Are you talking Bitmain or Avalon? The guy talking from Avalon is xiangfu

I'm emailing Janet. Several folks told me I should go straight to Yoshi so I'll look into that.

You mean like the holes that are already going to be there?


Okay, strap 'em on a C1 waterblock.

Yeah, funny you should mention... My contact at Bitmain hasn't gotten me any good info about getting chips, but recommended I wait until their next-gen chip is available but I won't get any info about designing with it until it's released so when you factor in dev and prototype time I wouldn't have a working board to test until probably August, so sellable machines about September. I'm gonna see if I can find someone higher on the food chain to talk to.

I um... I did get ahold of Canaan-Creative and they'll probably have a new chip out about the same time, maybe sooner. So if they're any more reasonable about getting datasheets and samples I might switch it up.

If someone in Europe wanted to license the design and manufacture it themselves, sure. But anything with our logo on it will be Made in USA (specifically, Missouri). Keeping things local - within our own oversight at least - also means quality control. We can't make sure that someone 8000 miles away is doing things right all the time.

Girlfriend? Wife? Obviously you don't spend enough time working on awesome stuff if you have time for friends and things better than friends.

Not outsourcing labor is one of the standard policies we've had since the beginning. The only reason to sell jobs overseas when people right here need work is greed, and some folks might have noticed that's not how we operate.

Also we're not building anything with a built-in PSU. So it doesn't matter if you use 120V or 240V. Just use whatever PSU you have.

As long as the United States still has people that need jobs, we will never outsource manufacture to a different country.

Please RTFT before asking questions that have already been answered all over the place.

FROM POST #140

That's what I intended to say the FET was doing. I doubt it's conducting all the time. But if a node voltage starts to go up because the chips aren't running full current, the FET will take the rest of the current and pull the node voltage back down. It's a safety feature to make sure your chips don't overvolt, but the power it's draining would have been burned off in the chips anyway so it doesn't really improve efficiency.


Sub-KW miners I'd call home miners. S5, Avalon4.1, SP20 are examples of this. Though the SP20 is pushing it given that stock settings request over 1KW and they're balls loud at that speed. Something more like the S3 would be a "definitive" home miner for me. And let's not overlook the entry-level quiet desk miner like the R-Box or U3.

I think we are specifically not going to offer bulk discounts on miners. Recall that one point of this project is to aid home miners, which means we're not going to incentivize bulk buyers to make things worse for home miners. We'll keep the price reasonable and the price will be the price regardless of the quantity.

The S5's screws are pretty spread out, so there's lots of room for board warping which would gap some chips. Unfortunately that's not something we'll be able to really get around since it's basically the same as the S1, but we can maybe arrange our chips in a better stagger formation to keep loads even. I'm gonna try to rig a temp sensor to measure the actual heatsink instead of the board, and base fan speed control on that. The diode array I think you're talking about is actually the inter-node level shifters for IO. We're looking at a three-chip-wide string instead of two-chip-wide so any one chip won't have as much effect on the node's voltage. Capacitor dividers will also be implied by each node's buffer cap, which won't be small. The Prisma used a node-level FET as a backup current shunt to help keep voltages down on underperforming nodes. This doesn't really help efficiency since the power is being burned off in a FET instead of the ASIC, but it does help manage node-level voltages. I'm not sure how necessary it really is.

A 2-chip string miner will require decent node-level capacitors but proper regulation will be at the mercy of each chip's functionality.

We're not putting off the TypeZero, if at all possible. But the USB stick miner is basically a tangible dev milestone between having nothing at all and having the full product. Since there's not a lot of extension required between having a one-chip miner and a two-chip miner, and between a two-chip miner and an eighteen-chip miner, and since our one-chip test bench will already be 80-90% of a working USB stick miner, there's no real reason not to go the extra 10-20% to make the complete product if 1) many people want them and 2) it's an affordable learning tool to introduce noobs to bitcoin.
We haven't stated outright that we'd be using sales profits from the stickminer to fund the TypeZero, but honestly that's what's going to happen. We got a bucket to put money in from whatever sales we have of whatever products and services, and then we pull money out of that bucket to stay in business. Profits won't be much from stickminers; selling 1000 sticks probably won't make enough profit to fund 50 of the TypeZero boards, but it's a start.

Once we have our breakout boards in hand and working, I'll complete a PCB design for stick miners. That'll be sent off to a board house for proto PCBs and while waiting for them, I'll work on TypeZero layouts. Then we'll assemble and sell stick miners while completing TypeZero layouts and getting prototype boards for it. Then once that's tested and working (and verified by some trusted hardware guys, probably talk to Philipma and MrTeal to start) we'll look into taking in money for a full manufacturing batch - contingent on chip availability from Bitmain, which it looks like might have a couple months' lead time. Which ain't exactly the best thing but if we can get everything else ready early and then the final step is dropping the BM1384s on the boards, testing and shipping, shouldn't be the end of the world. But it's not great.


I am nowhere with testing heat dissipation. All I know is, every existing BM1384 miner uses heatsinks on the chip tops. I'm assuming the manufacturer knows what it's doing, and since the S5 seems to have no problem drawing 10W per chip out the top (and supplemented by airflow over the board with the aid of the side panels) I'm gonna assume it works that way until I have the setup to do direct testing.

String design does need large capacitors to buffer current transients (and therefore voltage transients) at a node level, and the S5 does have smaller caps immediately tied to the VDD pads probably to compensate for trace and lead inductance and ESR of the node-level caps. Keeping a constant node voltage and current is a stiffer requirement for string designs than for parallel VRM designs.

If you can find a picoPSU-type board that switches 12V and provides enough current on all rails, you can lock your PSU on externally and run both your miners and computers just off its 12V rail.

Note that this is very similar to just buying another PSU, and not as cost-effective.

I'm not sure there's a way to convince a computer to stop running while it still has power applied. You could look into different standby levels in your OS and see what's possible.

We likely will be building one- and two-chip stickminers. They're great for learning, which is one reason we want to make even the USB sticks with adjustable clock and voltage. Noobs can learn the basics of pools and wallets and cgminer CLI and even a bit of hardware tweaking for $20 instead of having to buy a $400 machine.

Sure it makes no financial sense to buy a stick miner if all you're looking at is the bottom line profitability. But not everyone looks only at the bottom line profitability - myself included. I made more coin from learning about the hardware itself and using that knowledge for repairs and upgrades than I ever made mining. Some folks just want a first-tier tool to play with and try to understand how a thing works before going big. Why tear down a Corvette to learn about engines when you can take apart a lawn mower first?

If you crank the clock down it might work, at least for a while.

Yeah I'm still running some S1s at the shop. It's actually time to take them down to the bottom-end 0.8W/GH I just keep forgetting to do so and/or have no time. Full-range control of clock and voltage is key to a miner's longevity, so that's why we're working on what should be pretty much the most electrically efficient full-control topology available.

Actually, I don't even own a washer or dryer or lady. I've had a Dragon plugged in in the laundry nook for the last six months, heating the apartment and air-drying clothes for me. My last house had 200A service and lousy insulation so it actually was very full of miners, but now it's cheaper and easier to plug 'em in at the shop.

Snazzy board. What regulator controller are you using? I haven't really found one with the features I want but I can work around what I got.

Also, looks like I might be done with the BM1384 two-chip breakout board. I'll hopefully have time to verify it tomorrow and we can send off for some prototype PCBs.

We have a 3.3V UART input (so it can tie off a CP2102 USB-UART adapter like Novak cooked up for Prismas) with an integrated 1.8V level shifter for UART comms to the chips. The two chips on the board can be jumpered for series or parallel operation, and each node has a separate VDD and GND so you can externally power them on a common rail for parallel operation, or tie Node 1 GND to Node 0 VCORE and put a twice-voltage regulator across Node 1 VDD and Node 0 GND for string operation. Multiple boards can be tied in parallel or in series to test different matrix dimensions.

I could test a two-chip string with this board, which once I design the regulator would basically be a breadboard version of Phil's 2-chip USB stick. I'd go for a two-chip string instead of a parallel node because the higher voltage regulator at the same current output will be more efficient, and also because I can probably use the same parts as on the single-chip version.



The silkscreen is loaded with notes for proper use, just in case. If it works as expected I'll be quite pleased.

kilo17 - have you looked at an SP20 board and compared it to an S5 board? The S5 board has zero regulation, just some node-level current buffering capacitors. It has some passive comm level-shifting components. And it uses a standard protocol so no converters are necessary to talk directly to literally any controller ever. Each SP20 board has FOUR four-phase buck converters on it, the parts count for each of which approaches the total count for our entire board. And their protocol is non-standard so it'd have to be bit-banged by software, the timings for which would probably be implementation-dependent.
For another example, compare a Habanero board to an S3. The stock hashrate is about the same, but where the S3 has eight simple 53355DQP regulators, the Habanero has FOUR six-phase digital VRMs. Nothing against the Hab; if it were a woman I'd stick it in 'er, but the level of complexity required to support a large single low-voltage-high-current ASIC is about half an order of magnitude higher than a standard many-small-chips-in-parallel design and a full order higher than a well-thought-out string/matrix of small chips.

Please RTFT before asking questions that have already been answered all over the place.

FROM POST #1:



Regarding large-form chips, Spondoolies is the only company that's actually had good luck with 'em. Everyone else (well, I guess except KFC sorta) has pretty much crashed and burned. ASICMiner, Bitmain, Avalon all did quite well with small distributed chips instead of single giant chips. I'm hoping most outfits keep doing things that way, as it reduces complexity, increases modularity and flexibility and decreases requirements for cooling and power density. So basically, from where I'm sitting it looks pretty stupid to keep using that chip type. I like to think our design is the best design and our design is approximately impossible with giant chips.

Minions are also BGA and require ~100A VRMs. We very likely will never design a miner around that type chip (which appears to rule out Spondoolies' gen3 chip) because it makes literally every step of the process unnecessarily complex. Complexity means added cost and also increases the odds of machine failure.