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

[lightfoot]

Ok, I've pretty much taken the Jalapenos to their max limit (32gh, with 6 added chips and a lot of cooling) so now it's time to take on the singles. Much bigger board, has possibilities, but also has some interesting limitations.

My goal is to bring a 25gh single up a bit power-wise, maybe to 48-50gh or so. The unit I will be working with here is a standard single/25 which was bought by me running at 23.5gh tops. Which was a bit slow. In checking it out I found the chips ran slow, and one of the chips was missing from BFGMiner's point of view.

Update: Added two sets of FETs, driver, capacitors, and 5 chips. Hashing very nicely at 48gh and temps of 72c with a speed reducing resistor on the output fan.

If you're curious about the other things I do, check out my jalapeno thread at:

https://bitcointalk.org/index.php?topic=336782.0;topicseen

Here we go....

[1714877457]

1714877457

[1714877457]

1714877457

[1714877457]

1714877457

[1714877457]

1714877457

[1714877457]

1714877457

[lightfoot]

Reserved for thoughts and notes.

First thought: The Single board has some more beefy components than the older little_single. I wonder if early singles had the same components and they did an upgrade. I think they did; these FET drivers seem to be much more robust than the ones on the jalapenos. With two full sets of FETs it is almost overkill.

Also cooling is an interesting thing. I don't believe in running the singles open; what I have been doing is putting a speed limiting resostor on the output fan, adding a Noctua 120mm fan on the intake, and right now I'm running two fans on the heat sinks inside. One is blowing air down, the other up.

My thinking is that this take air from the outside, runs it down and to the left over the first sink and the FETs, then continues to run the air up through the second heat sink over the second set of FETs, then out of the unit. Nice wave form, which in my mind is better than sucking air down for the first one, then trying to suck what air is left again down for the second one. Reducing airflow is bad. The result is a nice 73c board temp and warm air coming out the end of it without a whole lot of noise.

I'm currently at 5 chips, might go to a max of six or seven chips because I have a pair that need reballing. 7 is about the limit anyway, going to 8 is a whole nother ballgame on the jallies, so why stress this unit for 4 more gh?

Edit: Went to six chips, 49gh speed, temps of 73c. That's about as far as I need to go; to get to 55-56gh is not really worth it since I have another single to boost, and power/cooling issues tend to snowball at 7-8 chips per side. Overall though a complete success.

C

[lightfoot]

First thoughts:

The Single I got was pretty stock. From what I have noticed, the first generation singles had a nasty habit of blowing up the FET drivers. A number of people have posted pictures of the results and they are not pretty. Which is not good. However the reason for this I think is not enough heat sinking on the chips. So when I flashed my unit to stock 1.2.9 software I made sure to include the following heat sinks on the bottom of the board:



Putting these sinks on gives some bottom cooling to the FET chips which I think will help the life.

Next up is to think about how to boost this unit. The normal way to do it is to add chips, and if your board is like this then you can do it pretty easily:



Note how this board has 8 chips to the left, and a full complement of components on the right. All you need to do (I think) is put chips on, as the 1 volt supply line is ready to go.

However if you're like me your board is more like this:



It is going to be a bit more of a problem.

Note how some components are missing. Especially the big capacitors, the smaller surface mount caps, and of course the FETs. And the FET driver. However the supporting small chips all appear to be there, so if we could mount the other stuff we would be in business. Equally important, the solder is there for the components, so it might be fairly non-impossible to add them. Some of the parts are kind of unique, but oddly enough I have a blown up jally sitting here.....

More to follow as I start working on this. You might be able to add a single chip on the right side, but that will probably stress the FETs out beyond their capacity (ie: They go boom). Edit: You can't. The 1 volt supplies on a single board are isolated from each other; the one on the right runs the right chips and so forth. Therefore there is no 1 volt power unless you were to jumper the pads which would be beyond stupid. :-)

More in a bit.

C

[lightfoot]

Thoughts after looking at the Jalapeno schematics:

I'm guessing the Single schematics are quite similar to the jalapeno, at least in theory. And according to those schematics, it looks like the 1 volt power supply is screaminigly simple: You have a bank of six FETs being gated by a FET driver that picks it's feedback up from resistors and capacitors on the board. All the big caps, and other things are nothing more than filter units, to keep the ripple to a reasonable level.

Which is interesting. The two coils operate one to a set of six FETs; the big capacitors that are flat filter the 1 volt line, and the big capacitors that are round filter the 12 volt line. The chokes are the final stage of the output; I'm guessing they absorb any transients on the line as a result of the switching.

So in theory I should be able to build a quick and dirty 1 volt supply that is isolated from the board by simply putting on the FET driver chip (the 1850), two FETs (one push, one pull), and get a rather noisy 1 volt supply. I still need a filter cap, and if I put the coil on it will hook into the main 1 volt power bus but then I would not be able to measure it in isolation, so I might need to hot wire in a capacitor on the FET side of the coil. Then if I see 1 volt on that side I know the rest of the parts are there. If not, then I'm kind of boned for awhile as not all the supporting tech is in place.

I'll give it a shot. One set of FETs, and a second 1850. Question: Do the FETs have to be different types? One I was able to find at ease at Mouser, the other one was more difficult and I may have the wrong part.

C

[lightfoot]

Found the other one at Digikey. Not quite sure why they did two totally different models, but I'll give it a go and wait a few days for the chips. I do have the ones from this trashed jally unit here, but since the power supplied blew up I have no idea what kind of shape it is in. Then again I could test them with a 2 volt power supply and a light bulb circuit like I do with my IGBTs; hook it up with the light bulb between source and drain, (a LED would be safer actually) then touch source to gate and the light should go on. Drain to gate should keep the light off. Or maybe you don't need to do that with FETs; with IGBTs the residual charge on the gate is enough to keep it on.

C

[lightfoot]

And... I took apart the box again and took a look at the existing FETs. Good thing I did because this board has some differences.

First, the driver is not an 1850; it's an ADP 1877. Looks to be a higher power version, which is good since those are taxed. Likewise the FETs are IRFH 5250's and what appear to be 16404's from TI. Both heavier duty components than the little singles.

I also verified that the chokes are the gateways between the power supplies and the 1 volt main lines. So I'm going to order six of each type of FET, and a controller chip. I'll get the 4.7uf electrolytics from my dead jally, and will get the 470uf ones as well, although I think this board has plenty of them.

Then I will try adding 4 chips and see what happens. It's going to be slower going; I have to map out the chips by adding them one at a time and then testing, which requires a lot of heat sinking. Maybe I'll get myself a good temporary sink to run for a few minutes for testing each chip. Once I have 4 I can put on an AL heat sink or something like that.

TO be honest I was temped to toss a chip on there and see what happens. But at the same time I knew I couldn't sink it, so I will wait...

In the meantime, what is the part number for the PCI-e sockets and the fan sockets on this board? I'm probably going to need another fan, and more power input on the 12 volt side rather soon. :-)

[helipotte]

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

lower errors but higher temps due to higher clocks.  They only have 2/3 chips.  Do not really want to try this with the already strained DC-DC stage

in the single.

BTW, thanks Lightfoot, for all your effort and information.

[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

lower errors but higher temps due to higher clocks.  They only have 2/3 chips.  Do not really want to try this with the already strained DC-DC stage

in the single.

BTW, thanks Lightfoot, for all your effort and information.

Hm. Does boosting the voltage boost the temps of the FETs on the jalapeno? I wonder because it's usually amps that cause the heat, and if you drive the chips at a higher voltage then the amps should go down or stay the same (maybe). I could see the chip temps going up more, but do the FETs?

I think if I tried this on my 8 chip jalapeno it would catch fire.

[helipotte]

I don't think increasing the voltage increases the heat.  Increasing the voltage allows the firmware to bring up more engines at faster clock speeds

and this is what increases the workload on the FETs.  My theory anyway.  Before I did this to one of my 3 chip jalas I noticed that only

about 39 of the engines where online and most of the chips ran somewhere around 275Mhz.  After I installed the 8.2K resistor, All 48 engines

where running and the average clock speed was about 285.  This jala hashes at about 13.1Ghs with a temp of around 42C.  The FETs aren't sinked

and only feel warm to touch after 24Hrs hashing.  8.2K sets the DC-DC to around 1.08-1.1V.  Varies per board, all 8 of mine are a little different.


I do wish BFL had tied the 1V to some type of control logic that the firmware could control.  Maybe automatically increase it when conditions

are ideal or when errors are high.  Maybe even make it user adjustable but stipulate this voids your warranty.  Kind of like motherboards do with

cpu voltage.  I can see how a setting like this could be a self-destruct button for an uninformed user/owner.

[southerngentuk]

Can't wait to see where this goes   

[freddyfarnsworth]

Can't wait to see where this goes  

Also watching

Adding info on PCIE headers plugs ect.
Before ordering make sure that the peg holes match a GPU connector. Been much confusion on that due to patent on PCIE Express by STB I believe it was. Aftermarket molex are called "something" JR.

http://www.frozencpu.com/cat/l3/g44/c155/s648/list/p1/DIYMod_Parts-Connectors_Pins-6-Pin_PCI-E_Connectors-Page1.html

https://bitcointalk.org/index.php?topic=379677.60

http://www.molex.com/molex/products/datasheet.jsp?part=active/0455590002_CRIMP_HOUSINGS.xml

finally to this:

http://www.molex.com/molex/products/datasheet.jsp?part=active/0455580003_PCB_HEADERS.xml

[lightfoot]

Thank you for the Molex information. I think a stock corsair power supply will be able to handle 12 chips which is what I will go for this weekend; if it works I'll order some edge connectors and start moving to 16gh or so.

Speaking of which, I need another fan. Oh great, totally forgot that....

C

[freddyfarnsworth]

Thank you for the Molex information. I think a stock corsair power supply will be able to handle 12 chips which is what I will go for this weekend; if it works I'll order some edge connectors and start moving to 16gh or so.

Speaking of which, I need another fan. Oh great, totally forgot that....

C

Fan : 120x120x25 is size Delta is brand gateway servers use em.

"Few" meaning three  leads on em...12v 4000rpm 45dba 130+ CFM 3wire

AFB1212SH
http://www.ebay.com/itm/Gateway-960-8006323-Delta-AFB1212SH-Back-Case-Chassis-Fan-30-Day-Warranty-/170822936596?pt=US_Computer_Case_Fans&hash=item27c5d72814

http://www.ebay.com/itm/Gateway-Delta-AFB1212SH-Back-Case-Chassis-Fan-/281046185168?pt=US_Server_Fans_Cooling_Systems&hash=item416fa874d0

http://www.ebay.com/itm/Delta-Elec-AFB0812SH-12V-DC-Brushless-Fan-/350343725894?pt=LH_DefaultDomain_0&hash=item51921d0346

[lightfoot]

Updates:

1) This is not easy :-)

2) Have the chip, 6 fets on, no choke, one capacitor on the 1 volt line. Note that oddly enough the 1 volt connections between the two sets of 8 chips are *NOT* connected. Note also that if you try to power with a 1a bench supply you will think you're dead. Man up and plug it into the big BFL supply. Either it works or it blows.

3) It seems to work. I see 2.5 volts at the choke inlet, remember choke is not on yet. Shorted the choke pads quickly and saw a nice solid 1.018 at the 1 volt test pad. Other 8 chips appear to come up on the other side.

Next step: Heat board, put on 5 more 1 volt capacitors, then try it out. I should get a solid 1 volt, in which case I'll check to see if I can put a chip on the board.

C

[lightfoot]

Well, that was a fun job. Microsummary: Ten chips on the board total, hashing at 35+gh, not bad. I really can do anything with anything and a Mouser catalog.

Quick details:

1) I don't have my darn heat sinks in, so I had to use some old small ones. Yuck.

2) I need to make a new back plate or something for this. I have the right heat sink, but it's balance on two chips right now, not great.

3) The 1 volt supply came up rock-solid. It must do the voltage sensing after the choke coil. Ok, makes sense.

4) Running one chip I heard a buzzing on the board. Added 5 12 volt caps on the right side. Added another chip buzzing still there. Turns out it was the POS BFL power supply buzzing probably because it's overloaded with 10 chips.

5) It really is working, maybe tomorrow I'll add the other six FETs to share the load. Which means removing that BFL heat sink on the left set of chips, not sure how to do that. Heat? Chisel? Putty knife?

6) Technically I have to remember that the 1 volt supply on the right here is handling the chip load with 1/2 of the FETs. So right now it's running equal to a 4 chip jalapeno.

7) I was too lame to put a fan on the second heat sink (it's the super-good one) so I just put a 120mm fan on the end blowing in. With the case on it should keep things reasonably cool, we'll see.

So now I have this thing running along with the Chili on one 500 watt supply. The other 500 watt is running an 8 chip, 2 7 chips, a five chip, and a four chip. Hm, it may be overloaded. Great... Edit: Nope, it's pulling 470 watts from the wall which is fine. So you can run a max of 32 chips per 500 watt supply. I'll have to remember that.

Pics tomorrow, have a date tonight.

C

[bcp19]

Well, that was a fun job. Microsummary: Ten chips on the board total, hashing at 35+gh, not bad. I really can do anything with anything and a Mouser catalog.

Quick details:

1) I don't have my darn heat sinks in, so I had to use some old small ones. Yuck.

2) I need to make a new back plate or something for this. I have the right heat sink, but it's balance on two chips right now, not great.

3) The 1 volt supply came up rock-solid. It must do the voltage sensing after the choke coil. Ok, makes sense.

4) Running one chip I heard a buzzing on the board. Added 5 12 volt caps on the right side. Added another chip buzzing still there. Turns out it was the POS BFL power supply buzzing probably because it's overloaded with 10 chips.

5) It really is working, maybe tomorrow I'll add the other six FETs to share the load. Which means removing that BFL heat sink on the left set of chips, not sure how to do that. Heat? Chisel? Putty knife?

6) Technically I have to remember that the 1 volt supply on the right here is handling the chip load with 1/2 of the FETs. So right now it's running equal to a 4 chip jalapeno.

7) I was too lame to put a fan on the second heat sink (it's the super-good one) so I just put a 120mm fan on the end blowing in. With the case on it should keep things reasonably cool, we'll see.

So now I have this thing running along with the Chili on one 500 watt supply. The other 500 watt is running an 8 chip, 2 7 chips, a five chip, and a four chip. Hm, it may be overloaded. Great...

Pics tomorrow, have a date tonight.

C

Sounds like you are having fun so far, nice accomplishment!  I do want to ask though... where do you get those square heatsinks you use on the back of the board?  I've had no luck finding any in my searches.  I also noticed you only put heatsinks on 1 of the 2 sets of FETs and on the ADP.  Any reasoning on that?

[valkir]

bump to follow! 

[lightfoot]

The heat sinks are Enzotech's. Nice stuff actually, waiting for another batch to come in.

Right now working on adding chips 3 and 4 to the blind corners, will then add the FETs and go from there.

Edit: Chip 3 on and working, doing chip 4 and the six FETs. Wonder if it will read 1 volt even without choke 2 on there (ie: does the chip have a single point of reference for voltage sensing on the rail, probably)

C

[bcp19]

The heat sinks are Enzotech's. Nice stuff actually, waiting for another batch to come in.

Right now working on adding chips 3 and 4 to the blind corners, will then add the FETs and go from there.

Edit: Chip 3 on and working, doing chip 4 and the six FETs. Wonder if it will read 1 volt even without choke 2 on there (ie: does the chip have a single point of reference for voltage sensing on the rail, probably)

C

Awesome, thanks, I got an 8-pack on order now

[lightfoot]

Ok. In the meantime all together, total of 5 chips on the right side, 8 on the left, both power supplies operational.

They way to test the power supplies for shorts or screw ups is to not install the choke, and check the voltage. If it's 2 volts or so the chip is good, if it's zero or something else you have a problem.

Once I put the choke on the overall voltage on the rail was a rock-solid 1.018 volts. Not bad, not bad.

The unit is now together and hashing at 48gh at 65c. Heat sinks are installed on the center FETs only since those get the least cooling. Oddly enough the stock fan was pointing *up* so I set the second fan to point up as well. I might try reversing it later, however I think the backwash of the fans going down will disrupt the cooling air going across the heat sinks (which it does nicely since the fins are parallel to the main air flow all around)

Edit: I changed it so the first fan points down, then the second up. Reason being I want the air to come in, be forced over the first heat sink, then zip up through the second sink, then out the unit. Keep the air flow smooth as possible, otherwise it bunches and your fan blades stall. Temps are 72c now, and quiet.

I'm still only feeding it with a single molex plug, part of the reason I am not going further on chips. The other reason is I am not feeling well today, so I'm done for the moment. Will post some pics as soon as I can find a USB cable for my phone.

C