My PSU popped.

[EsaT]

Where do people get these crazy ideas from? The cable design comes from the PCIe power specification, not ATI or NVIDIA, they use the same design!

By believing marketing departments, just like you: Go to check fuse box of your house and then tell us how many power lines your house has.

There aren't any separate 12V rails in PSUs but only one current source and artificially current limited wire groups! Corsairs, Seasonics and other UL Level 6 PSUs (not much others) just wipe their rears to those limits and give 12V out from any cables.

 

[weird_dave]

EsaT, did you even read post 40?

 

[HazardO]

I couldn't find that piece of info, but it doesn't make the situation any more misleading. My point was that it doesn't matter where the GFX was getting 12V from since the rated outputs on each 'rail' were the same. The fact that there may be just one rail doesn't matter, you can't pull all the current down one set of wires hence the separate rating for each output.

Glad it's all up and running again

I only found out about the single rail from the corsair forums when I wanted to know if the HX620 would cope with a 9800GX2. But you are right, the connectors and cables are probably not good enough to supply the full amperage through one connector. This is also a reason why the twin molex to 8pin PCIe adapters are not neccessarily a good idea

 

[wonko]

This thread made me curious on what sort of current draw / limits were being talked about here, and thought others may be interested in what I found - any corrections welcomed...
The HD5770 draws 108W maximum, and if this were all via the PCI-E connector is 9A at 12V.
The molex connector spec seems to rate each pin at up to 10A, so a single molex to PCI-E power connector looks up to the job, albeit only just. The HX620 is current limited to 18A on each 12V output (I won't say 'rail'!), which again should be OK (ignoring other loads).
By contrast, I found a figure of 197W for the 9800GX2. Again taking worst case that all this is via 12V PCI-E gives 16.4A. Really quite impressive and well over what a single molex 12V pin is rated for, but using 2 molexes in parallel would seem ok provided the 12V line from the PSU wasn't shared with anything else to exceed the 18A rating.
All assuming that the PSU and connector cables are also rated for the current, of course. I'd tend to trust a cable provided by a respected manufacturer of a PSU or graphics card, but might be suspicious of an unbranded power cable where someone had saved some money on copper or quality of connectors.

 

[weird_dave]

I only found out about the single rail from the corsair forums when I wanted to know if the HX620 would cope with a 9800GX2. But you are right, the connectors and cables are probably not good enough to supply the full amperage through one connector. This is also a reason why the twin molex to 8pin PCIe adapters are not neccessarily a good idea

The twin molex are not a fantastic idea, but they do get you out of a spot, works fine for 200W for GFX, just don't use 2 on the same output
I just checked the current rating for the molex connectors, 10A, which I guess is why there's 2

Edit:
I forsee a burning smell should one of the molex connectors come loose.....

 

[EsaT]

The HD5770 draws 108W maximum

~80W worst case, more like 60W in games.
http://www.xbitlabs.com/articles/video/display/radeon-hd5770-hd5750_5.html#sect0
And probably 1/3 or half of that through PCIe slot.
6 pin PCIe power connector itself is specified for 75W, and cards have to comply with that limit set in specifications made by PCIe SIG.

By contrast, I found a figure of 197W for the 9800GX2. Again taking worst case that all this is via 12V PCI-E gives 16.4A.

All graphic cards draw practically all power from 12V (like CPUs) because using lower source voltage would cause major and entirely unnecessary transfer losses.
And there aren't any excessive currents through single "contact".
http://www.xbitlabs.com/misc/picture/?src=/images/video/geforce9800gtx/9800gtx_power_full.png&1=1
(for 8 pin PCIe power connector max specified power is 150W)

The HX620 is current limited to 18A on each 12V output

No it's not! That "multi" rail crap is there on sticker just to satisfy those ones falling for that BS minefield.

For testing over current protection (OCP) we pulled 33 A from EPS12V (the maximum we could configure using our load tester)
http://www.hardwaresecrets.com/article/371/9

Although I can not say that these rails are or are not somehow electronically separated in the PSU's circuitry somewhere, I did find that there was no OCP (over current protection or "limiter") on either of these rails as I was able to load any given connector up 30 to 40A
http://www.jonnyguru.com/modules.php?name=NDReviews&op=Story2&reid=21

For fun, I put a 30A load on the 8-pin EPS+12V connector...
I did the same with the ATX main connector and again the PSU did not trip.
http://www.jonnyguru.com/modules.php?name=NDReviews&op=Story&reid=18

I asked Seasonic about this "descrepency" and it was stated that the "deception" is really quite innocent. Apparently the initial design was for quad rails, and if rails did need to be split up into three or four rails, it would be easy to implement without completely redesigning the platform. Multiple rails can be split up with a separate OCP circuit board, typically screwed to the top of a heatsink.
What Seasonic tells me is that there was an issue with high end video cards overloading a single 12V rail. So with Intel's blessing, Seasonic removed the OCP. Technically, Intel has removed the 240VA rail limit from the ATX12V specfication (although I've yet to see this published on FormFactors.org), and the UL simply considers devices with single output leads with output capability greater than 240VA (12V @ 20A) a different product class (level 6 instead of level 3.)
http://www.jonnyguru.com/modules.php?name=NDReviews&op=Story3&reid=3

 

[wonko]

The HD5770 draws 108W maximum, and if this were all via the PCI-E connector is 9A at 12V.

~80W worst case, more like 60W in games

The figure of 108W I found was from AMD http://www.amd.com/us/products/desk.../Pages/ati-radeon-hd-5770-specifications.aspx

And probably 1/3 or half of that through PCIe slot.

Doesn't surprise me - I was deliberately taking the worst case though...

6 pin PCIe power connector itself is specified for 75W, and cards have to comply with that limit set in specifications made by PCIe SIG.

OK, makes sense.

By contrast, I found a figure of 197W for the 9800GX2. Again taking worst case that all this is via 12V PCI-E gives 16.4A.

All graphic cards draw practically all power from 12V (like CPUs) because using lower source voltage would cause major and entirely unnecessary transfer losses.

Seem to be agreeing that my assumption it was mostly 12V is correct.

The HX620 is current limited to 18A on each 12V output

No it's not! That "multi" rail crap is there on sticker just to satisfy those ones falling for that BS minefield.

I was only looking at this to get a handle on the magnitude of the currents involved. I wasn't expecting a sort of Spanish Inquisition .

Anyway, I'm not trying to dispute what you've said, and interesting that some power supplies have dropped the 240VA output protection limit, so I'll revise my wording to "...has a specified limit of 18A...". It makes it an interesting problem for the user, though - if a specification says an output is rated to deliver X Amps, is it wise to ignore it and assume it can deliver more? The jonnyguru test you quote where 30+A is pulled without the PSU shutting down might prove there's no overcurrent protection above 18A, but it doesn't mean the PSU is really designed to deliver that.

 

[weird_dave]

Although I can not say that these rails are or are not somehow electronically separated in the PSU's circuitry somewhere

I'm rather surprised Jonny said that!

edit:
Typo, in a reply that short!