So I had a really long and thoroughly quantitative post written up addressing points of discussion from the previous 10 or so posts but the forum ate it and I don't really feel like going back over all those numbers again. Call it a lack of motivation if you want. Novak and I got back to the shop this morning after a 37-hour marathon drive (1630 miles) for supplies to find the power had gone out, all the UPS were depleted so the servers were powered down, the internet wasn't working and the hosting router had eaten its own hard drive, and also I haven't really had anything to eat except Twizzlers and sunflower seeds in about 23 hours. All in all it was a pretty fun couple of days.

Anyway. The point was, the S5 was designed at a time when the hardware scene was pretty different than it is now. It's fairly optimal when you consider the size constraints of the S[odd] miners, prioritizing hashrate first and efficiency second (since top clock on the BM1384 is still more efficient than bottom clock on just about every other chip brought to market). With higher chip density they could do better, but at increased cost (17x3 like in the S4+ could get you 1.4TH at 500W with 70% more chips, whooptee doo).

The 18-chip board will be more efficient than 18 sticks, because you have 1x support hardware instead of 18x support hardware and the regulator should run closer to 95% than the 87-90% on the stick.

I will probably also not release an 18-chip board with the BM1384. I'm assuming, with the dev timeframe we have, that Bitmain's next gen chip will be out and if I can design around that for the larger stuff I certainly will. Maybe if the BM1384 chip prices go way down it'd still be viable, but it'd be starting off at a disadvantage when the next chip will probably see 0.3W/GH around top clock instead of bottom.

I seem to remember the Avalon4 has an optimizer built in, where you can give it a hashrate and it'll iteratively drop the power until it finds the lowest stable point? We'll probably integrate an autotuning feature like that into it. If we can write an autotuning driver that can tune per board will be pretty great too.

I'm not sure if we're gonna have an input power measurement on there or not. Shunt measurements to take away from system efficiency - sure it's maybe a quarter percent of waste, but... yeah that's actually probably okay.

I think trying to find an optimal operating point based on power efficiency will pretty much always end you up at the top stable clock of the bottom possible voltage? If your only parameters are operating cost and hashrate you'll always end up finding the highest clock available at the highest efficiency voltage setpoint available. Being able to say "get me the most hashrate for this maximum power" or "get me the least power for this desired hashrate" are going to be different problems, but not difficult to solve.

I think as long as the heatsink is staying cool (which implies the chip is staying cool), there shouldn't be much problem. The buck on this guy is built for an upper limit of 800mV and about 20A which is a heck of a lot more power than any stickminer should ever use. The USB jack is rated for 1.5A.

Also if the new PCBs work he'll probably be getting a proper final design stick end of next week that can be babied.

But yeah, it'd be kinda lousy to see it catch on fire.

Well hey, nifty. I'm pretty sure the new revision Compac PCBs and the barebones 18-boards have shipped, so I might be playing with those next week.

I wonder when the other three guys are gonna start reporting opinions?

Dude that's not even fair. Maybe I should look into getting a better rate for our setup since 80% of our power consumption is "data center" uses.

No, because then it becomes special-purpose hardware.

I hope Bitmain comes out with another decent chip so I can play with it. I really want to like SFARDS and they do seem pretty open to working with the community, but I just really hate BGA.

But yeah, S4+ definitely caused a double-take. And who knows, maybe they'll make S2 upgrade kits next generation?

*crickets*

But if you want something not likely to burn down your office, don't get a Neptune.

I'm reusing one of the hubs I originally jacked up for Block Erupters about two years ago (https://bitcointalk.org/index.php?topic=241652.msg3343826#msg3343826). I pulled 5V in from my bench supply and used that to power the Pi and five sticks without any trouble. Doesn't take much for a crappy powered hub to be a not crappy powered hub.

Also, I mentioned somewhere else but Novak and I are gonna spend some time looking into building our own snazzy flexible powered hub, probably 7 port. We're looking at an internal regulator that can take in a range about 6V to 24V and steps it down to a clean 5V. The 5V rail would be tied to a barrel jack which could be used as a power input or output. We'd build it to handle at least 1.5A per socket (by which I mean every socket could deliver 1.5A simultaneously and it'd still work). I'm not sure yet if that'll be a fused limit, an implied limit, or a hardware-switched limit. But something like that would be pretty nifty for running sticks on, and also pretty nifty for hooking up to solar or a car battery or whatever else.

The rest of y'all, I'll get back to you sometime.

Actually, most server supplies are just fine for GPU rigs also, provided you have an external-on signal line which can kick on the supply along with your ATX. My breakout boards and I believe pmorici's do this. Heck, I got into server supplies by refitting them for automatic extra rails in GPU rigs.

Also, if he's looking for quiet fans for the S5 a server supply could easily be too loud. Something to consider.

If it's not reporting HW errors during the delay, the "no valid hashes detected" timeout is still based on U3 timings, which the duration is frequency dependent. The U3 would have four times the hashrate at a given frequency (as it has four chips) so cgminer says "well heck, I haven't seen a share in this long it must be hung and needs a hard reset" but that's not really valid for the stick because it's a single-chip device. The "we have a problem" timeout should be four times longer. That'll be taken care of when we have our own driver.

Looking good, Phil. I won't get to play with my meters until probably Thursday. Thanks for getting some good numbers up.

At freq 175 it should converge to 9.63GH

Regarding folks chipping in, I'd just as soon not take large chunks of coin from anyone, especially not if it's considered prepayment for sticks. Even though we have what appears to be a functional proto stick independently tested, I still don't have the final version PCBs in hand yet, let alone tested. I dunno. Maybe I should be confident enough in the prototypes? The only reason I'm buying the chips now instead of after I have tested final versions is because I gathered from Bitmain communications that chip availability was limited and needed to be taken care of soon.

Granted there's more evidence of the product than most people who take in money for Bitcoin gear ever provide, but I learned very well from my last boss the dangers of selling things which don't actually exist.

So, update.

I just got final prices from Bitmain on chips. We're organizing a two-part order, 1200 chips for me and 800 for the guy in Germany. With 1200 chips I can use 1000 as intended earlier for Compacs and Amitas (probably 600 and 200, respectively) and have 200 chips left which will allow me to do a full set of 4 of both 18-chip (half S1) and 30-chip (Prisma) boards for testing.

The portion I'll need to put up for chips and shipping is about 16.6BTC using current values. I'll be shifting funds into the 1BURGER address for now. Being as I'll be all over three states until Thursday I don't expect to take care of final business and submit payment until the end of the week. I don't want to lean on y'all for footing any of that bill, but the way things have been going, I'd probably get lynched if I didn't at least alert folks to the opportunity.

Ah, Hackaday - the internet's premiere source of information for all things Arduino and Internet of Things.

Also the hack on that page only described how to change power limits on one model of Pi, not address older versions with hardware limits and not how to keep any of it from catching on fire due to overdrawing. If you wired power to the header instead of the mini jack you might get away with it. Also the problem of port spacing allowing only one stick still stands.

But yeah. For sticks, just get a not-crappy hub. Even a pretty crappy hub can be made to work well (as evidenced by my $9 11-port powering 5 sticks and a Pi).

For sticks, but he's talking about the TypeZero boards which will certainly not be USB-powered.

Even if you could get a single stick worth of current out of the Pi's USB ports, since they take power from a stupid micro cable (I really have trouble understanding why the designer would use the literal worst socket) you'd be drawing double current and you'd smoke your power-in.

The machine I use as an at-a-glance reference for how things are handling heat reported about 8C lower on the hottest day we've had so far than it did on the less-hot-than-that hottest day I had before upgrading the exhaust about two weeks ago. I'm not sure if it matters how large or small the room is as long as you keep the hot side fairly well sealed from the ambient and have enough exhaust pressure to keep hot air leaking and mixing with ambient air to a minimum.

If you can't keep things sealed and ducted, I'd recommend a tall room with strong exhaust at the top.

Yeah, I've had two of these for over a month and haven't even bothered firing them up yet. One hasn't been unboxed. The other came in with half the caps already mushroomed. I recognize the garbage brand as one of the most common failures I saw in the couple years electronics refurb was my full-time job. If I fire them up at all it'll be after removing or replacing all those capacitors, and not on the stock PSUs.

I just set up a pair of ~9KBTU exhaust fans in our hosting running off a single 120V 20A circuit doing a reasonable job of cooling what is currently about 38KW of machinery. Once I finish ducting everything properly (hot/cold aisle separation helps a lot for controlling ambient temperature) it'll be even better, probably good for 50KW.

What kind of ambient temperatures are you dealing with? Are you going full AC? 35% seems like an embarassingly high number.

With hubs you could tie who knows how many boards to a single controller, and it's not just limited to a Pi but anything with USB connectivity that'll run cgminer.

The short answer is "no". The longer answer involves quoting the first post of this thread (I'll leave finding the other times this question has been answered thoroughly as an exercise for the reader):


Since the goal has always been to create boards which will run as quiet desk miners, a secondary purpose being four of them will mount on an S1 chassis, no we will not be making it compatible with the S1 controller because that either makes it unnecessarily complex (having two completely separate means of interface) or makes it useless for its primary purpose (specifically, being a simple and flexible miner on its own merits) by tethering it to very special-purpose hardware not everyone has.