GekkoScience BM1384 Project Development Discussion

[sidehack]

Yes, you can see in the first picture the temp sensor is the little square guy in the center of the ring of ASICs.

I've also thought about doing a plate of aluminum across the back. The bottom layer has a few signal traces in it for the temp sensor, but otherwise is for the most part absolute ground plane - which means chips aren't belly-pad-sinked into it directly like on the S5, except on the bottommost node. This thing is designed for top cooling. I can change that layer around if necessary to give it a bit more heat flow into the rear face of the board without a lot of trouble, but then you have to make sure your heatsink is electrically isolated which tends to increase thermal resistance. That's a big reason most string boards are top-coolers - having 15 different voltage-potential planes all rubbing up against the same block of aluminum can get fairly risky.

If I retool for a different chip whose dimensions are any larger than the BM1384, I'll probably have to either go with fewer chips or give up on the CPU cooler. It was darn difficult to fit 8x 8x8mm ASICs in a 30x30mm space with room for routing. Any bigger would be impossible. But I could just build for a basic heatsink, maybe 80mm square with a case fan mounted, and make the chip field any size I wanted to fit underneath it. Of course I'd keep the same size and screw positions on future boards which could mount up to the same heatsink.

[1714153665]

1714153665

[sloopy]

No long, wall of text from me sidehack and friends...

Just a quick note to say I keep following this thread and things look very nice!

I hope you all had a great Christmas, and if you are as excited as me for the pods, then that is when Christmas is coming for us

[Mudbankkeith]

Yes, you can see in the first picture the temp sensor is the little square guy in the center of the ring of ASICs.

I've also thought about doing a plate of aluminum across the back. The bottom layer has a few signal traces in it for the temp sensor, but otherwise is for the most part absolute ground plane - which means chips aren't belly-pad-sinked into it directly like on the S5, except on the bottommost node. This thing is designed for top cooling. I can change that layer around if necessary to give it a bit more heat flow into the rear face of the board without a lot of trouble, but then you have to make sure your heatsink is electrically isolated which tends to increase thermal resistance. That's a big reason most string boards are top-coolers - having 15 different voltage-potential planes all rubbing up against the same block of aluminum can get fairly risky.

If I retool for a different chip whose dimensions are any larger than the BM1384, I'll probably have to either go with fewer chips or give up on the CPU cooler. It was darn difficult to fit 8x 8x8mm ASICs in a 30x30mm space with room for routing. Any bigger would be impossible. But I could just build for a basic heatsink, maybe 80mm square with a case fan mounted, and make the chip field any size I wanted to fit underneath it. Of course I'd keep the same size and screw positions on future boards which could mount up to the same heatsink.

I see your point about a back plane cooling, if its not all a common ground plane, then:- "magic smoke"
Some plastic standoffs and a secondary "case fan" will be  a safer base plate.

[sidehack]

This one is all a common ground plane. If I made it not a common ground plane, the chips would heatsink out of the board better but putting a common heatsink to it would be riskier - hence why Bitmain went with top-cooling the S5 and dozens of tiny backside heatsinks on the S7.

Though, to be fully correct, it's not actually entirely a single ground plane. There are still vias on the belly pad of each chip which tie into internal node-level ground planes, which could still short against a heatsink if it were not properly electrically insulated. Soldermask takes care of most of that, but if it gets nicked or a via wicks a solderball to the surface you could get smoke.

[AJRGale]

would the thermal pegs I once mentioned be useful in such situation?
eg, make one hole for the belly, solder the peg in, its only connected to the bellies power plane and the sink..

[sidehack]

No, actually that would make things worse in pretty much every way (including cost).

[Mikestang]

The first picts look awesome, so glad to see this coming along.  My last U3 just died, so I am anxiously awaiting a replacement in the form of your pod.

[sidehack]

So I got the pod hashing under its own power. There's still the problem of it only running about 80% expected speed, and I'm still having trouble with it not wanting to light up at higher voltage/hashrate but I'll keep on it. Looks like in addition to forgetting a divider on the reset line, there also needs to be a limiting resistor in the high FET bootstrap because at high switch rates (basically, CCM) it was dropping out the internal LDO on the '53 and the chip was turning itself off. So that's another component I might have to add, which will mean more manual pasting because I'm not buying another stencil for a 200-part board with 3 different components. First I'll probably test driving it off the board's main 5V rail, bypass the internal LDO entirely. Hopefully I can get it stable at 300MHz soon; it was stable at 200MHz but ran 300MHz off an external buck. Surely the internal buck, what with the lower output impedance and better ripple filtering, can eventually learn to outperform it.

[cavaliersrus]

cool that its under its own power keep up the good work and updates sidehack

[baazju]

Love the updates in this thread. Thanks for your hard work, really cool to follow along and read this.

[AJRGale]

I had a thought about the possible bigger (say 9mm or 10mm) chip and the heatsink mounting, how does a heat-spreader (mentioned back in the past, you didn't like it dew to extra cost) sound for that? i'm happy to pay an extra $15 (be about that price to mill up the copper) for 80²mm by 5mm chunk of copper bolted on top of them chips, to mound a cpu sink on..

The way I see it, the price on that pod will change dramatically thanks to the possible cost of the chips and components, so this pod could be just $50 or $150, maybe more if you're unlucky (like me and price differences between us and au)

If not, its going to be an extra cost to do a custom heatsink, even though you have been mainly aiming at reusing heatsink hardware that is available these days (you looked into S1/2/3, Rockminer, New-Rockminer, U3, CPU sinks, etc)


As for the pegs, i was thinking of it for the backside heat-spreader that wes mentioned, since the pegs are at the same size as the body pans and made of copper, so heat will move though that better then vias and copper on the board, covered by solder mask.. but yes, $5 Each for them pegs are expensive.

Buy yes, that's my thoughts, I'd be interested on what you say..

[sidehack]

The pegs would also provide just as much risk, if not more, of shorting into the heatsink as vias.

A custom heatsink might not be much cost if it's fairly simple and in sufficient bulk. Hopefully. And also it'd be easier to design for than a CPU cooler because there's far less variation. Laying out the pod as it is was super freakin' time-consuming because of so many constraints.

If the pod gets made at all, it'd probably be $50 to $75. If I can't make it for less than that, it probably won't get made - or it'll be made with a new better chip.

[Morguk]

What's the potential ETA for this puppy?

[sidehack]

Don't really have one. Unlikely absolute best case is about the middle of February to start building, but if I have access to a new chip I'll redesign and build that one instead.

[atmtech]

Looks very promising Sidehack I'm sure the minor problems will be solved.
Missed out on getting some of the stick miners.
Any chance on making more sticks??
Also have you thought about building a USB hub into the pod allowing  users to plug in
Sticks  I would gladly pay up to 25 extra just to avoid the hassle of which hub to buy.

[sidehack]

If I build a hub it won't be as part of a pod. Really doesn't make much sense, since the only doubled-up part would be the USB jack. I'd need to build a separate high-current 5V source and all the brains, and it'd be better off as a separate thing. So no, until just now I hadn't thought of building a USB hub into the pod and it took only a few seconds of thought to analyze and reject the idea. I've had in mind for months to build a good standalone stud hub so if anything I'll just keep on that project.

Regarding making more sticks, I could direct you to my most recent post on the stick sales thread.

[cavaliersrus]

there are plenty of hubs out there that do the job if your connecting to raspberry pi it sometimes can be a pain ... how is that 49 usb hub working hack the one that uses the computer powersupply working does it work with a Pi ?

[jstefanop]

Nice work! Thats a LOT of poly caps for just 8 chips. Are you having issues with transients? Figured a single poly cap our a couple ceramics between serial plains should be enough for that design.

[sidehack]

Well, I like to overbuild things. Who in their right mind would put a 16A-rated buck on a USB stick? But y'all are having endless fun with it. I don't want to scrimp on node-level buffering so I have a good full-voltage output cap bank and good node-level caps as well. The per-node polys are 470uF Compac leftovers (which will probably be 470 or 680 pulls from S1 and such on the final) and a 100uF S5 pull tossed in for good measure. You gotta remember, I'm designing this to use scrounge parts so I have to play with what's available and keep costs down.

Good node-level current availability is essential for ASICs to initialize at low node voltages, since there's a bit of a spike when the chip first fires up which might not be available from idling chips higher up the line, and if you're running at 600mV per node you really don't have much room for sag. Part of that will be overcome by good node-level caps, and part by dynamic voltage control - we'll probably have the controller init the string at a higher node voltage and baseline frequency, then ramp down to user setpoint voltage as the frequency ramps up to user setpoint. This should ensure minimal excess power draw and reliable startup.

[toptek]

Missed out on getting some of the stick miners.

If can't find any and live in the US I know of two places online that sell them in the US


http://holybitcoin.com/  has some in stock

https://asicpuppy.com is out of stock but does sell them or did .