I'm interested in this, even though I've given up on 5+ GPU setups - too much aggro getting them to work, and logic board / PSU costs increase non-linearly above 4 GPUs per board.

But my incompetence at getting a 5-GPU rig working (took ages, though *did* manage it eventually) may be explained partly by this issue.

I count 2 +12V wires on the fat EATXPWR connector, and another two +12V wires on the ATX12V power connector. My boards mostly use the four-pin ATX12V sockets as they aren't 'extreme performance' boards and aren't rated for BIG wattage CPUs, but the 5-GPU board I was messing with had the 8-pin version, which has four +12V wires.

So for the logic board and expansion slots, you've either got four or six wires, each of which is 18AWG according to the ATX PSU spec (though most of the high-power PSUs use 16AWG - my Corsair does, for example).

18AWG cable is 'recommended' for up to 40A at 12V with cable lengths 3 feet and below (a reasonable assumption in computer PSUs), according to this link:
http://www.rbeelectronics.com/wtable.htm - I am not 100% sure of the veracity of its claims, so YMMV, but I'm assuming that the electronics guys writing that website mean that the 'recommended' gauge in the table won't be running within an inch of its life.

Hence if the table suggests that a 12V requirement of 40A would be *ideally* served by 18AWG cable for 3 feet or less, then I assume that this *doesn't* mean that said cable will be hot to the touch and starting to soften its insulation...

On top of this 'absolute' max current the cable will carry, there's the ATX spec, which IIRC allows a max of 240VA on any wire. So for our 12V feeds for our GPUs etc., the ATX standard says 'no more than 20A please chaps'.

So I'm guessing that even with my 'small' logic boards, the four 12V feeds can give a maximum of 960W *just to the CPU and expansion slots* - that's at the maximum 20A per wire.

But if 18AWG is OK for 40A on short stretches, I can't see it even getting *warm* at 20A. Temperature of cabling is much more likely to be influenced by the hot air expelled by the GPUs and other components in a bitcoin rig, if what I say above is anywhere near true.

Using 'warm cables' as a diagnostic measure is therefore a BAD idea since it's going to result in false positives ALL of the time. If 18AWG is good for 40A as 'recommended' then I doubt it'll get hot until over 50A - at which point you're talking about 1200W through ONE wire, which no GPU will pull. Note that I'm ignoring the CPU and chipset here, because as a bitcoin mining rig, the CPU is underclocked to the minimum and only one core runs. If you're running a 250W CPU as well then the power available to the PCIe slots *may* become a problem. But the cables shouldn't get hot, surely??


Personally, I'm much more concerned with GPUs pulling power through the nice PCIe extenders a lot of us use. The ribbon cable used, even dual layer, is much thinner than 18AWG and I *really* don't know the safe ampacity of the conductors. Personally I'd like to keep the GPUs pulling ALL their required power from their 6-pin connectors - even a single 6-pin connector (with three 12V conductors at 18AWG) would be safe to a 720W load, assuming the wires followed the ATX spec. Ignoring the spec, using one conductor to supply an entire GPU would be safe up to 480W. I'm happy with that, which is why I'm not too worried when I see 6-pin converter cables that simply loop two 12V pins to the same cable, and the same with the ground. The cable *itself* can carry the current.

The ribbon cables are a problem though. Molex-augmented extenders are IMO irrelevant unless your logic board's *LOGIC* limits the 12V current supplied to the PCIe slots artificially. If you *really* need the current in the extender, then the limit isn't the Molex or the 18AWG yellow wire - it's the thin conductor in the ribbon cable that the yellow wire is spliced to.

I've got a Molex-powered extender from Cablesaurus. The Molex connector takes a single stretch of 18AWG and splices into the ribbon cable. How can the Molex augmentation supply more power than the ribbon cable can already safely take? In fact, since the ATX spec allows 20A on those 18AWG Molex yellow wires, the powered extenders could possibly make fire *more* dangerous, by allowing the thin ribbon cable to draw more current than the slot itself would supply.


Can anyone tell me the wire gauge and rating of the ribbon cables used in our PCIe extenders? It's not usually a problem, since IME the power drawn by GPU cards is primarily pulled from the 6-pin power feeds, with only residual power taken directly from the board (obviously only talking about GPUs with 6-pin power connectors here, of course!). I've got a Mac Pro which has a mad number of sensors to query, and it's telling me the exact amperage going through the two PCIe power extensions (I've got Y-cables on these...), along with the exact amperage going through each PCIe slot on the logic board. If this behaviour (preferentially using the 6-pin supplies before taking from the PCIe slot) is some cool Mac OS X thing, then my theories may be useless - because none of us run Mac OS X as a main mining OS, and all my big-power GPUs are on Linux boards. But if this is how the GPUs are designed, and also due to higher resistance on the PCIe extenders (making the distribution an electrical issue), I don't see how splicing an 18AWG power cable into the *start* of the existing extender ribbon cable will reduce the resistance of that ribbon cable.

Sorry for big post, thinking out loud.



TL;DR = I don't think powered PCIe extenders are of any use beyond the edge-case where your logic board *deliberately* restricts available current to the PCIe slots when all are populated. And on the other hand, the Molex connector is capable of delivering enough current to melt and set fire to the ribbon cable it's spliced into. I don't use them. Happy to be proved wrong, or any assumptions challenged though