What is this "external clock" ? Is it provided on the board, or is it just a connector where the customer can provide a clock ? Does every chip have its own external clock, or are they chained ?

"Demonstrated performance of 64GH/s per board" - Does this mean you have a working board with 16 chips, each doing 4GH/sec, or is it an extrapolation from the one test chip ?

"It's just normal variance" is a perfectly fine explanation, as long as you can show that falls in the normal range. The 40-day/50% explanation is a plausible one.

There's nothing wrong with CPPSRB itself, but if I add up all the shares in the block list http://eligius.st/~wizkid057/newstats/blocks.php , I get a total of 29618979908 shares found, with total difficulty of 27904252293. This means average luck over the last 19 months (all of the stats) is only 94%. That seems awfully low to be just a case of bad luck over such a long period.

Any explanation ?

Price/performance would be worse, because of the fixed cost for the power regulator and other components.

What's not open about bitfury ? All the information needed to port other mining software to the Raspberry PI is available.

I'm running the same board now without a heatsink, but just with an old undervolted fan pointed at the top of the board. Board temp is about 60 oC and running fine (41-42 GH/sec).

It's probably a problem with the chip itself. Unfortunately, there's some variation in performance of chips.

In some cases there's room for a little improvement by tweaking individual chip clock rates. I'll make a firmware update for that.

Of course, you're not just paying for a piece of silicon wafer, but also for the months of design work, the cost of the mask set, as well as a risk premium.

I don't know how serious the danger is. The voltage regulator has a current and temperature limit, so it should protect itself. From what I've seen, it just turns itself off when it's not happy anymore. The card will stop mining, but a quick power cycle will fix it. Of course, if this happens when you're not watching, you could lose a lot of hashes, so I wouldn't get it too close.

Of course, increased heat will decrease lifetime of the chips, but how important is that given the increasing difficulty ? A few months at 120% rated capacity may return more than a few years at 100%.

Correct. But if some chips are misbehaving, and you notice a lack of solder on the jumper, you may want to peek under the edge of the chip to see if there's enough solder on the contacts there.

It's possible to fix poorly soldered chip. Take good quality solder tin, and run a fresh bead along the edge of the chip. If you're lucky, it will wet the contacts of the chip. Only attempt this if you know what you're doing and have some wick ready to fix any bridges.

0.8 mm not 8 mm And I prefer not to hand solder 0402 (0603 is okay), but use my reflow oven instead.

The purpose of the solder jumpers is to route the signals around a broken chip. Normally, the communication signals from the Raspberry PI go through each chip to the next one. If a chip is broken, it could prevent communication with the next chip.

With the solder jumpers you could remove a chip (or disconnect it by cutting the copper traces), and then connect each of the two blobs of the solder jumper with some extra solder tin so the chip is skipped, and the software can talk to the next one.

Yes, SJ = Solder Jumper. It doesn't matter if they are covered with solder tin or not, as long as they aren't connected to each other (unless you want to disable that particular chip).

But your lack of solder tin on the jumpers may indicate that the chips don't have enough either.

I do everything with 0.8 mm chisel tip. But I agree with fpgaminer. Use good solder, and make sure it's fresh. Don't let the flux evaporate, and if you do then wipe tip clean and start again. My favorite solder wire for rework is Multicore crystal 511.

I usually take an X-Acto knife with a fresh blade and a metal ruler. Cut across the traces. Shift the ruler a tiny bit, and make a parallel cut. Now, using tip of the blade peel away the little strips of copper between the two cuts.

Powers of two are a little easier to work with, and there's no disadvantage, so why not ?

Because most pools work in powers of two.

I've tried various kinds, but this is Multicore 96SC LF320 AGS88 lead free (Farnell order code 5091100) and so far I like it the best. The problem I have with some kinds of paste is that they are very runny, falling off the squeegee, and slumping when on the board. This one is nice and firm, but not dry.

When you measure them off the board, then R02F = 10k, and R01F = 1.0k or 1.5k.  When you measure them in-circuit, you also measure part of the regulator IC. The resistance will then depend on polarity of your probe, and measuring voltage used by the multimeter.

The 1.1k - 1.3k people are referring to is not an accurate value, but it can be used to judge the effects of the pencil. Also, if you have multiple boards, measuring before/after gives you a reasonable shot at some consistency.

The resistor is in a regulation feedback path, so we don't want long wires that can influence stability of the control loop. Also, potentiometers are notoriously sensitive to wear and dirt.