Topic: Specs on BFL Single power adapter? - page 5. (Read 13075 times)

Thanks, that is very helpful.  It makes planning for efficiency a much simpler task, certainly less troublesome.

The 50% rule is now more the 50% myth.  At one time PSU had a very "steep" efficiency curve.  It might be 70% (or worse) efficient at 20% and 80% but 85% efficient at 50%.  This allowed the PSU supplier to cheaply sell an "85% efficient" unit but only in a very narrow band.  Worse many didn't provide efficiency curves.  The "buy a PSU to load it at 50%" was a guestimate to get best efficiency.

"80Plus" and hardware review websites changed all that.  80Plus cert requires 80% efficiency at 20%, 50%, 80% 100% of peak load.  The higher ones (like Gold) require even more.  Hardware review websites now stress test most major brands and efficiency curves are available on 80Plus website for all certified power supplies.  So the curves are now much flatter.

Today I have no fears running a high quality unit (like Seasonic) at 100% load 24/7 and even with cheaper units I would be comfortable loading them to 80% 24/7.  If a unit can't handle that it is junk and you likely shouldn't be buying it anyways (poor efficiency, poor regulation, etc).  Spend $20 more and get a higher quality unit.

TL/DR:  The 50% rule hasn't applied for almost a decade now.  It is more an urban legend now.  The difference between 100% load and 50% load is often less than 2% efficiency.

I will see what I can do on Mon or Tues.  I think I have a spare adapter on my desk at work.

This would actually be incredibly useful.  Any decent ($60+) clamp meter should be able to accurately measure DC current.

One would simply need to cut the cable right down the middle.  We are talking about the cable between the PSU brick and the barrel connector.  With clamp meter no need to cut the insulation you just need to separate the 12V and DC wires.  You just need to separate the two wires otherwise they current in opposite directions cancels out the meter (it will read 0 amps which is correct).  Put only the 12V wire in the clamp and measure current under load.

You might be able to convince BFL to buy a quantity of them as well, if the minimum order is like 1000 or 2000 or something.

Before we get ahead of ourselves.

I asked the OEM to price out 1 PCIe 6pin to 3x barrel connector.  Some have asked for 3x adapter, some for 1x or 2x.  Given need for a large unit size I figured a single design would be best.

It is entirely possible the OEM will come back with some insane unit requirement (like 10,000).  I don't want to get anyone's hopes up.  I talked to superviser in person, explained the situation and the need for a small run.  He said he would like me know by Tues.  I will let everyone know when I know.

No need for anyone else to put in requests.  Once I find out min order and unit pricing if it is viable I will start another thread to gauge interest.

No one has measured actual power consumption as far as I know.  It would involve tearing apart the current power cable, and no one has been willing to do that yet.  I might see what I can get taken care of as far as that goes, I think I have a spare around somewhere. But my unit is downtown and I'm at home, so it'll have to wait until next week.

I believe Gigavps knew what he was talking about when he specified 83W per device.

The power draw of the FPGA can be easily measured using a clamp meter or by feeding it from a well-known PSU you have the efficiency charts for and looking at the kill-a-watt readings.

Running the FPGAs with an ATX PSU has another benefit: the ATX spec mandates very tight voltage regulation - probably much tighter than any wall wart is capable of.
The cleaner the power, the less load is being put on the onboard voltage regulators.

PS. When BFL issued a statement regarding the power usage they were promising 19,3 watts IIRC

Maybe I missed the information somewhere.

Does someone already measured the correct power-consumption of the BFL single ONLY (without the included power-supply)?
Or does even BFL did a statement about this?

All I've read is that the include power-supply is getting quite warm/hot, which is at a total power-consumption (BFL single + power supply) of ~80Watt a clear sign that the included power supply has massive power loos because of a very bad efficiency of the power supply. I also read BFL's advice to better use a higher efficiency power supply to support more BFL singles in a cluster.

Maybe the BFL single for itself has only ~55W ?
Maybe there is also a little difference between the different revisions?

If the power efficiency of the power supply is really so bad, it would maybe make sense if BFL offers an option to exclude the power supply. Beside of a small discount it should make the shipping especially to NON-US destination cheaper. Or they should include a professional adapter cable to use the BFL single with a ATX power supply.

Couldn't do the measurement for myself already, but I will do it if I will receive my 2 units (rev 2)  (they should be shipped end of next week). 

While 50% provides the best efficiency, the difference in efficiency between 50% and 80% is minimal on a Gold rated PSU - perhaps 1-2% difference. I wouldn't want to load a PSU more than 80% if you are running continuously, with the exception being if you got some efficient server power supplies that are relatively recent - since they are much more durable than most consumer stuff.

The manufacturer promises 83 amps - that's 996 watts at the +12V output.
As DAT rightly noticed, this is a single rail unit so no additional rail topology issues for you to overcome.

HARDOCP were able to squeeze 100% load out of the unit at 84.9% efficiency (120V AC voltage) but while doing so they measured ripple in excess of 130mV at the 12V output. Temperature delta at exhaust was 20°C.
When using 100V AC input voltage, the ripple rose to an ungainly 390mV - over three times more than that ATX spec allows (120mV max) and six times worse than the Superflower Golden-Green-based Rosewill Lightning 1300W (available for $250 from NewEgg).

The design was clearly beyond its comfort zone when delivering 100% of the advertised load.
Your total load, however, should never exceed 990W (11 singles at 90W). Typically it would be closer to 910W.

You should be able to get away with that PSU but I'm not entirely comfortable recommending it as it's an ATNG-made device - this particular OEM doesn't exactly have a history of stellar-quality builds.
It's a bit of a wild card and it's very hard for me to estimate the longevity under 910W+ load 24/7.
Personally, I'd go with a stronger design - although the price is tempting remember that the FPGAs are very expensive and the $100 saved on the PSU doesn't justify the risk of destroying your singles in a catastrophic PSU failure.


DAT, is this the picture you were looking for? I'm glad to be of help.

I'm not a power supply expert, but I've heard 50-60% (or lower) is ideal, but up to 80% is still ok.

I'm starting with 2, but I might eventually get up to 10.  I'm thinking a 1250 watt PSU should be fine for 10 (64% load), and would probably even be ok for 12 (77%).

It's funny how stuff made in Asia is often cheaper in the U.S. than it is in Asia.  It's cheaper to buy it in the U.S. and have it shipped back here (Asia -> U.S. -> Asia).

Thanks, and the rule of thumb for maximum efficiency for PS load would be about 50%? Or is there a much broader range, say 30-70% or 35-65% or something else? I see that some people want to load each PS closer to its rated capacity, that doesn't sound good(optimal).

Theoretically you could have one single per pair of +12v wires, so 3 per 6-pin connector and 4 on the EPS12V connector. Although I didn't look up the part numbers to fit the EPS12V connector.

Icarus uses the same connector as BFL, so they would be able to coexist on the same PSU. Same with any device that uses a 2.5x5.5mm DC barrel plug and draws 75 watts or less.

Damn, I wish I could have that kind of prices on this side of our planet 

I'd be interested in some, too.  Two mini-plugs per PCI-E adapter would be nice.

I would be interested in a few too.

Would each adapter be designed for one, two or three singles?

Hopefully it pans out cause I for sure would buy 10+ minimum.