Hacking BFL Monarchs and servicing them while times are weird.

[lightfoot]

Ok, it looks like BFL's Monarch board is shipping, pictures are starting to pop up, and I'm looking at the one on Ebay and going "MMMMM It would be FUN to FIDDLE WITH IT". Yep. It looks very cool....

Edit: I just bought one on Ebay for $1k worth of Bitcoin. Yep, by Wednesday it will be MINE! BWAHAHAHAHAHAHAHAHA

Ahem. So I thought I would start a thread for people to post their pictures, thoughts, and the like. I really am not interested in whining posts, so I will attempt to moderate this thread specifically to technical discussions.

And in a "support the people who bought them and put my bitcoins where my mouth is" I've just sent 8 bitcents to an early owner so he can get a P3 meter and a temp sensor so we can see what they do and get some data.

Congrats to BFL for shipping, let's see what we have...

C

[1714189278]

1714189278

[1714189278]

1714189278

[1714189278]

1714189278

[OgNasty]

I have provided wattage and temperature readings to lightfoot.  I will let him discuss the technicals.

As far as operation, these things were plug and play.  They make less noise than the singles, and I was impressed with how quiet they are.  They do dump a ton of heat, but can also operate at extremely high temperatures.  I don't have a thermometer in my mining closet, but the wall was over 100 degrees and these are able to operate with only a house fan pointed at them, which is nice since in the same environment my CoinTerra rig is nearly worthless without a direct A/C line running to it.

Actual performance at the pool is a bit shy of what was expected.  I tested one on GHASH.IO for a quick comparison and saw 660GH/s vs 620GH/s on P2Pool.  With my kill-a-watt saying they use 520watts per Monarch, that gives a real efficiency rating of ~0.8W/GH.

[lightfoot]

Ok, Ognasty was nice enough to take pictures, and I'll upload the key ones here. Or to some sharing site then go from there.

First, the money shot: A shot of the back of the board:



Thought: Holy fuck. That is one serious way to cool down the 12 volt to .6 volt power subsystem. The FETs are on the other side of the board along with the 2708 type gate drivers. At high frequency switching dropping that much one can expect a lot of heat from the high side FETs as they buck the 12 volts down to .6 as they are switching on and off (switching current is a lot more complex than on state current).

That heat sink and fan is covering it all up to the chips. Good idea, but man that is big.



Image number 2 in terms of interesting: This is the switching system for the FETs. Seeing it up close is interesting, the rows of chips in the center are the FET drivers, high and low side. Six channels per side, those bottom chips are probably 6 phase power sequencers. Equally interesting are the little heat sinks covering the high side FETs; wonder how hot those are getting as most of the heat on that style FET is dropped into the board and not out the top (they may be ballast sinks, to react to instant temp/load changes until the board itself saturates the heat and draws it out the back through the copper substrate). Hm.

010NE2LI is a reasonable choice for the low side FETs, you want two because switching issues are minimal on the low side (most of the time they are on to keep the rail close to .6 volts). The high side ones however are another story: You can't put two FETs in parallel because the issue on the high side is gate capacitance. They have to switch from full off to full on *fast* since during the time they are switching they have to drop all of the current at 12 volts going down to .6. Dropping 100 amps at 12 volts will incinerate a FET unless you transition quickly. Thus the lower the gate capacitance the better. Paralleling FETs doubles the current but also doubles the capacitance and leaves the FETs in that state a lot longer.

Interesting.

Then the chokes, and some white stuff, might be the heat sink compound from the chip to water block. Probably more efficient heat transfer than the singles since those had 8 little chips with not much area on top. If this is pulling similar power on one much bigger chip then surface area works in their factor. Hm....


Back of the chip. The temp here would be interesting, they're pulling all the heat up through the chip top and into the heat sink, this would tell how efficient the heat sinks are.

More thoughts later. Very interesting though....

[Taugeran]

OgNasty. Are you willing to uncap it (remove water cooler) so we can get a die shot?

Plx

Need ASIC pr0n

[lightfoot]

At this point, that could be tricky :-) Has BFL ever posted a picture of their chip anywhere?

Someone buy me a unit and I'll "unbox" it in real time. Then try to put it back together again :-)

Meantime I got some of the temp numbers from Og. I'll look at them tonight, post some of the highlights tomorrow. First glance though is that the heat is being taken off the chips very efficiently with those water blocks.

[MrTeal]

At this point, that could be tricky :-) Has BFL ever posted a picture of their chip anywhere?

Someone buy me a unit and I'll "unbox" it in real time. Then try to put it back together again :-)

Meantime I got some of the temp numbers from Og. I'll look at them tonight, post some of the highlights tomorrow. First glance though is that the heat is being taken off the chips very efficiently with those water blocks.

[lightfoot]

Well there is that... Very pretty.

[OgNasty]

OgNasty. Are you willing to uncap it (remove water cooler) so we can get a die shot?

I don't think so.  I'm sure lightfoot won't be able to fight the urge to buy a Monarch much longer...

[Rabinovitch]

Guys, almost 3k USD for 1 Th. WHY do you need it?..  

[Taugeran]

....

I think I went blind for a moment.

<3 ASIC pr0n

[jadefalke]

At this point, that could be tricky :-) Has BFL ever posted a picture of their chip anywhere?

Someone buy me a unit and I'll "unbox" it in real time. Then try to put it back together again :-)

Meantime I got some of the temp numbers from Og. I'll look at them tonight, post some of the highlights tomorrow. First glance though is that the heat is being taken off the chips very efficiently with those water blocks.

Mr. Teal, will you build up an own Board for that chips, like the habanero?

[crazyates]

Mr. Teal, will you build up an own Board for that chips, like the habanero?

Correct me if I'm wrong, but I don't think BFL is selling raw chips like they were with their 65nm stuff. They're only selling them inside the Monarchs.

[lightfoot]

Well, now I did it: I am an official proud owner of a BFL 700gh monarch, just bought on Ebay for "$1,000 in Bitcoin". Great... :-)

My guess is I'll have it by Wed/Thu and I'll sit down and take it apart. Because no one else is doing so right now, so why the heck not?

:-) Then... Then we shall see what we can do with it. Bwahahahahaha!!!!!!!

Ahem.

C

[k9quaint]

Well, now I did it: I am an official proud owner orderer of a BFL 700gh monarch, just bought on Ebay for "$1,000 in Bitcoin". Great... :-)

My guess is I'll have it by Wed/Thu and I'll sit down and take it apart. Because no one else is doing so right now, so why the heck not?

:-) Then... Then we shall see what we can do with it. Bwahahahahaha!!!!!!!

Ahem.

C

FTFY and good luck

[lightfoot]

OgNasty. Are you willing to uncap it (remove water cooler) so we can get a die shot?

Plx

Need ASIC pr0n

I will do so by next weekend. One is coming in from Ebay by Wednesday.

[Taugeran]

Mr. Teal, will you build up an own Board for that chips, like the habanero?

Correct me if I'm wrong, but I don't think BFL is selling raw chips like they were with their 65nm stuff. They're only selling them inside the Monarchs.

I'm sure if you contacted sales they would entertain you if you had a large enough production volume

[lightfoot]

FTFY and good luck

No actually this one is in hand from a guy, I've talked to him and seems legit. I will at least run it for a day or two before taking it apart, wonder what kind of thermal mating compound I will need.

One big thing will be baselines, need to know what the temps are on the backs of the chips and the FET drivers. Watch it with the scope to see what the clock rates are, stuff like that.

We'll find out Thursday. I'm really curious to see what this thing looks like and what makes it go.

C

[lightfoot]

Well, it's here. 700gh sticker, solid little thing, BFL Monarch.

Can't power it up because I have to go out. Will try to get it running in the Danger lab later this evening. Bad assed looking thing though, the board is *solid*

C

[lightfoot]

She's up. Running bfgminer 4.5.0 and started hashing with no problems. Running right now with both plugs and a Corsair 500 power supply, plug area is a bit warm. Cables are not. Board temp at back is 119f right behind the chips, 105 overall pretty much all around.

Hash speed 660gh. Not bad. It's making a high pitch squeak from the power supplies, probably the chokes resonating from the frequency. Interesting.

More in next post. How this thing works....

C

[lightfoot]

Now for the really interesting thing: The output from chiliflash.

DEVICE: BitForce SC-28nm SHA256
FIRMWARE: 1.4.2
Serial Number: 4320250
ASIC Installed: 2
IAR Executed: NO
PLL Latency: 48
Channel Parallelization: YES @ 16
Max Queue ID: FFFF
Scan Interval: 50ms
Total Engines: 1800
CORTEX-00: 4480MH/s
CORTEX-01: 5603MH/s
CORTEX-02: 6375MH/s
CORTEX-03: 6062MH/s
CORTEX-04: 6495MH/s
CORTEX-05: 5616MH/s
CORTEX-06: 3080MH/s
CORTEX-07: 2240MH/s
CORTEX-08: 4840MH/s
CORTEX-09: 6800MH/s
CORTEX-0A: 6160MH/s
CORTEX-0B: 6495MH/s
CORTEX-0C: 5280MH/s
CORTEX-0D: 5100MH/s
CORTEX-0E: 5852MH/s
CORTEX-0F: 3456MH/s
CORTEX-10: 6495MH/s
CORTEX-11: 4763MH/s
CORTEX-12: 5629MH/s
CORTEX-13: 5629MH/s
CORTEX-14: 4400MH/s
CORTEX-15: 4320MH/s
CORTEX-16: 5950MH/s
CORTEX-17: 3960MH/s
CORTEX-18: 5616MH/s
CORTEX-19: 6375MH/s
CORTEX-1A: 5950MH/s
CORTEX-1B: 6062MH/s
CORTEX-1C: 6375MH/s
CORTEX-1D: 6270MH/s
CORTEX-1E: 5100MH/s
CORTEX-1F: 3464MH/s
CORTEX-20: 4113MH/s
CORTEX-21: 6160MH/s
CORTEX-22: 5629MH/s
CORTEX-23: 6800MH/s
CORTEX-24: 6375MH/s
CORTEX-25: 6375MH/s
CORTEX-26: 6270MH/s
CORTEX-27: 5629MH/s
CORTEX-28: 6375MH/s
CORTEX-29: 6062MH/s
CORTEX-2A: 5184MH/s
CORTEX-2B: 6800MH/s
CORTEX-2C: 6688MH/s
CORTEX-2D: 6270MH/s
CORTEX-2E: 6270MH/s
CORTEX-2F: 6270MH/s
CORTEX-30: 6160MH/s
CORTEX-31: 6495MH/s
CORTEX-32: 6495MH/s
CORTEX-33: 5629MH/s
CORTEX-34: 6800MH/s
CORTEX-35: 5950MH/s
CORTEX-36: 6375MH/s
CORTEX-37: 5434MH/s
CORTEX-38: 5720MH/s
CORTEX-39: 6495MH/s
CORTEX-3A: 3960MH/s
CORTEX-3B: 4763MH/s
CORTEX-3C: 5720MH/s
CORTEX-3D: 5720MH/s
CORTEX-3E: 6576MH/s
CORTEX-3F: 6375MH/s
CORTEX-40: 6928MH/s
CORTEX-41: 6034MH/s
CORTEX-42: 5950MH/s
CORTEX-43: 6688MH/s
CORTEX-44: 6576MH/s
CORTEX-45: 5525MH/s
CORTEX-46: 6060MH/s
CORTEX-47: 6368MH/s
CORTEX-48: 6375MH/s
CORTEX-49: 6165MH/s
CORTEX-4A: 6270MH/s
CORTEX-4B: 6576MH/s
CORTEX-4C: 6165MH/s
CORTEX-4D: 6165MH/s
CORTEX-4E: 6464MH/s
CORTEX-4F: 6464MH/s
CORTEX-50: 6576MH/s
CORTEX-51: 5950MH/s
CORTEX-52: 6464MH/s
CORTEX-53: 6464MH/s
CORTEX-54: 6165MH/s
CORTEX-55: 5970MH/s
CORTEX-56: 4932MH/s
CORTEX-57: 5572MH/s
CORTEX-58: 6688MH/s
CORTEX-59: 6270MH/s
CORTEX-5A: 6480MH/s
CORTEX-5B: 6368MH/s
CORTEX-5C: 5656MH/s
CORTEX-5D: 6368MH/s
CORTEX-5E: 5865MH/s
CORTEX-5F: 5239MH/s
CORTEX-60: 6480MH/s
CORTEX-61: 6368MH/s
CORTEX-62: 6256MH/s
CORTEX-63: 5865MH/s
CORTEX-64: 5474MH/s
CORTEX-65: 5775MH/s
CORTEX-66: 5775MH/s
CORTEX-67: 5865MH/s
CORTEX-68: 6464MH/s
CORTEX-69: 6368MH/s
CORTEX-6A: 5970MH/s
CORTEX-6B: 5474MH/s
CORTEX-6C: 5775MH/s
CORTEX-6D: 5775MH/s
CORTEX-6E: 6160MH/s
CORTEX-6F: 5474MH/s
CORTEX-70: 6368MH/s
CORTEX-71: 6075MH/s
CORTEX-72: 5865MH/s
CORTEX-73: 5572MH/s
CORTEX-74: 5005MH/s
CORTEX-75: 6160MH/s
CORTEX-76: 5174MH/s
CORTEX-77: 5474MH/s
CORTEX-78: 6165MH/s
CORTEX-79: 6160MH/s
CORTEX-7A: 5572MH/s
CORTEX-7B: 5775MH/s
CORTEX-7C: 5474MH/s
CORTEX-7D: 6064MH/s
CORTEX-7E: 5390MH/s
CORTEX-7F: 4776MH/s
Total Processing Power: 746416 MH/s
ASIC CORTEX Count: 128
ASIC Channels: 128
Queue Depth:383
Critical Temperature: 0
Total ASIC Thermal Cycles: 0
Total PCB Thermal Cycles: 0
OK

That is very interesting. More in a bit.

C