I'm a little confused..  if you really meant 5970 you might want to edit the topic and OP

I'd wager you're right

Ah, multi-rail fail.  Appears that PSU has 8 12V rails, each capable of ~300W.  So 1 card per rail (~190W each).

I'd wager you've got 1 rail dedicated to the CPU 4/8pin, 1 to the 24pin ATX, and the other 6 divvied up between peripheral connectors.

As long as you can balance the rails properly, yes.  5 6970s is going to be ~1000W.

Yes, that works fine.  Cablesaurus' original 1x -> 16x extenders were exactly that; 1x -> 1x extenders with the end of the connector cut off to allow the card to seat properly.   I'm currently usisng 4 of those.

You *absolutely must* use powered PCIe extender cables on these PCI -> PCI-e adapters!

The maximum power output on normal PCI slots is 35W (I believe, couldn't find a reference offhand), where PCI-e must supply 75W.

Its a gamble whether or not your card requires that much from the slot, but worst case scenario is the card won't power up.

Get extenders with molex power and save yourself some headache.

Which would be perfectly fine.  You do not need a full 16x connection to mine.  I have 4 cards running well on 1x -> 1x extenders.

[Edit] NM, I see where the OP mentions 1->16x extenders where the illustration indicates otherwise.

I picked up a bunch of momentary NO (normally open) push button switches, a couple of mini breadboards, and some jumper wires from sparkfun.com

They sell male-female jumper wires that fit the motherboard header perfectly, and plug right into the breadboard.  10 seconds later and I've got a 'real' power switch

It's not the cheapest solution, and these parts can probably be found cheaper elsewhere, but I had picked up all this stuff for another project.  The jumper wire is perfect for quick prototyping.

Wires: http://www.sparkfun.com/products/9385
Switch: http://www.sparkfun.com/products/97
Breadboard: http://www.sparkfun.com/products/7916

These guys are more or less in my backyard so I didn't have to pay shipping either, which saved a bit.

If I remember I'll grab a pic of the switch when I get home.

Great find, aluminum tape is so useful

Will definitely be giving this a try.

Since you're talking about shorting, I assume you mean you're using it in a 16x slot.

Does it work in a 1x slot?  It could simply be a bad one.

The 'current' price is the last transaction price.  If that 0.17BTC had sold for $1 instead, you'd have seen the price drop $4.30 until the next, hopefully sane, transaction. 

The price stays relatively even only because a) nobody will buy for more if they don't have to, and b) it doesn't make sense to sell for significantly less (unless you're just dumping 'free' or 'cheap' coins without regard to price)

If you watch mtgoxlive.com, you'll see that kind of thing happening all day long.

Carefully re-read the first sentence of Cluster2k's post.

If you have enough transaction volume, the fees decrease enough to make those transactions profitable.

There are lots of factors in play that determine your max stable clocks.  On one of my boards, anything over 960 causes hard locks.  An identical board runs the same cards at 1010/320 with rock solid stability, and could probably go higher.  Yet another different board tops at 970.

OP needs to give more info if he wants any useful answers.   

Whoa, isn't 1.63V waaay high?  Maybe you mean 1.163V?

Heh, now this is getting a little off topic, but the OP seems to have disappeared anyway.

I do think this deserves it's own thread, a nice fat DIY watercooling guide would be amazing

That's quite a large order to gamble, especially when even the major retailers have had trouble finding these cards.

Replace all the 6850/6870 with 6950 / 6970 and that all makes a lot more sense.

I nearly ordered 6850s instead of 6950s for the same mistake.

Not exactly - as of this post we're at block 145669, and the size of the block reward doesn't halve until 210,000 blocks.

There are 64,331 blocks until we reach that point, and difficulty adjustments attempt to keep blocks generating at the rate of 1 every 10 minutes.

64,331 * 10 min = 643,310 minutes until that point, or

10,721 hours
446.74 days
~14.9 months

There's well over a year before you have to worry about that, though at current exchange rates ($4.87/BTC) it may take well longer than that to pay off your hardware.

330mh/s is pushing a 5830 pretty hard.  If your temps are hitting ~90C that could easily be thermal throttle.

Bitcoins, silly!

This is back-seat-driver, off the cuff stuff, with-no-experience-in-these-things, but:

The vendor that produced this card likely a) has their FPGA instructions stored in a memory chip, and usually these are re-programmable (aka 'firmware update').  If not, it means they're b) loading the instruction set on the fly, usually as bytecode contained in their driver software.

If the card happens to have an open-source driver, and you're really lucky, it may be possible to figure out either a) how to update the firmware with your own instruction set, or b) figure out how they load it on the fly, and load your instructions instead.

If you're unlucky, they don't have any open-source driver or equivalent, or the bytecode they use is signed in a particular way to avoid people doing exactly what you're trying to do.  This is more common in, say, a smartphone than an off-the-shelf video recorder card.  In either of those cases, your only choice is to interface with the FPGA directly as EM suggests.

Regardless, it's going to take some work and a lot of trial-and-error to get anywhere with it.  By the time you do, the 10-15mh/s may no longer be worth the power it consumes.