SLI and CrossFire, Pushing Power Supplies To The Limit

gonaads said:
Correct me if I am wrong, but wasn't the whole idea behind "modular" was for ease of install/setup? You don't connect the sections of wiring that you do not need. This way there is less wiring floating around inside your box. So there is a cleaner install, better air flow and all around neatness.

Yes/No?

Yeah, I think that was the thinking behind it. In my case, it was "if I could just get the stinkin wire outa there, I could have removed that bad/dieing drive years ago..." The extra wires can add more clutter, and also make space (for working in there/cooling, even tighter).

As to my system, I'll give somewhat of an idea, albeit not having SLI video (long ago I had Voodoo2 SLI on 1 card, aka the Quantum 3D Obsidian 2, but those didn't draw the power of todays cards...). The Radeon 9600 pulls a bit less then the newer gfx cards, I'm rather certain. But:

- Athlon 64 3500+ OCed at 210 MHz for the HTT, (was 217 MHz for the HTT, with my older XMS 2400 DDR, prior to my getting 1 GB of Corsair XMS Pro 400). The CPU is running at 2,310.01 MHz according to wCPUid... Default is 2.2 GHz

- 2 10,000 rpm SCSI drives (Seagate Cheetah)

Edit: Rather then estimate, I'll give Seagate's power info on the model numbers:

http://www.seagate.com/support/disc/specs/scsi/st39102lw.html

POWER DISSIPATION (watts/BTUs) Active ____13.8/47
Idle ______11.2/38
POWER REQUIREMENTS: +12V START-UP (amps) _1.7
+12V TYPICAL (amps) __0.60 idle
+5V START-UP (amps) __0.96
+5V TYPICAL (amps) ___0.80 idle
IDLE (watts) _________12

http://www.seagate.com/support/disc/specs/scsi/st336704lw.html

POWER DISSIPATION (watts/BTUs) Active ____/
Idle ______10.91/37.2
POWER REQUIREMENTS: +12V START-UP (amps) _2.2
+12V TYPICAL (amps) __0.58 idle
+5V START-UP (amps) __0.86
+5V TYPICAL (amps) ___0.79 idle
IDLE (watts) _________

- 2 internal CD/DVD drives, one of which is a CD burner (The external DVD burner doesn't matter, cause it has it's own power transformer)

- About 4 PCI boards, from the SCSI card (don't use IDE drives at all), NIC (prob with the integrated NIC, nForce 3, and drivers in winXP, though it works fine in Linux), DVD decoder card, and probably one other PCI board

- 5 USB devices, including a scanner, that has no power adapter, so draws all power from the USB port, at the 500 mA maximum The remainder pull about 100 mA a piece for the most part...

- 8 to 10 fans case fans, a few about mid-range in terms of RPM...

Hadn't added all the power consumption up, though looking at it, I do know my old 400 watt PSU was deficient on the 12 volt rail for just the A64 alone, as it put more on the 3.3 and 5 v rails then some... I'm running about 40 processes now (just in the browser), including about 4 or 5 folding like apps, under BOINC, just to give an idea what's running here :D

Sazar said:
But you are introducing additional resistances and therefore making the supply of power less "clean".

Actually, this does make some sense, that it could add a bit of resistence, come to think of it... How great it would be, I'm not sure off the top of my head though...
 
Last edited:
Well let me see my rig has 5 hard drives , 2 x dvd burners , 2 gigs ram , loads of USB devices , printer, a 7800GTX , 6 case fans its a 939 pin 3800 X2 .. umm not a single problem with my power supply guess I need more stuff to really push my PSU . And 46 running processes .
 
Last edited:
Mine isn't having a problem either... However, and that said, what Sazar is saying about adding increased resistence to the circuit does make sense. I haven't done the math on this, but from a computer engineering standpoint, his statement is not without a degree of logic... Afterall, it would require looking at the specific materials themselves, and also whether any air is present between the mettle contact in the cable, and the mettle contact in the pins on the PSU, that current would have to traverse... Air is of course a natural insulator, though if the connection is close enough, the current can "arc" accross... Lets just say that I wouldn't rull out the possibility for a bit of a power drop over that connection there, though the specific I haven't really looked into myself...

In theory, I can see a possibility for what he's saying (and there probably would be more to read up on). In practice, I do however have my experience...
 
And thats what I am looking at also is the REAL world experience over a year running 24/24 I cant say I am not pleased.
 
Sazar said:
It looks nicer, less cables, potentially better air flow (if you don't care about doing your own cabling clean-up and all round neatness.

But you are introducing additional resistances and therefore making the supply of power less "clean".

Efficiency therefore is reduced. It should not make a difference for most systems, but crossfire/Sli systems are not most systems :cool:


This is quite true, any time you add a connector into a circuit it will raise the total circuit resistance.

But.

If the connectors are made of low resistance materials such as Gold, then you are reducing this added resistance to negligible levels.

Now with the work I do on Automotive electrical systems and their computer systems any time there is a need for splicing wiring the best option is to solder the wires than to use crimp spices or some other form of connector. Now granted this is not the situation with a PSU. But the modular connectors on PSUs when they are configured must take into account any added resistance that they may contribute to the circuit. Also the Power that goes through these connectors is considerable. So any loose connection at these modular pieces can cause arcing which will create suface resistance on the connector pins which in time will cause major resistance if not fixed. One piece of advice is that if you do have a modular type PSU don't wiggle the connectors when inserting them. Any uneven force can open these connections and cause looseness that will cause current flow problems.

Anyway, my two cents worth. :)
 
Yeah, I've only had it since last October... And there is my experience which tells me one thing, at least wrt what I'm running. It is correct to say that I don't have an SLI setup with the lattest gen cards however. I do have a lot of stuff on here, but no doubt am probably running closer to 75-80% load, rather then 95%... (Beyond 95%, I don't think I'd tend to want to push any PSU...)

But lets just say, that when added resistence was mentioned, the sense of his argument was quite apperent to me, and from what I've learned thus far (albeit I still have a ways to go in my degree in comp engineering), it makes perfect sense... Now, how much resistence is added, I don't know... Beyond that, I couldn't comment much at present... But there is a rational basis from which he could get such argument; and hence I wouldn't discount it out of hand. I of course couldn't ignore my experience either, however that can be placed; if this all makes any sense :)

This is quite true, any time you add a connector into a circuit it will raise the total circuit resistance. But. If the connectors are made of low resistance materials such as Gold, then you are reducing this added resistance to negligible levels.

I'm not sure the material, but I know it's not gold, or gold appearing. Looks more like a metalic (silverish) color, though it coudl be titanium as easily as steal. Not sure of the material used though...
 
Last edited:
"Turbo-Cool 510 spec: full load @ 50°C.
"500W" Home PSU spec: full load @ 25°C, no load @ 70°C."


No name brand on the "Home 500W" ?

And a study bu the company that is selling the items? Where there are no guidelines on any study like this?

All I gathered from that was "Finally, we ended up using PC Power and Cooling's massive TurboCool 850 SSI. This just goes to show that if you really want to build something that pushes the bleeding edge, make sure you have the right power supply."

From the PcPower&Cooling website. Be sure to visit our reseller area!!!!! :D
 
Steevo said:
"Turbo-Cool 510 spec: full load @ 50°C.
"500W" Home PSU spec: full load @ 25°C, no load @ 70°C."


No name brand on the "Home 500W" ?

And a study bu the company that is selling the items? Where there are no guidelines on any study like this?

All I gathered from that was "Finally, we ended up using PC Power and Cooling's massive TurboCool 850 SSI. This just goes to show that if you really want to build something that pushes the bleeding edge, make sure you have the right power supply."

From the PcPower&Cooling website. Be sure to visit our reseller area!!!!! :D

If you look at any enthusiast setup, not a kid with lots of money but a true enthusiast setup, you will see only one type of PSU in there.

Dismiss their comments, thats your perogative, but what they say (for the most part) makes sense and is borne out in real world tests.

Read some independent reviews of their products.
 
Heh, I have SLI, but nothin like the Quad SLI setup the new Dell Renegade or w/e will have, just need to buy that second 256mb gfx card and i'll be good for a while..
 
Ok, since nobody believes SAZAR I'll do the numbers for you.

5V is required to be +/-0.25V at every chip in your system. The power supply is regulated inside the PSU, not at the chip, or even at the card connector.

-Typically PSU voltage reference stability is 1% with age and temperature variations and sense resistor tolerance of 1% which means +/-0.1V of your 0.25V margin is gone before you leave the PSU box.

-Now with a 20A load (100W at 5V) that means you have 0.15V/20A = 0.0075 ohms for wire plus connector drop allowed.

A good, gold plated connector is rated at 0.001 ohm new and degrades to 0.0025 ohms with time and a few plug unplugs. (Quadruple this for an ultrax system, 4 pins per power feed 0.01 ohms for an aged system.) Hint, those crappy white Molex connectors are not gold plated so are even worse. You also have about a 1 foot power line, plus 1 foot of return line of 18AWG wire, which is 0.010 ohm per foot or about 0.015 ohms total (return wire is doubled so only counts half).

If you can count and haven't fallen asleep yet you realise your voltage drop budget is already well out of range. You got 0.0075 ohms you have used 0.025.

Now you know why they stopped using 5V for everything and went to 12V. It halves your current load and doubles your voltage tolerance. Divide 0.025 ohms by 4 and you get within your 0.0075 ohm tolerance.

The above neglects volatge drops through the MB to PCI-e and AGP connectors and then the drops through the MB to card connectors.

Designing power supplies for high power, low voltage digital equipment is a bitch. I used to do it for a living. Adding extra pins where they are not needed is absolute stupidity. I'm amazed performance PC's are as stable as they are with 400-500W loads.

Conversion to 12V feeds instead of 5V, 24 Pin MB connectors, beefed up PCI-e power connectors, parallel 12v feeds to the back sides of cards, and on board voltage regulators on the video cards and MB are the result of just how ugly power distribution inside a PC has gotten. Now some marketing idiot decides to add another voltage drop by using modular connectors.

Notice how the better brands like Antec resisted going modular until they had to because of marketing pressure from so many of the second string brands going modular. BTW all those extra connectors cost money and add labor hour cost to your PSU price while degrading performance.

PS You don't want minimal cables. You want to run every strand of copper you can get from the PSU to the loads to minimize voltage drop. That means that you should use every cable provided with the modular PSU to spread the current around and reduce voltage drops. So why pay extra for modular???

Amen, sermon over.
 
One other thing to take into account is the strands of the wire itself. Stranded wire is by it's nature, compressable. It deforms to provide a high-reliabilty, high contact area connection. For vibration prone, and flexing applications. Technically speaking, stranded can carry more current than solid wire. Now you could have two top PSU manufacturers and one has 12 strands per wire. Then you have the other and their's has 20 strands per wire. Who's is going to be the better? Now throw this into the equation. The one with 12 strands is 16 AWG wire and the one with 20 strands is 14 AWG. Now which one conducts better?

Remeber the small the wire the more resistance but if there are more strands then what?

Also current is moving electrons, they travel "through" the conductor and along the surface of the conductor. They move along the surface better because it is easier, so more current is distributed along the surface areas than in the center. Stranded conductors have a lot more surface area so they can carry more current for a given overall diameter. So, back to my question, The one with 12 strands is 16 AWG wire and the one with 20 strands is 14 AWG. Now which one conducts better? So If a modular PSU uses a more stranded wire setup that say a non modular PSU then would the resistance values still be the same?

More "surface", more current flow, and more heat dissipation (current flow creates heat) and less resistance and less voltage drop. Heat creates resistance.
 
gonaads said:
One other thing to take into account is the strands of the wire itself. Stranded wire is by it's nature, compressable. It deforms to provide a high-reliabilty, high contact area connection. For vibration prone, and flexing applications. Technically speaking, stranded can carry more current than solid wire. Now you could have two top PSU manufacturers and one has 12 strands per wire. Then you have the other and their's has 20 strands per wire. Who's is going to be the better? Now throw this into the equation. The one with 12 strands is 16 AWG wire and the one with 20 strands is 14 AWG. Now which one conducts better?

Remeber the small the wire the more resistance but if there are more strands then what?

Well, the effect of running resistors in parallel, vs. running them in series is different. It would not be beyond reason that something could be extrapolated from this, to apply to the resistence in the wire itself, albeit the connectors are end points...

http://physics.bu.edu/py106/notes/Circuits.html

Parallel circuits

A parallel circuit is a circuit in which the resistors are arranged with their heads connected together, and their tails connected together. The current in a parallel circuit breaks up, with some flowing along each parallel branch and re-combining when the branches meet again. The voltage across each resistor in parallel is the same.

The total resistance of a set of resistors in parallel is found by adding up the reciprocals of the resistance values, and then taking the reciprocal of the total:

equivalent resistance of resistors in parallel: 1 / R = 1 / R1 + 1 / R2 + 1 / R3 +.

A parallel circuit is shown in the diagram above. In this case the current supplied by the battery splits up, and the amount going through each resistor depends on the resistance. If the values of the three resistors are:

9d.GIF


With a 10 V battery, by V = I R the total current in the circuit is: I = V / R = 10 / 2 = 5 A.

The individual currents can also be found using I = V / R. The voltage across each resistor is 10 V, so:

I1 = 10 / 8 = 1.25 A
I2 = 10 / 8 = 1.25 A
I3=10 / 4 = 2.5 A

Note that the currents add together to 5A, the total current.

Actually however, it's more the

(Quadruple this for an ultrax system, 4 pins per power feed 0.01 ohms for an aged system.)

that I find problem with, because all 4 pins aren't involved in the same connection. The given cable used to connect up one's drives, runs both 12-volt, and 5-volt, with a ground (the black wire in a non-rounded power cable) for each. In fact, I just looked at one of these cables and it's 4-pins at both sides, supplying both 5-volt and 12-volt. The quadruple, if we're talking the 5 volt feed doesn't hold, as half those pins are involved in the 12-volt power feed (one for the 12-volt line, and 1 for it's ground). Therefore, doubling this value, rather then quadrupling it would seem more accurate. The pins that carry the 12-volt power feed, are not part of the circuit path for the 5-volt power feed, and as such shouldn't necessarily be computed as such...

BTW, when Saz mentioned additional resistence, I had already indicated that there was sense in what he was saying, and it was the connectors themselves I was thinking of. I didn't want to be the first to come straight out and say it however, lest I was mistaken... Gonaads himself had agreed with Saz on this point; and I also reped Saz for his comment there, mentioning that from a comp engineering standpoint, there is a logical basis to what he suggested there.

However, I also mentioned my experience and what I have noticed from it as well. There is a basis to what he said, and the argument was well thought out and presented there. At the same time, I don't suspect that he would expect us to ignore the testimony of our own eyes and experience, either... It's not really a matter of believing or not believing, at least here ;)
 
Last edited:
ok this has devolved into the esoterica, when we start relating to Coloumbs law and particle physics then I think we lost most of the laymen out there. Still nice find there guys and makes for good reading so you are all repped for taking the advanced physics classes in college.

Thank god someone had to take them with me :)
 
Heh heh weeeeeeeeeeeee, we got a physics and electronics 101 class. Actually it's more like advanced now. :p



But the bottom line is as with any connector that is of connect and disconnect design. You have to ask, "how many times am I going to be disconnecting it/them?

Say you put together a system with a modular PSU. You connect all the pertinent harnesses that you will need. O.K. that's it, that's all the plugging and unplugging you are going to do. Now say 3 or 4 months later you wish to add something to your box and it will need it's own power. The connectors that are plugged in to the modular PSU are all used. So your setup has a couple more modules that can be used. You connect one and plug in whatever it is that you installed. There, you're done. Now say, 4 to 6 months down the line you go and upgrade your system. Well maybe a year down the line to make it more real world. You unplug everything and redo your system. Maybe you have to disconnect a module or two, maybe you don't. The point of this long thought here is that you are not plugging and unplugging these modular connector every day like you would a hairdryer. So the resistance buildup could possibly not come into play here. Why, because these connectors are not going to loose any clamping force from just being connected (with the every 6 month disconnect, that may or may not happen). Now also if these connectors are coated with some form of dielectric grease before they are plugged in they would stay clean of corrosion and any other uglyness that could possibly cause a resistance rise.

Food for thought.
 
gonaads said:
Heh heh weeeeeeeeeeeee, we got a physics and electronics 101 class. Actually it's more like advanced now. :p



But the bottom line is as with any connector that is of connect and disconnect design. You have to ask, "how many times am I going to be disconnecting it/them?

Say you put together a system with a modular PSU. You connect all the pertinent harnesses that you will need. O.K. that's it, that's all the plugging and unplugging you are going to do. Now say 3 or 4 months later you wish to add something to your box and it will need it's own power. The connectors that are plugged in to the modular PSU are all used. So your setup has a couple more modules that can be used. You connect one and plug in whatever it is that you installed. There, you're done. Now say, 4 to 6 months down the line you go and upgrade your system. Well maybe a year down the line to make it more real world. You unplug everything and redo your system. Maybe you have to disconnect a module or two, maybe you don't. The point of this long thought here is that you are not plugging and unplugging these modular connector every day like you would a hairdryer. So the resistance buildup could possibly not come into play here. Why, because these connectors are not going to loose any clamping force from just being connected (with the every 6 month disconnect, that may or may not happen). Now also if these connectors are coated with some form of dielectric grease before they are plugged in they would stay clean of corrosion and any other uglyness that could possibly cause a resistance rise.

Food for thought.

Not really. You make a good point but the mere presence of the modular connectos implies there will be added resistance and reduced overall efficiency.

Would you rather a powerstation have transformers/boosters and provided current to a location in as straight a line or with multiple "modular" segments?

Get the drift?

It may not be a dramatic difference in most cases but you are simply getting a cleaner/better psu w/o modular design (if you get a good psu to begin with).

Go modular for looks or clean design or whatever. If you want a better quality PSU, go non-modular.

Leejend explained that about as well as can be expected.
 
I agree whole heartedly Sazar. But truthfully I think the main reason for these modular PSUs was/is just for the sake of ease of use, installation and for the customizer that wants a super clean wiring setup. No more no less. Hell if you're good at setting up systems you can make a great super clean install with a conventional PSU.

Ya gotta remember that these modular setups are only going to be used in home systems. I really don't think anyone in their rite mind would use these in server applications.

The points I was making were to how minute these values can be for a home system. Granted that some of us will have higher end systems but then we are a tad bit more informed as to if the drawbacks of these PSUs are worth it or not to use.
 
Gonaads, this thread is about SLi and Cross-fire solutions. Those are hardly run of the mill household items :cool:

This is why I brought up the better psu's to begin with.
 
Sazar said:
Not really. You make a good point but the mere presence of the modular connectos implies there will be added resistance and reduced overall efficiency.

Would you rather a powerstation have transformers/boosters and provided current to a location in as straight a line or with multiple "modular" segments?

That might not be the best example; because public utilities are almost certainly going to have segments in laying the power out. In fact, there's already 2, as 220-volts is what comes into a modern house (some older homes only got a 110-volt line), which itself gets broken down to multiple feeds for 110-volts for general use, and 220-volts for driers, hot water heaters, and the like. But they transport 440-volts over long distances (and from the power generation stations), for specific reasons. They probably won't find one 1 cable that's 20 miles long to get it around town either...

And yes this can get ugly. Telephone rather then power, but when I first got aDSL, I had to get a lower quality aDSL (609 Kbps) even though I'm only 9,000 ft from the CO, because Qwest layed cable of different gadges. When I called up, they said there were 5 different segments that went from like 24 gadge to 22 gadge, to yet something else; with the first change being < 500 ft from the CO. It happens, and yes it can have an effect... Albeit I guess it got cleaned up some, as I now have 1.5 mBit...

Leejend explained that about as well as can be expected.

I think the point was that the values wouldn't necessarily be quite as bad as Lee was suggesting. In fact, looking at the math he was using; I would have to say it wouldn't be as bad. Looking at the 5-volt power connector, he quadrupled it for 4 pins in the connector. However a standard drive cable doesn't just supply 5-volts, but 12-volts also. A +5-volts (or -5-volts if one were looking at electron flow, rather then conventional flow), a +12-volts, and 2 grounds.

Quadrupling the resistence values for 4 pins is in error; because only 2 of those pins has to do with 5-volts. The other 2 are in reference to the 12-volt feed that's available in case the drive does (and in fact the motors on the drives tend to use 12-volts, with the electronics using 5-volts), needs it. Those other 2 pins aren't part of the same circuit; resulting in his values being 2x higher then they should be...

Actually, I don't think any of us disagree that added resistence is added. Some of us might not agree on what the actual numbers will be, but since you mentioned the added resistence, I don't think that either Gonaads, or myself, had said it wouldn't be added... Also, the cables I'm looking at don't use the lousier white molex connectors on either end, so if both ends had a higher queality connector then standard... But yes, there would be more resistence...
 
O.K. We all tend to agree that there would be more resistance. But you have to agree that if these manufacturers put these values into their equations they are going to compesate for these small variations. Also isn't a computer with all it's expansion cards Memory Modules CPU and Video card "modular"? You plug in the connectors from the PSU to the MoBo, to the HDD, to the Vid Card, and so on. You plug the CPU into it's socket, you plug the Mem Modules into their sockets. I think you get the picture. So now, does each and every one of these have a connector to deal with? And doesn't every single one have the possibility of added resistance over the course of time and/or due to disconnecting them and reconnecting them? Now I said home use to simplify my statement. Maybe it was too simple. But SLI/Crossfire systems is what this was/is about.

But in all truth there is no way of knowing what system configuration will have more or less resistance than another. There are way too many variables. Solder joints, quality of the wire, wire guage, type of capacitors, the connectors, the plating of said connectors, the quality of the solder used at that moment, on and on it goes. Plus these PSUs have voltage "ranges" with + or - percentages, load variation "ranges" with + or - of that. You have Hold-Up Time, Load Transient Response, Overshoot. All these values must be precise as to the model of PSU and should be tested as to these values. Now except perhaps in uber-mission-critical applications where the miniscule voltage drop from an extra connection point is unacceptable, this is probably a non-issue. I think very few of us use our PSUs to operate heart-lung machines. And I think that with that miniscule voltage drop that may cause a drop in your frame rates by 2 or 3 I don't think it is of much consiquence.

It makes for good discussion on deep technical viewpoints but for the rest of the SLI/Crossfire junkies out there, I don't know.

I will say this was/is a fun thread. :)


[edit]
This may not be as to the modular PSUs but it is one hell of a read for anyone who whats info on the energy efficiency of computer power suppies.

http://www.efficientpowersupplies.org/pages/Energy_efficiency_computer_ps_EngTr.pdf
 
Last edited:

Members online

No members online now.

Latest profile posts

Also Hi EP and people. I found this place again while looking through a oooollllllldddd backup. I have filled over 10TB and was looking at my collection of antiques. Any bids on the 500Mhz Win 95 fix?
Any of the SP crew still out there?
Xie wrote on Electronic Punk's profile.
Impressed you have kept this alive this long EP! So many sites have come and gone. :(

Just did some crude math and I apparently joined almost 18yrs ago, how is that possible???
hello peeps... is been some time since i last came here.
Electronic Punk wrote on Sazar's profile.
Rest in peace my friend, been trying to find you and finally did in the worst way imaginable.

Forum statistics

Threads
62,015
Messages
673,494
Members
5,621
Latest member
naeemsafi
Back