Chip's power consumption is <6.6W/Ghs
ATX power loses(82%), DC/DC power loses(87%), Fan(5~20w), etc
So
650W * 0.82 * 0.87 - 20W =
443W, About right.
so an equivalent bfl number would be something like:
60w / .87 / .82 + 2W (fan) + 5W (host) = 91 W (host can be discounted for many since it can run several units)
or "bfl's worst case" 60W * 1.2 = 72W ===> 72w / .87 / .82 + 2W (fan) + 5W (host) = ~108W
BFL's single uses 72W
at the wall, there is no ATX psu or DC/DC or extra for fans etc, just the host usage to be added. All speculating of course coz it ain't alive yet, but it is definitely no 620W, or 108W.
@jgarzik, have you ran it without the ups to see if that isn't messing up? Checked/reseated internal connections, it has been on a long and bumpy ride?
There is an onboard DC/DC module that converts the 12V to 1.2V or whatever core voltage the chips are running at, along with whatever they're using for IO.
We have been quoting our power as power at the wall, not the chip power. If we want to talk chip power, it's .8w GH, so using the Avalon system of measure, 48w per Single, or 24w for a Little Single.
Even if we have worst case scenario, it's 1.2w per GH and that gives you 72w via the Avalon measuring system and probably somewhere around 80w - 85w at the wall. We are not expecting 1.2w GH/s, that is just the upper limit of the power draw per chip under the worst possible conditions.
Earlier Josh had posted that they were aiming for 0.8J/GH and that the possible worst case was 1.2J/GH. I'm a little confused by this too as BFL has always said that 60W would be at the wall with the included power adapter, but if 0.8J/GH is just the chips you will have to factor in both the onboard convertor losses, other draws (microcontroller, fan, etc) and the brick PSU to get the final power draw.
Hopefully we'll know the answer in a month's time.