If you use boards like mine, you don't need an optoisolator or relay to trip the supply - the external turnon takes an active-high signal (3-20VDC), so it'd work just fine on GPIO from a Pi.
I believe chigieureitor (or however the heck his name is spelled) rigged up a setup like that for his mine in Venezuela.
|
|
|
I'll probably talk to some of the same reviewers I talked to for the Compacs last year - people I trust whose opinions cover important aspects of a successful product. Phil is really good at understanding customer mindsets, guys like NotFuzzyWarm are well-rounded engineers, and CrazyGuy as a top reseller can comment on marketability.
|
|
|
Prototype 2pac and pod PCBs are on order. If the 2pac works as expected, I'll have enough to send out a few "review" sticks to some good folks on here like I did last time.
VH is still hammering away at the last details of the driver.
I'll probably put up donate/loan/buy addresses on Friday.
|
|
|
Unfortunately I don't have a stripped-down 54-chip board anymore, just some functional ones I don't really want to take apart.
|
|
|
Hot air and a bit of pressure worked just fine for me on heatsinks, but then I don't have fancy tools.
Once the chips are clean it shouldn't take half an hour to figure out the routing. Just by looking at how the ground/power planes interact you can figure out node spacing. S7 won't be as easy as S5 since on the S5 the diode level shifters pretty much gave away where the inter-node data connection was but S7 chips integrate level shifting so it's pretty much just traces from chip to chip.
You looking at a 54-chip or a 45-chip?
|
|
|
Topside heatsinks can be removed with hot air to soften the epoxy. White epoxy will dissolve in acetone, not sure yet what cuts the black epoxy. Different epoxies were used on different batches.
As for layout, not really sure offhand. I'll check something over and might post a picture later.
|
|
|
I like that bitcoins have an actual economy of utility, where they can be directly exchanged for goods and services in a variety of ways. Most altcoins don't have that, most of 'em never will, and it's really the only thing I care about.
|
|
|
Now it's time to wind down with a movie and popcorn. Tomorrow I'll see about fetching some PCBs. Maybe VH will have some ramp code to test by then. Morning will probably be eaten up with building and packing, and it's sandwich day to boot, so probably nothing to report until later in the day. But I might start getting a funding plan together.
|
|
|
The Monday-Tuesday 9-day span before Thanksgiving I worked 91 hours, and that's with taking Sunday off and leaving early Friday for burger night. That was fun. My best single week was something like 110 hours in 7 days. Part of that is cheating; I did a 24-hour turnaround trip up to Chicago and back so, even though I slept about 2 hours on the road, I counted the whole thing. Didn't even go home, just got to the shop and napped in the breakroom for about 2 hours until Novak came in for work. That was a fun week.
The stick won't be wider. I'm going to try double-siding it, with node-level components on the top side (where everything else is) and just the ASICs on the back. This means the heatsink will be on the other side from the current Compac. If I'm able to manufacture the prototype PCBs without issue, I'll be satisfied with that design.
The BF16 sticks won't require this, since they don't need as much for node-level components. No level shifters or anything.
|
|
|
So I worked up a 2-chip string hardware test today that's only about 5 components different than the final stick will be. It's currently up and running, swapped out the previous parallel test board. So that's proof the thing works. There's still ramp code to worry about, but with extra node-level capacitance to help overcome initial power burst the thing lit right up first time on cgminer. So, I think that's pretty much good to go. I can never find my camera when I actually need it, but as soon as I do I'll post pictures. And there's still the ckpool link a few posts above to keep watch on. Also, for anyone keeping track, that's 11 hours for the workday plus 6 yesterday makes 17 hours on the week already. EDIT - camera was, oddly enough, sitting right next to the test machine where the stuff was hooked up. Right where past me knew it'd be needed next, funny present me didn't think to look there. ![](https://ip.bitcointalk.org/?u=http%3A%2F%2Fgekkoscience.com%2Fmisc%2Fcompac%2F2-chip_test_boards.JPG&t=665&c=8j9HJWgPKGEOeg) Here we have both mockups. The unplugged one has two BM1384 in parallel running off a CP2102 USB adapter and a stock Compac regulator. The plugged-in one has two chips in series running off a modified Compac regulator (1.2-1.6V for 0.6-0.8V node-level) and it's chugging away. It's pulling an unnecessarily high current for the hashrate because the voltage was set at a midpoint for testing reasons.
|
|
|
Just heard from VH, and 2-chip communication is working. All we need now is functional ramp code (ramping up the power use of the chips to help maintain balanced current between the two nodes during mining startup) and the baseline 2pac driver will be done. I'll try and throw together a 2-chip string tomorrow to start testing without ramp code, so I can at least get some efficiency ballparks. If the test setup can be made to start and mine without ramp code, it's probably enough for me to start cracking valves on orders and funding. The long-term average here should be 11GH, but low hashrate means high variance. http://solo.ckpool.org/workers/1BURGERAXHH6Yi6LRybRJK7ybEm5m5HwTr_COMPAC
|
|
|
Well I got the 2-chip BM1384 test board working better today, and got VH a buttload of data, so hopefully he can grind through it in the next week and get something working. Once 2-chip and ramp code looks good I've got a revision to make to the board before it's fully duplicating the final version, and then we're off to the races.
I'll look around tomorrow for a domestic board house that's reasonable before sending off the order to China. Being able to do all the other assembly in-house is nice for cost and time savings, but there's no way in heck I can make my own 4-layer PCBs for 0.5mm pitch QFN ICs.
On a related note, anyone know a good source for not-retardedly-expensive simple convection reflow ovens? I found one I liked with a US distributor, but the price for the stand it came on was approximately what I'd have paid for the whole thing; just the oven itself had only a 12" square bed and still cost approximately twice what I paid for my car. My IR oven's lack of consistency is starting to really get annoying.
|
|
|
Time's only one of the problems. Doing all the manufacture and maintaining a 100KW datacenter does eat up time, but that's only part of it. Waiting 2-3 weeks for a prototype PCB slows things down a bit. I know VH is putting all the time he can into it, but the lack of documentation on BM1384 stuff isn't helping his time efficiency a whole lot. I'm at the shop right now actually, bus-sniffing him some more data out of a makeshift S5 setup.
I'll probably start shopping around again for an affordable domestic prototyping place here soon, but right now where I'm getting 'em from is China (which sucks for a variety of reasons) because it's about 1/4 the cost of what I've seen in the US and to say I'm on a shoestring budget would be an insult to shoestrings.
|
|
|
I was pretty sure I had said from the beginning that I wouldn't do CPU cooler compatibility this time, for a long list of practical reasons, but that must have been said somewhere else because I don't think it's mentioned in this the thread that actually matters.
Yep, I bailed on CPU coolers months ago. I could try and make a small U3-sized thing that'd mount up to a cooler but it'd be power-limited (probably only 4-6 chips) and to do any just compactness (without custom machining the heatsink like U3 does) would require double-sided assembly, something I'm still not too sure on.
|
|
|
You might wanna actually read the description then.
|
|
|
Just yet another reason why we all switched to mainly GPU mining
Speak for yourself, dude. And probably in the Altcoin mining section.
|
|
|
Yeah and I've still got a couple cases of 'em at the shop, but trying to fit all the geometric constraints for CPU cooler compatibility and still fit a respectable number of chips in there sucks. Just putting a big square heatsink on there and giving myself all kinds of room for the chip field is way easier. I am intentionally making it the right size to find replacement fans easily, and there's a lot of variety in 80mm case fans depending if you want maximum cooling or silence or whatever.
|
|
|
Remember, those prices are still tentative. Could end up higher, hopefully not by much.
|
|
|
I was going to reply sarcastically with a search result, but searching let me know I hadn't actually said much about this in a while and a lot of things have changed since then.
The heatsink has not changed since then.
I gave up on CPU cooler compatibility a long time ago; it was an okay idea for a while but presented way too many problems to really be practical. From now on I'll be working around an 8cm square heatsink, atop which will sit a standard 2/3-wire case fan.
The pod is designed to be 12cm square with mounting holes at each corner for an enclosure or tower stack standoffs. It will ship with rubber feet adhered to these corners. The heatsink will be screwed on through an aluminum base plate to help maintain planar contact of the chips underneath. The pods will take in USB from either a Mini or B plug, and will take power through a barrel jack or PCIe 6-pin. There is a second 6-pin for daisychaining power to multiple boards from one master cable. All these connections are lined up along one side. On the manual-adjust pods (short-term plan) there's also a small potentiometer which can be turned to adjust core voltage. BF16 pods will implement software-adjustable voltage control with internal calibration.
|
|
|
Because of buck overheads and other things, the low-end hashing efficiency of the 2-chip BM1384 would only be at best a marginal improvement over the existing Compac. However, overclocking is where it'll really shine. The old stick's 5W performance was around 12GH, where 15GH is hopefully a conservative estimate for this one; 7W should see over 20GH instead of 16GH and the top-end 10W would get you about 27GH instead of the old stick's maybe hitting 20GH if it didn't burst into flames. Cooling should also be better because of the reduced heat density of two chips.
The pod, as it's laid out so far, is about a 5" square. I'm making both versions with the same formfactor for convenience, if someone wants to make enclosures or stack 'em into towers or whatever. The same hardware should work for any of 'em.
Heard a bit from VH earlier, he's making progress on the 1384 code. I'll be at the shop tomorrow doing some more testing and getting him some more info, but it's looking really promising so far. Code's a bit behind schedule but it was an optimistic schedule anyway, and I gotta take the blame for a lot of it for not getting hardware setups and some necessary testing done in a timely manner. Course there's also approximately zero documentation so that doesn't help. But if good news keeps coming to me at pace, I might have some good news for y'all in about a week.
|
|
|
|