Hacking a BFL Jalapeno to 32GH and beyond....(???)

[lightfoot]

Sending Lightfoot a 25 GH BFL single tomorrow with some PSU's. This unit is underperforming at ~23 GH, so I figured it would be best to get the expert under the hood to tinker and see what he can find, maybe add a few chips into it.

Post pics, tell us what you find. 

Well, I'm starting to look at it, this is my first-ish single (I had one 30gh one, but it was a bit different). However it does exhibit that weird symptom of not starting all the chips if it's warm; when I brought it in from the cold post office and fired it up I got a good 24-25gh. Shut down this evening and rebooted and got 12. Yep, 12. And a bunch of errors I think. Shut down, took to kitchen, put in window so 32 degree air could chill it down, took upstairs, fired up (30c), 25gh.

I wonder if that's the firmware. A quick initial review says yes, this does not have the support chips on the right side like the 30 I looked at, and it does not have heat sinks on all the FETs (only half of them. Odd). It also does not have heat sinks *under* the FETs, which is interesting, but it does have a nice little sink under the CPU.

Need to ask the owner if the sink was taken off or anything like that. It looks like the baffle is not there, nor are half of the standoffs. Odd if this has never been opened, in a way it looks half baked.

Comparing it to my Jallies, it's going to have some heat dissipation problems. My 7 chip jallies hold at 27gh, but aren't this odd. I wonder if there is something in the single code that is disabling the chips on startup when hot. Odd....

[bcp19]

Sending Lightfoot a 25 GH BFL single tomorrow with some PSU's. This unit is underperforming at ~23 GH, so I figured it would be best to get the expert under the hood to tinker and see what he can find, maybe add a few chips into it.

Post pics, tell us what you find. 

Well, I'm starting to look at it, this is my first-ish single (I had one 30gh one, but it was a bit different). However it does exhibit that weird symptom of not starting all the chips if it's warm; when I brought it in from the cold post office and fired it up I got a good 24-25gh. Shut down this evening and rebooted and got 12. Yep, 12. And a bunch of errors I think. Shut down, took to kitchen, put in window so 32 degree air could chill it down, took upstairs, fired up (30c), 25gh.

I wonder if that's the firmware. A quick initial review says yes, this does not have the support chips on the right side like the 30 I looked at, and it does not have heat sinks on all the FETs (only half of them. Odd). It also does not have heat sinks *under* the FETs, which is interesting, but it does have a nice little sink under the CPU.

Need to ask the owner if the sink was taken off or anything like that. It looks like the baffle is not there, nor are half of the standoffs. Odd if this has never been opened, in a way it looks half baked.

Comparing it to my Jallies, it's going to have some heat dissipation problems. My 7 chip jallies hold at 27gh, but aren't this odd. I wonder if there is something in the single code that is disabling the chips on startup when hot. Odd....

That's an interesting point to consider, but it sounds weird to program a firmware to disable thing just because they are hot.  I know there's a schematic of the Jalapeno board out there, but never heard about one for thier 'long boards'.  Is there one?  I'd be interested to learn if this is indeed software or hardware related, but I honestly can't come up with a component that would cause a slow down just cause it is hot.

[lightfoot]

That's an interesting point to consider, but it sounds weird to program a firmware to disable thing just because they are hot.  I know there's a schematic of the Jalapeno board out there, but never heard about one for thier 'long boards'.  Is there one?  I'd be interested to learn if this is indeed software or hardware related, but I honestly can't come up with a component that would cause a slow down just cause it is hot.

Maybe, but BFL has some pretty smart software to go through the chips and enable/disable the engines based on errors. Maybe they have something else going on in their firmwares, remember that the first generation singles had a bad tendency to blow up the FETs so they put the 30/60gh limiters in the 1.2.8 and 1.2.9 code. Actually made sense.

A quick review of the board shows this to be very different from the earlier model singles I have seen; the heat sink actually makes *sense* as it's a model that

• Covers all the chips
• Is lower profile between chips and fins
• Is oriented so that sir flow through the unit blows *ALONG* the fins instead of butting up *AGAINST* the fins
• Doesn't have that weird baffle
• Has a heat sink on the *bottom* of the board. It's small, but it's the thought that counts

They also seem to understand that the FETs *and* the FET driver chip are heat sensitive; they have a really nice little Heat Zone silk screen on the back at those places. No heat sinks but they did think about it.

Also this has all eight chips, but chip 0f is not there. Odd. Using my powers of having a boatload of jallies I think I can figure out which chip is 0f, assuming they used the same labeling scheme. But why would they change that?

I wish I could have a copy of this code; it will be sad to reformat it. But there are no LEDs on the back to cross-check, so I don't know if that chip is honestly broken, or just idled. However given the exceptionally slow speed of this unit at 7 chips (it hashes at 22gh, my 7 chip jallies hash at 26-27) and the exceptionally low errors, I think it's been seriously de-tuned.

[lightfoot]

Ladies and Gentlemen:

May I present the first 8 chip jalapeno.



And a quick snapshot of the screen to prove it's operation:



Yep, a 30+ Gh Jalapeno.

However, before you go for this, know that the heat load has simply gone beyond insane and into plaid. I have mentioned that going from 6 chips to 7 is an order of magnitude, this is another order of magnitude. Possibly an even larger jump.

It runs so hot that when I fired it up with too much heat sink compound on the water block it blew from 40 to 80 in seconds, then hit the BFG limit I have set (I throttle at 80, the BFL code throttles at 90). There is no chance in heaven this will work with the air cooled heat sinks, no matter how neat the heat pipes are.

Note that it has heat sinks on the top and bottom of the FETs. Note that the water block is oriented lengthwise along the axis of the FETs to get as close to them as possible. And note that the FETs are still rather warm, I'm thinking of extending the heat sinking across the top of the board on the back.

It throws that much heat. The pump is now running at full speed, that's buying me a few degrees.

And note also that it pulls so much current on startup that it failed on the special stock-ish Danger-Jally test supply. The first time I fired it, it found no chips. The second time the LED on the power side went dim, then bright. Had a few bad moments there as I wondered if I shorted the chips.

But after taking that deep breath I plugged it into the smaller of my ATX supplies (the 300 watt one) and up it came with all 8 lights blazing. Bad moment there. :-)

I'd like to thank everyone on this thread for their thoughts, comments, encouragement, and donations. I'm really amazed it works at 8 chips, but it does.

I now see why BFL looked at what it would take, and said "Um, I think we need to redesign". They made the right choice.

And I now have the only 8 chip jalapeno in the wild.
CZ

[arekm]

It goes to the point when it will become easier to immerse entire jalapeno into (non-conductive obviously) coolant. Not sure what is used these days but... mineral oil works.

[eoakland]

That is insane.  congrats man, long time coming.  you have truly maxed out the jally...no more room to add chips.  that's super crazy. 

[freddyfarnsworth]

It goes to the point when it will become easier to immerse entire jalapeno into (non-conductive obviously) coolant. Not sure what is used these days but... mineral oil works.

Use a tank of freon (3M Fluorinert), ala Cray in the old days

Interesting info:
Each cable between the modules was a twisted-pair, cut to a specific length in order to guarantee the signals arrived at precisely the right time and minimize electrical reflection. Each signal produced by the ECL circuitry was a differential pair, so the signals were balanced. This tended to make the demand on the power supply more constant and reduce switching noise. The load on the power supply was so evenly balanced that Cray boasted that the power supply was unregulated. To the power supply, the entire computer system looked like a simple resistor.
 
The high-performance ECL circuitry generated considerable heat, and Cray's designers spent as much effort on the design of the refrigeration system as they did on the rest of the mechanical design. In this case, each circuit board was paired with a second, placed back to back with a sheet of copper between them. The copper sheet conducted heat to the edges of the cage, where liquid Freon running in stainless steel pipes drew it away to the cooling unit below the machine. The first Cray-1 was delayed six months due to problems in the cooling system; lubricant that is normally mixed with the Freon to keep the compressor running would leak through the seals and eventually coat the boards with oil until they shorted out. New welding techniques had to be used to properly seal the tubing. The only patents issued for the Cray-1 computer concerned the cooling system design.

Back to Back 8x8 16chip Jalleys in freon...

[godsdog]

It goes to the point when it will become easier to immerse entire jalapeno into (non-conductive obviously) coolant. Not sure what is used these days but... mineral oil works.

Maybe add a Peltier Cooler in ceramic container to it but that would add 12v..70 watt+ ....

[miaviator]

I'd donate something towards florinet?  What is it called? we used it in the late 90s with liquid nitrogen (and cheap cooling coils) for cooling.

Fluorinert

http://www.3m.com/product/information/Fluorinert-Electronic-Liquid.html

You can submerge the entire device.  At one point we didn't want to pay for liquid nitrogen so we used AC coils or a cheap dehumidifier  to cool the flourinet and recirculate it.  Dump it directly onto some well attached heat-sinks and it'll cool whatever you can throw at it.

[lightfoot]

Yep, that's the stuff they used to cool the Crays. But in order to need that, I'd start working on ways to put the chips on top of each other on the boards.

No, not... yet... :-) Although I am going to the hardware store today for some very long bolts. I'll use those to mount the Sedion cpu cooler to the other 7 chip jally board and bring it to 8 chips. Then I'll put a 7th chip on the 6 chip weird board and make it the air-cooled one.

C

[lightfoot]

In the meantime I bumped the board with 1 temp sensor to 7 chips and elected to use the water cooling. Given that it's running with only a single sensor I don't want to stress it with air cooling alone (stress it. Hah. Ho ho he he he!!!). However it is at a nice 45c and hashing away properly.

So here's the latest recommendations if you want to speed up your system:

2 chips: Flash the unit with the right 1.2.9 firmware. Will cut errors and bring you to 7-8gh if you have good chips.
3 chips: 10gh. Good if you want to run with the whole case on.
4 chips: Most you should run with the stock supply. Need case off, can have the top fan on with the standoffs and BFL lid.
5 chips: Lid needs to come off, run with a different supply. Heat sinks on FETs.
6 chips: Most you can run without bottom cooling. Heat sinks on FET drivers too.
7 chips: Water cooling time. Small water block needed. Corsair 80 water block recommended.
8 chips: Serious water cooling time. Get a Corsair 100 or better.

If anyone would like me to work on their units, please PM me. Chips can be found, and I've gotten to the point where I have not missed a chip yet. So I feel more confident in this, my protocol is good.

Next up: What to do with your SC25/30/60.

On a side note, if you have a real SC30 and want to add chips, drop me a line. I think it's possible to do if the unit has fans on both ends of the case. Check yours and see.

C

[lightfoot]

Update: I cleaned up my mining table, and re-ran all the wires. Resulted in an hour shutdown, always a pain but sometimes necessary.

Found that my two 7 chip units, the 5 chip unit, and my 8 chip chili are pulling 560 watts from my power supply. Given that it's a CX500, I figured it was time to order a new one. Fair enough. I'm also running the 8 chip unit on a second crummy power supply, I'll be replacing that with this new CX500.

That should allow me to balance things a little bit better and give me some room to run test units/jallies in for repairs. Still, 500+ watts is a lot of heat in the room; I have a window open now in the middle of winter.

Next up: Picking up a Raspberry Pi and running that. Also how does one run Ez-miner on an android? I'd like a smaller miner system and one that can run on a cell network as a backup would be nice.

C

[elaramus]

Lightfoot, congrats on making it to 8 chips!!! How many chips have you successfully reballed?

[lightfoot]

Lightfoot, congrats on making it to 8 chips!!! How many chips have you successfully reballed?

Three so far with one failure. It's not easy; I ordered some new soldering iron tips to help with the job. The problem is if you don't do it right and carefully you can tear off a pad on the bottom of the chip when it comes off the screen. Bye bye chip.

8 chips also pulls a lot of power, I am officially overloading my 500 watt corsair. It's handling it, but I have another one on the way to replace this junker 300 watt unit. The corsair does have a 99+% power factor though, very nice.

[drmadison]

PM'd you about this but that cable I had was shipped out last week - should make it there in the next couple of days...hopefully that will help with some power issues

[int03h]

<- Tips Hat. Well done sir!

[synapse]

nice summary, you have inspired me to do a 4 and a 5 chip jally.
or maybe a 3 and a 6 :-)

[lightfoot]

PM'd you about this but that cable I had was shipped out last week - should make it there in the next couple of days...hopefully that will help with some power issues

Thank you for sending it. It's amazing, beautiful,

And almost set my supply on fire. Specifically I plugged it into two 7 chip jallies, and my supply. Ran it for a few hours and checked it. The ATX molex connection was *hot*.

So I pulled one unit off and back onto the hard-wired connections and all is well. Moral seems to be that 7-8 chip units pull an incredible fuck-ton of power, so hard connections are where we need to go.

I'll think about it a bit more over the next few days, but for the smaller jallies it will be perfect. Thank you for sending it to me.

C

[lightfoot]

nice summary, you have inspired me to do a 4 and a 5 chip jally.
or maybe a 3 and a 6 :-)

Go for it! The key *is* heating the bottom of the board up, I have not failed a single unit. The other key is proper placement and fluxing; I spend about 10-15 minutes with a 6x loupe and tweezers getting the chips balanced on the balls from two directions. Makes for a perfect melt with 450c at 60-90 seconds 1/4-1/2 inch above on air speed 2 (really low, more like a ball of hot air than a flow)

I would also say it's better to spread the chips out than pile them on one board. The temps and power issues go up a lot between 4 and 5, and 5-6.

C

[lightfoot]

On a side note, I think I see the issues with the singles; BFL really screwed with the firmware to try and keep them from blowing up. However with the later model heat sinks this is not so much a problem, however you can't just flash them. You need to also add some heat sinks and ensure the airflow is right.

I think I have it down pat here; is anyone interested in the details and such once I get it done? So far I have a 23gh single running at 28+ with no heat problems, but a number of temporary sinks.

C