GekkoScience BM1384 Project Development Discussion

[sidehack]

So Novak and I have been rolling around an idea for a miner since about December. We were inspired by ASICMiner's BE300 sample chip test results and thought we had a pretty solid idea for a machine built around them.

Unfortunately, AM's BE300 doesn't really exist outside of those test chips, and it's quite probable it never will.

Bitmain's BM1384 (the chip inside the S5), however, is a fair substitute. Its bottom-end power efficiency compares to the BE300's midrange, the core voltages are a bit more manageable and the chained comms setup appears to be easier to work with. We ended up buying some sample BM1384 from Bitmain and the chips arrived on Monday, so now we can start playing with them. The datasheet is incredibly sparse regarding any kind of protocols, but fortunately the signalling appears to be identical from the BM1380 (S1/S2/U1/U2) and BM1382 (S3,S4,U3) so it shouldn't be too hard to figure out.

I did a bit of prodding on an S5 yesterday and got the comms topology and level-shifting pretty well figured out. Some prodding on a U2 has given us info on USB interfacing to the chip chain. Novak's tearing into U3 cgminer drivers to iron out data specifics and I'm going to work on the regulators and digital IO requirements.

The next step, which I'll be working on this afternoon, is to design a simple BM1384 breakout board that consolidates power lines and puts signals on a simple header. With a two-chip board on independent planes we can test one-chip comms, parallel node comms and string comms to get a basic understanding of topology changes.

It occurred to us that a simplification of this breakout board could be an effective USB stick miner which might be capable of 10GH around 3W of draw. We may design a PCB for this if there's enough interest.

The primary goal is to build a simple board which would be USB-connected to a controller, and capable of adjusting both core voltages and clock speeds using cgminer flags. We're looking at a single board capable of 300GH at 150W, downclockable to around 150GH 50W. At mid- or low-range settings it could be run off a brick with a quiet 120mm fan and heatsink and be a decen Jalapeno-formfactor home desk miner. The board will be designed specifically so that four of them would mount to an S1 chassis. Couple that with a 4-port USB hub and a Pi or something as the controller, and you have an "S1 Upgrade Kit" which will aim for 1.3TH at 600W clockable down to 600GH about 160W. These are board-level power estimates. This board is codenamed "TypeZero". We're trying to iron out a design that would incorporate the overall efficiency (both in power and cost) of string topology and the flexibility of standard VRM topology. Basically we're trying to take the best parts of several decent designs and make one thing that should be modular, flexible, reliable, affordable and efficient. Lofty goals to be sure, but hey why not aim for the top?

A secondary goal would be to design another board which could mount to a Prisma chassis. Modifying our design to a similar power density as a Prisma (1KW board-level) we could see a 2.4GH miner downclockable to about 1.1TH at 300W. There are other prerequisites to meet before considering moving forward on this, however.

We're a pretty low-fund operation these days, but we have some big jobs coming in which might be able to provide capital for a batch. In the worst case, we might try and open a preorder (grumble grumble) once we have a tested working prototype and solid final design. Manufacturing feasibility is going to depend heavily on the chip cost from Bitmain. With the "upgrade kit" savings of not having to manufacture heatsink chassis and controllers (and then pay to ship it all everywhere), the bulk of BOM expense is going to be ASIC costs, most likely. We've always liked doing things for home miners, and trying to keep prices fair, so we'll do our best to get these things made in a way that doesn't suck for everyone.

If it takes forever for anything to happen and Bitmain's already out of BM1384 and selling the BM1386 or whatever, it shouldn't take much to modify the design for the new chip. Assuming they don't make major changes to the comm protocols, which they haven't done yet so that's encouraging.

So, questions? Comments? Sandwiches? We're not really getting paid to design all this stuff so "keep up the good work" sandwich donations are more than welcome (1BURGERAXHH6Yi6LRybRJK7ybEm5m5HwTr).

[Gladimor]

I will be watching this with great interest

[TheRealSteve]

It occurred to us that a simplification of this breakout board could be an effective USB stick miner which might be capable of 10GH around 3W of draw. We may design a PCB for this if there's enough interest.

*cough*

So, questions? Comments? Sandwiches? We're not really getting paid to design all this stuff so "keep up the good work" sandwich donations are more than welcome.

Where do we send the sandwiches - and do you have any preferences?
Alternatively, where do we send BTC donations?

[MCHouston]

So, questions? Comments? Sandwiches? We're not really getting paid to design all this stuff so "keep up the good work" sandwich donations are more than welcome.

I have some questions.

Are grilled cheese sandwiches okay?

At the current rate how long are you looking to have your first ones ready to be sold?

How much funding are you needing for "preorders"/manufacturing?

[SFMiner]

I love the idea of developing a board for the BM1384, but potentially being able to use it with the BM1386 (or whatever Bitmain decides to call their next chip).  I was hoping the BE300 would materialize as well.

I would be even more interested in a design that could use Bitmain's existing S1 or even S3 controller boards, which would save the cost of buying a bunch of RPis and USB hubs, and the cabling issues those create. I imagine that might be tough, but is that even possible?

I'm definitely interested, would love to send BTC for food along with my food for thought.

[ezeminer]

I'd be interested in buying some awesome 0.33 GH/w.
As well as delivering some ham and cheese sammiches.

[Biodom]

very interesting; we needed that
feels like a kickstarter campaign

[AbuGarcia]

If you guys do finally make a sizable pcb, please make sure that the asic chip height is the highest point on the pcb.  Also lay them in rows and not alternating.  I would recommend that you make pcb's that can be bought as blades and the end use figure out the cooling.

[philipma1957]

I have an interest.  I can't spare a lot of btc  upfront but I have a house full of parts. Be willing to  do something to help.

I know 3w 10gh usb sticks would be a good seller at the right price point.

A large tube/prisma style miner :

1th to 2.4th

300 watts to 1000 watts would sell . 

[ElGabo]

I got 66 S1-s laying around and collecting dust.....  

If you've got working prototype or something I can put a little money in it....  

Edit: Tip jar for the donations?

[sidehack]

Actually, since the BM1384 takes most of its heat out the top, we'll really only have to worry about it being the highest point on the board on one side of the board. That's also going to be a limitation for a USB miner.

Rows are logical. We're nowhere near layout stage yet but there hasn't been any talk of not putting the chips in rows.

I was even thinking about putting edge connector pads on the board in case someone wanted to backplane them, which could get interesting. Basically a backplane with power and a USB pair to each socket, have a hub chip on the backplane and there you have it.

We may try and come up with a closed-form solution for a one- or two-board Jalapeno miner but in general the boards would be available barebones.

I don't know enough about the S1 or S3 controllers to say if that'll be straighforward or not. We want something flexible, so the most flexible option is simply a USB with a cgminer driver. It's not specifically an S1/S3 upgrade, it's a modular mining board that will be sized to mount on an S1 chassis but can also be run in infinite other configurations. With that in mind, making it work for an S1/S3 controller will not be a priority. Not when you can get unpowered hubs for few dollars and oughtta be able to run a bucket of 'em off a single host.

No idea on when they'd be available. Right now we're working on being able to communicate with a single chip. Once that's done it shouldn't be too far to get a passel of 'em going, but making sure the whole thing is rock-solid reliable will mean we don't just throw it together. Plus prototypes for any PCBs will be at least a week getting made. And then if we haven't fetched a pick-and-place yet...

Funding for an initial batch will depend heavily on chip costs. We're guessing probably around $30k for 500 boards but hopefully that's a high estimate.


Sandwich donations can be forwarded to 1BURGERAXHH6Yi6LRybRJK7ybEm5m5HwTr
We tend to lean pretty heavily on the Bacon Texas Cheesesteak Melt from Waffle House. Just throwing that out there.

[ElGabo]

Small tip sent for some sandwich.  

In my opinion no need to use the controller of the S1 S3 etc....

But if somebody have these old miners laying around the chassis and the heatsinks are very useful.

The shipping price of the miners are very high compared to the price of the miner. If you able to send only the boards the shipping will be much more lower.

[sidehack]

Novak's poking around on the S1 controller now to see what its interfacing is like. Thing is, with our modular board size there'd be 4 UARTs required per S1 and it appears the S1's controller on has (or breaks out) two. It'd be nigh impossible to do a chain-up/chain-down for two boards because the chips are hardware-addressed. Unless we set a high bit on a common line that the user could jumper as 0 or 1 so the controller could tell the two boards apart? But that kinda makes routing suck.

If we intended on doing an S1 Upgrade Kit only, we could probably make it directly compatible. But that's not what we're going for.

Also, thanks for the sammich money. Tuesdays are usually sandwich days around here.

[ElGabo]

This upgrade thing was just an idea.

Will follow this work. 

And keep up this good work for us, the community......

[GenTarkin]

And that my fellow miners is the ONLY way to make profit in btcland, ripoff all the miners w/ a new project to drool over. Yay, for something else we all dont need.

[notlist3d]

Very interesting.  I wish you the best of luck with it sidehack.

I personally would love to see you make not a upgrade as much as a all in one.  I would love to be able to get it without getting S1's.  Just the miner it's self and maybe raspberry pi to run it.

One idea I have that could be a big market on upgrades is if you did it would be a dragon (A1) upgrade.  There are tons and tons of dragons out there that are at end of life.  If you could make a upgrade there that is something you would have no competition currently.

[ElGabo]

And that my fellow miners is the ONLY way to make profit in btcland, ripoff all the miners w/ a new project to drool over. Yay, for something else we all dont need.

?

[Xian01]

And that my fellow miners is the ONLY way to make profit in btcland, ripoff all the miners w/ a new project to drool over. Yay, for something else we all dont need.

How about we cut sidehack some slack for now, huh ?

[novak@gekkoscience]

One idea I have that could be a big market on upgrades is if you did it would be a dragon (A1) upgrade.  There are tons and tons of dragons out there that are at end of life.  If you could make a upgrade there that is something you would have no competition currently.

Looking at the A1 it would not be super easy to fit in the mechanical parts required for a dragon upgrade, although, the boards would be very similar in size to an S1 blade.  It's definitely worth considering.

--
novak

[MCHouston]

Funding for an initial batch will depend heavily on chip costs. We're guessing probably around $30k for 500 boards but hopefully that's a high estimate.

PMd