Hacking a BFL Single/25 to 48gh and beyond.....

[lightfoot]

Ok, some pictures of this whole thing.

First, the board with 5 chips added to it, and the extra FETs and parts



Close up of the left-side FETs. Remember that the board sides are mirror images of each other.



Good shot of the board when I put on the first choke. Nice solid 1 volt.



And finally a shot of the other FETs and the driver chip during install.



Overall not bad, I'll put a sixth chip on it tomorrow or Tuesday based on how I feel.

[lightfoot]

Ok, here is an interesting question/problem:

Normally when I add chips to a jalapeno I plug it into the "danger supply". This is a stock BFL supply current limited so that if a chip is shorted the board will come up with a "beaconing" 12 volt LED instead of blowing the 1 volt supply to hell. It's worked well (this is how I found that old style chips can short internally) but with the single, there's a problem:

I can't fire the unit up on a jalapeno supply. And even an 8 chip jally will not boot with a jally supply, you need to use a big supply. Which means if you short, you're dead.

What can I do to protect the 1 volt supply lines as I add more chips? I'm confident that I can add factory chips, but I worry a bit that a reballed chip might have a slightly out of line ball. Normally not a problem, but if it shorts the 1 volt line....

Any thoughts?

C

[freddyfarnsworth]

"This is a stock BFL supply current limited"

Limited to how many amps ? use automotive fuses 5 7 10 15 20 25, they are cheap start low on the feed lines. (newer plastic style with legs).
electronic fuses are available in fast blow, but harder to get.

Not sure how BFL does it, prolly a very fast sense digital circuit in the supply somewhere.

EDIT: Just thought about all the horror storys of the older BFL jally supplys so all bets are off on this one ....

[lightfoot]

Gonna watch this closely.  I have a single that pulls 62Gh at about 64C.  Thinking about replacing the feedback resistor like I did in my jalas to put

about 1.1V into the ASICs.  Wondering if the mosfets can take this though. With the jalas is was a marked improvement with every one. +1Gh and

Ok, I'm ordering the resistors. 0603 size, 8.2k. Question: Do you know which resistors to do on the single board? I'm sure there is one per side, haven't tried tracing them out yet. The MOSFets on the Singles are more durable; they should handle it.

C

[helipotte]

Look for a 7K resistor between two ASICs.  On the jalapeno it is R11. Not completely sure on the resistor with singles but the ASIC layout looks the same as a Jala.

I am going to mod one of my singles this week with a 7.5K resistor.  The one I am changing the resistor in only does about 57Ghs with stock firmware.  Hope this

puts it over 62Ghs.  Will post results.

[Goredox]

To check the 1 volt, can't you check the resistance to ground of the 1 volt supply like you could on a Jally???

[lightfoot]

To check the 1 volt, can't you check the resistance to ground of the 1 volt supply like you could on a Jally???

You know, I didn't think about that. If I could find the 3.3 volt test point I could also find that as well (3.3 is shared between sides). But I seem to recall a loaded jally has a very low resistance to ground. What would it be, probably under 1 ohm.

I'll do that.

[lightfoot]

And in the meantime, it looks like I just bought a 100% BROKEN single. So now I'll have another thing to figure out and keep me busy...

C

[helipotte]

To check the 1 volt, can't you check the resistance to ground of the 1 volt supply like you could on a Jally???

You know, I didn't think about that. If I could find the 3.3 volt test point I could also find that as well (3.3 is shared between sides). But I seem to recall a loaded jally has a very low resistance to ground. What would it be, probably under 1 ohm.

I'll do that.

Each side of one of my singles the 1V line measures 14 ohms to ground.  This one has the revision B chips.  Maybe the revision A chips are lower?  I always check the

1V line after the resistor mod to make sure I did not bridge it to ground.

Also, installing a 7.5K resistor was very successful.  Gained 3Ghs (57-60) and error rate went down to around .34% (was 1.9% at 57Ghs).  Temps went up 4 Degrees

though.   Still this mod seems to be a very easy way to push an underperforming single/jala up to speed.

[bcp19]

To check the 1 volt, can't you check the resistance to ground of the 1 volt supply like you could on a Jally???

You know, I didn't think about that. If I could find the 3.3 volt test point I could also find that as well (3.3 is shared between sides). But I seem to recall a loaded jally has a very low resistance to ground. What would it be, probably under 1 ohm.

I'll do that.

Doesn't the ATMEL chip need 3.3V?  I seem to remember seeing something when I flashed my friends LS about 3.3V on the program... Also, looking at my little, isn't that gold rectangle a 1.0V test point?

[_Miracle]

I've had much restraint in not cracking my case open. It's a 50 hashing closer to 48. There was a fried one on ebay that looked interesting to tinker with.
I'll just watch you and learn for awhile.

[lightfoot]

The one that looks like a metallic woodchuck was chewing on the side? Yeah, I picked that one up to check out. I'm wondering if the guy fried the FTDI chip, when the USB system won't talk that is usually it. Fortunately I have a spare chip here, so I could get it working again if that's it. Otherwise I'll have a ton of chips to re-ball.

In the meantime I put a reballed chip on my unit here. The concept of using resistance was useless, as the resistance between +1 and ground is basically 0 ohms. Which makes sense if you think that 8 chips can pull over 120 amps thus R=e/i or 1/120=.0083 ohms.

However it's working, and with 14 chips populated I'm cruising at 49gh and 76c. I did try running one fan pointing up and the other down, but I could hear one of the fans "singing" inside which means it was stalling, and temps went over 81c even with the output fan running at full speed. With both fans pointing up, the temps are 76, no stall, even with the resistor in series with the output fan to quiet it down.

On to the next thing. I'll check out this oddball single and see if it can be hacked too. Never a dull moment....

C

[_Miracle]

"The one that looks like a metallic woodchuck was chewing on the side?"

YES! Had he never heard of a multimaster?
It's best in your hands than mine.

[lightfoot]

"The one that looks like a metallic woodchuck was chewing on the side?"

YES! Had he never heard of a multimaster?
It's best in your hands than mine.

I'll post pictures and funny commentary, don't worry. Maybe it was shards of metal or something. But to be honest the problem was probably that he didn't have the bottom plate in the little slots in the ends. It's um.... made that way.

C

[southerngentuk]

Also, installing a 7.5K resistor was very successful.  Gained 3Ghs (57-60) and error rate went down to around .34% (was 1.9% at 57Ghs).  Temps went up 4 Degrees

though.   Still this mod seems to be a very easy way to push an underperforming single/jala up to speed.

This sounds g8, any chance of a pic identifying that resistor.

[helipotte]

Also, installing a 7.5K resistor was very successful.  Gained 3Ghs (57-60) and error rate went down to around .34% (was 1.9% at 57Ghs).  Temps went up 4 Degrees

though.  Still this mod seems to be a very easy way to push an underperforming single/jala up to speed.

This sounds g8, any chance of a pic identifying that resistor.

R36 and R11 I think.  Will be between two of the ASICs. Should have 828 on the top.  It will be a 7K resistor.

[lightfoot]

Very interesting thing here:

Last night I added another chip to this single: It now has six chips on the right, 8 on the left. I fiddled with the fans a bit, then put everything back to stock condition (both inside ones pointing up, etc) and it hashes at 48-49gh constant with a 76-77c temp (as opposed to 75 before)

However it's now running with 2% errors instead of <1%. And in looking at the stats the errors are coming from the right side, but from all sorts of chips and *not* simply the one that I put in.

Now, my 1 volt supply is missing about 4 of the filtering capacitors because I stole them from a jalapeno, but I'm wondering if this is a filtering issue on the 1 volt line. A bit more noise coming into the chips causing increased random errors (my code shuts down bad cores to save on heat, so what remains are truly random errors). It's running fine and the overall effect on hashing is minimal (<1gh) but still something odd to note.

I'll see about ordering some more capacitors and putting them on the next time I get some chips to try out 7 chips on the unit.

C

[lightfoot]

Update: I did switch the heat sinks so that the best one (the old style BFL with the heat pipes) is on the 8 chips at the exhaust end of the unit and the newer one (black, fins facing properly) is on the intake side. Also put heat sinks under the board on the center FETs (6 chips starts the warming trend) and it's hashing 49gh at 72c. Very good, stable.

I also put another capacitor on the board's 1 volt supply, that was hard as those open capacitor pads are near the last chip which is on there with normal lead solder (which has a significantly lower melting point than ROHS stuff) which made adding the component dicey. 425c for 40 seconds, air angled as much as possible away from the chip. Hashing errors are down from 2.2 to 1.8, most all the errors seem to be on the last chip added, so it might be a bit of a clunk (or took a bit of damage from the reballing)

Overall things are working well otherwise, I'm going to wait for the chipmunk-chewed single to come in and continue research on that.

C

[lightfoot]

Update: Today I took apart one of the single/30 power supplies because although it seemed to provide power it did not light the LED. Fan would run. Found that although the power supply is a pretty solid design (and the neutral goes to GROUND as it should) I did figure out why the LED wasn't working; cold solder joints. Looked soldered, but wiggling the LED would cause it to light.

Used soldering iron to reflow solder, now works fine. Also touched up the power cable lines as well, I wonder if that is the problem with some of these supplies; the wires run in parallel, but if one of them was badly soldered by the power supply company then it would go high which would result in a lot more resistance on the remaining wires, and lower power going to the single. Which would either shut down chips or keep them from coming live on startup.

Interesting.

[freddyfarnsworth]

Sounds fun did you ever get the rest of the caps from mouser or want to see if it makes any difference on your 1v supply issue.

edit: saw the chewed on one, thought he wanted to much as it looked like he arced the power input area pretty bad.