Hacking BFL Monarchs and servicing them while times are weird.

[lightfoot]

Pricey, but looks like one of the later generation ones. They do make good heaters, quiet and the water cooled radiator warms a room evenly....

[1714539815]

1714539815

[1714539815]

1714539815

[1714539815]

1714539815

[DaveF]

Not from me.

I was going sell mine for the price of shipping, but I was too lazy to get a quote and pack them up.  Sorry about that to whoever that was.

If you still have it and are US based and ever get motivated ping out to me, I'll take it. I had a few and as lightfoot said, dead quiet and throw more even heat then the R4 or an underclocked 1 board S9.

Not real reliable at times and you have to have a separate controller but still a neat piece of mining history.

-Dave

[DaveF]

So os2sam sent these to me (thank you) and one has been working fine with no issues and the other one not so much.
At 1st it was not seen by the PC at all, then it came alive and started working. Then it stopped again and is no longer seen.
Someone a while ago had a link with a bunch of schematics and test points for power. Does anyone have a copy?
Figure I would give it another quick look at before just shelving it.

Thanks,
Dave

[os2sam]

Dave,
What mining software are you using?

Those two miners have different firmware on them so one will work with CGMiner and one won't.  I couldn't get either to work with mainline BFGMiner.  So I had to use the old BFL supplied hacked version of BFG Miner.
Sam

[lightfoot]

I love watching this old thread. Post a pic of the non-working miner, I think if it has two PCI power plugs it's gotta use the old 4.20 software.

[DaveF]

Dave,
What mining software are you using?

Those two miners have different firmware on them so one will work with CGMiner and one won't.  I couldn't get either to work with mainline BFGMiner.  So I had to use the old BFL supplied hacked version of BFG Miner.
Sam

They both worked with the hacked BFG.

This one did not even show up as attached to the PC at first. Then it did, and it mined fine for a while. Then it just disappeared. Checked cables and power and such and all looks good.

Figured I would do a quick voltage test in some spots to see if it's even getting power where it should.

Beyond that, it probably is not worth the time. I picked them up as ulta-quiet 700 watt space heaters

-Dave

[os2sam]

They  did kick out some heat when I was using them.  Had to open a window in the winter in Wisconsin.  Well I hope you get the other one to work too.

[lightfoot]

So...... It's um.... been awhile.

While cleaning out some junk here in the lab I came across a box with my Jalapeno, some Titans, and a pair of Monarchs I bought on Ebay long ago. Everything was rusty, dusty, haven't been run in forever and last were used as space heaters because unlike Antminers they are kind of quiet....

And I figured "Why not, I wonder if they still work?" So I cleared my bench and started looking at all of this stuff.

First was to get the Jalapeno working. This was my 3 chip model, my 8 chip water cooled super Jally and the 7 chip air cooled Jally are upstairs somewhere. But I found a power supply, plugged it in, and fired it up. Fan was unbalanced (broken blades, these used to be tough to get) so I dug out a spare, swapped it, and then had to find BFGMiner.

Not easy: The Github repository seems abandoned and many of the modules needed don't.... seem to be in the right places. Finally found CGMiner, compiled it on a Pi, had the compile totally barf, found that the code was not great and the latest version of gcc doesn't allow all that redefining of variables, found the switch to ignore that, and got it up and running.

Next up, found a pool, connected the Jally and sure enough I'm mining at a blistering 10gh/s. At this rate my .01 minimum payout will take.... forever? Yeah. Forever. Even hitting the minimum share amount is like a couple of times a *day*. Not once a second anymore, maybe I'll just solo mine. Wonder if p2pool is still around.

Found another jally, gave it a whirl, it works too. This was a bog-stock model that I never flashed, changed, or hacked. 4gh speed. Which is about what they came with, I remember when we were flashing like nuts to get them to 5-6gh, then we started adding chips....

The Monarchs..... man, those are a different story. One literally exploded its FETs when plugged in, and the later model went to thermal shutdown pretty much immediately. I'll see if I can get it running, these were a lot more fickle, and my guess is the water coolers are shot, the FETs are insane, all of that stuff....

Do not plug those in.

Next week, I'll take a look at the Titans. Fujicoin anyone :-)

[DaveF]

So...... It's um.... been awhile.
....
The Monarchs..... man, those are a different story. One literally exploded its FETs when plugged in, and the later model went to thermal shutdown pretty much immediately. I'll see if I can get it running, these were a lot more fickle, and my guess is the water coolers are shot, the FETs are insane, all of that stuff....

Next week, I'll take a look at the Titans. Fujicoin anyone :-)

I know you know what you are doing but since I have have played with some of these in the last few years.
For the one that shutdown.

Do you see 'drip' from the coolers anyplace? And the obligatory did you have all the molex plugs in?
At least on my monarchs I had some that had 3 plugs and others with 2 and others with 4 and you needed all of them in.
Fan...pump...other and other no idea on the other.....

-Dave

[lightfoot]

Never seen one with 4 molex plugs. The ones with two do burn the plugs, I can replace those.

But I took the water loop apart and found some interesting stuff:
1) The loop was low on water. This is a big problem as the pumps *cannot* self prime. They need a full amount of water to work, otherwise you go into thermal la-la land.
2) Cutting those junk hoses off shows that it uses 7/16 inside diameter tubing with flanges so you can replace it with real tubes.
3) The real problem: The propolyne glycol or whatever they put in there precipitates out of the water into thin strings of gruel. This is a massive problem because:

4) It clogs the little water radiators inside the pumps.

If you pull the tubes off the pump, put real tubing on it, and immerse the right tube into a bucket of clean water and suck on the left tube (to prime it) you will get the propolyne out and some of the slime will appear. Dump that water and flush again with clean. Then turn on the pump and see if it pumps water through in a nice stream. If a dribble then you have to take the water block apart (8 screws) and you will find that on the back of the copper plate is what looks like a fine radiator with a white rubber thing in the pump.

What happens is the pump pushes water through the white things which forces the water *through* those tiny fins to pick up all the heat from the copper. If those find get clogged your pump will not flow and will not conduct worth a damn. Clean those fins with a toothbrush and soap and rinse it well.

More on refilling in a bit. Remember the pumps cannot prime themselves so there can't be air bubbles in the water system.

[NotFuzzyWarm]

More on refilling in a bit. Remember the pumps cannot prime themselves so there can't be air bubbles in the water system.

Quite true but nonetheless the system MUST have a small amount of air in it to allow for thermal expansion as the fluid gets hot. On the Intel liquid CPU cooler (a sealed system much like the BFL setup) that I use on my Render Beast the radiator is about 7/8 maybe 9/10 full and I'd guess the same should hold true for the BFL coolers. The radiator should be mounted so its outlet going to the pump inlet is on the bottom and fluid return on top so that air pocket is not circulated through the system.

[lightfoot]

Interesting point @notfuzzy: Thermal expansion is going to be needed, but any air that hits the pumps seems to fuck them over. Yaknow, I cut open one of the radiators and it is basically a two sided thing with one side going from the port down to the bottom of the radiator, where the bottom connects the two sides, then back up to the other port for output. If the pump catches a water bubble it's going to lose prime.

Putting the radiator upside down is a master stroke, it would solve the bubble problem, keep the pumps primed at all times, and of course BFL didn't do it :-) But I never thought of that....

Currently I'm running with an absolutely full radiator and rubber hoses that do have some give. Not sure how much added pressure there is, but maybe I should look into either a small air bladder type expansion tank to allow the fluid to expand without introducing air to the mix. Hm.....

[lightfoot]

Having fun dredging up old memories. One problem is that this 725gh miner here will mine at 725gh. Full blast out hot 71c chips, fans running at max, very much running on the edge with the FETs running at 90c even with the big heat sink on the back. Yes it will run, but not forever and any error will probably nuke it.

However. On the 3 power plug Monarchs there was a hack from BFL that allowed you to send it a command through USB to either set the clock speed (from F0X to FAX I think, maybe higher in hex) or the voltage from V0X (makes it a 3gh miner) to VAX (and maybe higher). The system would accept the command and auto-tune the voltage to produce the fewest errors, so it was best to just stick to playing with the "F0X" commands.

There was sort of a way to do it on the Bfgminer command line but it never worked for me. What I did was use the following command in a DOS/CMD prompt:

set /p x="F5X" <nul >\\.\COM12

Where COM12 is the COM port assigned to the Monarch's USB device (find it in Device manager, then expand the Serial ports). I put that in a batch file called mine.bat
Then I wait a few seconds and run bfgminer.

With F5X I get a nice 437gh speed, temp of between 50 and 57C, and most importantly I think the power efficiency goes from 1w/gh (at blazing full speed) to .5w/gh. Result is a nice warm 200 watt space heater that will mine 8 times as much coin for the heat as a BFL 50mh tube unit.

NOTE: You have to set this each time you power cycle the miner. When you power cycle, the miner will auto calculate speed and such for max hash rate which will turn your miner back into a seriously on the edge thing.

NOTE: I don't know if this works on the older 2 plug miners.

NOTE: Of course 8 times 0 is zero but hey, it is working :-)

Now to try and figure out why cgminer is barfing. I'm guessing they changed the response format of messages and never bothered to document it. Still I should be able to throw something into debug mode to find it out......

[NotFuzzyWarm]

It doesn't help that *all* flexible tubing allows water vapor to diffuse through the wall leading to coolant loss with no visual evidence of leaks or 'sweating'.. The material used makes the difference between almost no significant loss per year to 'oh crap!' in the same time. Norprene has the lowest diffusion rate.

Good article about mounting pump/radiator and touches on the tubing used to reduce loss: https://www.cclonline.com/article/2005/Guide/Water-Cooling-Kits/AiO-Water-Cooling-Best-Setup-Don-t-Get-It-Wrong-/

[lightfoot]

Meantime now that I have a cool running unit I want to stick the thing in my ATX box here with my Raspberry Pi's and just run it in slow and cool mode. Maybe I'll hit a block.

The real problem is the mining software: The driver.bflsc.c files in cgminer don't work.... well with the board. At 400mh speeds, the board is reporting only about 41 and 38gh speeds with a lot of "failed to find nonce" type errors. Bashing the code a bit, I can see that the error nonce being returned is always the same value, but the core number bounces around. Unfortunately cgminer can't show individual cores like bfgminer can, so that's not going to help. Running bfgminer shows 446/407gh speeds.

So about 90% of the cores don't work. But if I use bfgminer and look at the summary values per core I can see the core naming scheme is different. It labelels the chips as BFL 0aa, 0ab, 0ac.... through 0az then 0ba-bx then 0ca-cz... all the way up to 0ex. That is different as it's not splitting the engines by die but mushing them all together.

So 26*4 (a-d)+24 (ea-ex) =128 cores total. Or 64 cores per chip (each hashing at 4gh. I really wonder if they just took their old chip design and instead of putting one per chip they put 64 of them per bigger chip using a smaller fab process). But that's not important, if they are using 128 cores and the software wraps at 64 cores then yes that could be the problem.

Need to figure out how they did chip addressing. Hm. If they counted by hex numbers (0aa-0ae) then skipped to 0ba-be then 0ca-oce then that could be a problem. Only 5*5=25 cores would work which would still give us 100gh of speed instead of 400. Not right. About 12 cores should be working. Hm.

[Easteregg69]

X gets repressive as you steps up.