for the other... or if a quad-rail/SSI PSU is used on an ATX12V boards you can say 12V1 and 12V3 feed the CPU via the 4-pin connector.

So please note when I say "12V1" I'm talking about the 12V1 in the ATX12V specification that feeds power to the motherboard via the ATX connector and power to the peripherals and not the 12V1 that supplies power to the CPU as it would on an SSI power supply. :D
 
I see said the blind man to his deaf wife as he stuck out his wooden leg to feel if it is raining :p


now why did I thing the CPU was powered off the main connector?
just looked at the EPS12V spec and sure as @##@$ in B&W +12V3 for the 10 & 11
 
I can see how some think that an SLI certification for a PSU does not necessarily mean much. For example, I know many in the home theater system community thumb their noses at receivers that have THX certifications.

The flip side of the argument is that when those PSUs were certified, they were with a pair of 6800GTX video cards in mind. Now those same PSUs are expected to power a pair of OC'ed 7800GTX video cards. So maybe there should be an SLIv2 certification standard? :confused:
 
Ok.. Sorry.

Let me actually take a stab at this.... Perhaps you're asking about how the rails are split up?

It's not listed on PCP&C's website, but I'm 99% sure they use a standard SSI 12V load distribution.

So you're going to have one 12V to the board, two to the CPU(s) and one to the peripherals. PCI-e may be on the peripheral lead or one each of the two CPU leads.
 
beowulf7 said:
I can see how some think that an SLI certification for a PSU does not necessarily mean much. For example, I know many in the home theater system community thumb their noses at receivers that have THX certifications.

The flip side of the argument is that when those PSUs were certified, they were with a pair of 6800GTX video cards in mind. Now those same PSUs are expected to power a pair of OC'ed 7800GTX video cards. So maybe there should be an SLIv2 certification standard? :confused:
Click to expand...

SLI cert and THX cert is a bit different. At least in the case of SLI, they're at least trying to factor in something(power requirements) that isn't a personal opinion(THX channel levels).

THX certification just means the gear supports configuration such that it can fit into a system that will produce what THX thinks is the best sound. You can find plenty of people with a differing opinion on this issue, but if two videocards require X amperage/voltage, it isn't really a point of argument.

The only real meaning of THX certification to me is that the vendor thought this piece of equipment was high-end-enough to justify the expense of THX certification. Useful as a policy statement, if little else.
 
eastvillager said:
SLI cert and THX cert is a bit different. At least in the case of SLI, they're at least trying to factor in something(power requirements) that isn't a personal opinion(THX channel levels).

THX certification just means the gear supports configuration such that it can fit into a system that will produce what THX thinks is the best sound. You can find plenty of people with a differing opinion on this issue, but if two videocards require X amperage/voltage, it isn't really a point of argument.

The only real meaning of THX certification to me is that the vendor thought this piece of equipment was high-end-enough to justify the expense of THX certification. Useful as a policy statement, if little else.
Click to expand...
OK, I see what you're saying. So if I'm following your thought, you're basically saying THX certification is subjectively "awarded", whereas SLI cert is more objectively "awarded". :confused:
 
beowulf7 said:
OK, I see what you're saying. So if I'm following your thought, you're basically saying THX certification is subjectively "awarded", whereas SLI cert is more objectively "awarded". :confused:
Click to expand...

Sorta, I suppose, but I'm not really focusing on the awarded part, as on the concept itself.

SLI certification is pretty much just a quality/quantity of power rating. You can really argue good, clean power, and how much you need for a SLI rig. That is just raw numbers, determined by the requirements of the two videocards.

THX on the other hand is a set of audio specifications that describe a listening environment, and by extension, hardware requirements to create that environment. This listening environment is one that is "tweaked" to make it sound "better" to the THX people on the assumption that it'll sound better to everybody, and should thus become some golden standard. Sorta like the early days of A3D vs EAX, where A3D was actual 3d positional audio, while EAX was just some funky reverb tweaks---I admit, I'm no fan of creative, btw. In both examples, we have people trying to create a defacto standard which would then conveniently create an instant monopoly for their company.

Personally, I think the concept of THX is ludicrous when it is applied across the spectrum of media available. Even for media that was created end to end with THX in mind, I still think there is a great deal of "opinion" in the specification, opinion that isn't really backed up with any solid fact.

In the end, if you have a properly calibrated THX system, and media created with THX presentation in mind, you can rest assured that it is being presented as the producers intended. That said, try listening to something at THX reference levels without pissing off your wife, your neighbors, and any dogs within a 5 mile radius.
 
A real life scenario

I had the System listed below hooked up to an enermax SLI certified EG565V-PE with a 6800 overclocked to 425/825 and the CPU @2750/1.475VDC and had a stable system. When I upgraded to a 7800 GT I could not overclock both the 7800GT @485/1.118 and the CPU @2700/1.475VDC without getting a windows system crash message. I could overclock the 7800 GT or the 3800 X 2 but not both at the same time and be stable. I strrongly suspected the PSU was too weak for my system.

I ran this PSU estimator. The one advantgage this PSU etimator has over others is that you can plug in an overclock and it will factor that in to the PSU wattage estimate.

http://www.extreme.outervision.com/index.jsp

And the answer was about 485W. The EG565V-PE is rated at 535W. Definitely seems like i'm pushing my luck with the Enermax.

When I bought a silverstone ST60F I was able to overclock both the 7800gt and 3800 X2 without any Windows system crash messages.

I am wondering what will happen if I ever decide to go SLI with a second 7800gt but at the moment I'm happy with the silverstone. And I can't complain about the enermax since I was pushing it to its limit..
 
EnderW said:
wouldn't the issues with other split rail supplies affect them too?
Click to expand...

Not necessarily.

Because although there are four rails of lesser amperage each, the total 12V capability is quite high and I believe the PCP&C rails are on separate planes so the rails don't necessarily interact with each other.
 
I'll add this in here, since I didn't realize this discussion was already started, this article goes a little more in depth than the CNET article:

Here is a good article from extremetech on PSUs having problems with the new highend dual card set-ups. It talks about PSUs that handled their test, and some ways to work around the problem with other PSUs.

http://www.extremetech.com/article2/0,1697,1932957,00.asp

I'm going to try and delete my other post now ;-)

bigsarge
 
I actually like the ExtremeTech article better than some of the others I've read because they pooled facts and look at different aspects.

The fact that 7800GTX can pull as much as 11A each. The power supplies that they used. For example: The Silverstone ST65ZF. The PCI-e's is fed by a rail that supplies "only" 18A. Good stuff there.

For the record, they could have solved their problem without switching out the PSU. Of course, in their defense, they didn't initially know what the problem was. ;) If they had used a Molex to PCI-e adapter for one of the video cards, it would have worked fine because then you're moving one of the cards off to 12V4 which supplies 16A (even though the sticker says 8A.)

Ok guys... I've got an X-Finity 600W in my hands and an idea in my head. I'm going into the lab with a soldering iron. I've got an idea.
 
jonnyGURU said:
I actually like the ExtremeTech article better than some of the others I've read because they pooled facts and look at different aspects.

The fact that 7800GTX can pull as much as 11A each. The power supplies that they used. For example: The Silverstone ST65ZF. The PCI-e's is fed by a rail that supplies "only" 18A. Good stuff there.

For the record, they could have solved their problem without switching out the PSU. Of course, in their defense they didn't know the problem initially. ;) If they had use a Molex two PCI-e adapter for one of the video cards, it would have worked fine because then you're moving one of the cards off to 12V4 which supplies 16A (even though the sticker says 8A.)

Ok guys... I've got an X-Finity 600W in my hands and an idea in my head. I'm going into the lab with a soldering iron. I've got an idea.
Click to expand...

Depending on what you might already have on that line already

Does anyone know if the 11 amps per 7800GTX is stock speed or overclocked

BTW that was a nice rreview you did JG
 
You must log in or register to reply here.
Back
Top