Topic: GekkoScience BM1384 Project Development Discussion - page 69. (Read 146665 times)

I started work today on an 18-chip board which should mount to one half of an S1 heatsink. It'll be designed to wire up to an external power source and controller. Basically it'll be used as a testbed for software, cooling and such. I'm going to be working on a better regulator for it than the stickminer uses, though to be honest the stickminer's so overbuilt that, with a better inductor and increased output caps, that thing could actually do it. Which is pretty sexy. But I can make one more efficient, and that's sorta the key, so I'm gonna try. The controller I'm eyeballing won't be as easy to adjust the output voltage as the one on the Compac, but that doesn't mean it'll be in any way hard. Just it won't be as easy. That buck driver (TPS51219 if anyone's wondering) is a pretty decent little part.

We also need to work on integrating fan speed, voltage control and temperature sensor into both hardware and software. That's gonna rely a lot more on Novak than the project has so far, but he's also got his own projects that I need to help with so we'll probably be trading back and forth a lot more in the coming weeks. I'm hoping to have the 18-chip base board ready to send off for proto PCB by Friday, but I also have to do some renovations in the hosting room before this weekend. The next couple days are going to be fun.

And one of these days I'll take the time to do some more experimenting on the Compac stick and see what needs to change to get the Vcore noise down. I have an idea or two I just need to get around to testing.

well Slacking off can restore ones mind.  Playing a baseball  game can be fun.

If by "focus on the project" you mean "sleep in and play baseball", then you're mostly correct.

Friday I noticed that part of the speed issue I was seeing pool-side from the seriaL-board on the S5 controller was rejected shares. I had about 25% rejected, which accounted for the ~4GH (of ~16GH) missing poolside. So I swapped the paralleL-board onto the S5 to see if it was a hardware or software issue. I immediately started seeing 25% rejected shares, which snuck up to about 50% later in the day. Right now the ~17GH poolside is from the 50% non-duplicate shares on the USB controller and the 50% non-rejected shares on the S5 controller. One of these days we'll have software capable of doing something right, I promise!

Today I'm gonna test out some ideas for cleaning up the Compac regulator output. If we can get a functional design this week with a low-noise output prototyped I'll probably send off for proto PCBs end of the week. We'll have 'em in hand for testing a week or two later. I also should have some more heatsinks inbound, samples of the final version which I can attach to some Compacs if they work and get 'em out to my testers. I also have to focus on some hosting stuff at least one day this week, and actually next week I'm off for two days (yep, vacation - first one since Christmas week when I didn't come in to work for a couple days because of debilitating illness) so hopefully I can make progress before Friday. But the weather was really really nice all weekend and it felt like a waste to spend it working instead of out running around doing fun stuff since it's not like I get paid anyway.

Also, and I'll be honest, my weekend slacking was somewhat motivated by the knowledge that, even if I made progress, I had no means by which to report it.

Great stuff. These 3 days give sidehack some time to focus on the project without any question on the forum 

The forum is back : very good :-).

And the news about THE project.

to update this gear is now 16-17gh  which was hoped for by sidehack and look at the charts it is right there!

The L-board over there on the left. We call it the L-board because it's a three-chip breakout board so the chips make a right angle. Like the letter "L". The first chip will always be the first chip, but you can put a second chip either in parallel (alongside, as in the previous picture) or in series (above, as in this picture). Two power inputs, one for Vcore and one for 2x Vcore to be split by the string configuration. I also wired up the LED circuit from the Compac design onto it so I can see the flashing LEDs. Whoever said that was a good idea was right. Except that something with the 3.3V comms on the S5 controller sorta prevents the LED from working properly, or it strobes so sporadically I don't wait long enough to see it.

In any case, I rewired the schfifty-three to output 0.8 to 1.3V so I could safely take the string voltage up from safe-for-one-chip-if-something-is-very-very-wrong to can-operate-at-150MHz-at-half-this-voltage without anything exploding. I just checked the per-chip voltages in operation with my trusty crappy DMM and they're within 2mV of each other, so that's a pretty good split (less than half a percent difference). I did light this up on my bench supply to watch the power consumption and the power draw profile on chip init is pretty cool. It goes from very low to full in about a linear sweep of several seconds, which tells me they coded the init to ramp up the power slowly to avoid severe imbalances. If that works how I think that works, I could probably pull the 1800uF cans off the board and it'd still be alright. The S5 has 100uF tantalum per 2 chips in a node, and I have 200uF ceramic per one chip node right now, but the U3 init (expecting parallel chips, no balance needed) was way too abrupt and the first chip was very much outpacing the top chip until power was so out of whack that they both shut off - in well under one second.

So, for now I have two working two-board setups, one a standard parallel on a USB adapter and one string on the S5 controller. Novak's in command today, and will probably be figuring out what the heck all the code's doing so we can try and get some Compac- and Amita-specific drivers going in cgminer.

I'll probably be spending the day odd-jobbing and trying to cut down the noise in my Compac regulator. Hopefully I can work out some simple changes and a fresh PCB layout we can prototype in-house and not have to waste more time and money on prototype PCBs that don't really work. I won't be able to run out full Amita prototypes until we know what node-level capacitances we'll require, which looking at the init power profiles, will definitely require some working driver to test.

hey hack very excited to see how well project is coming along and cant wait to see future upgrades and progress reports

If everything's working proper it should level out around 16.5GH. I don't have remote access to the stats page (shifted it off the VPN-capable subnet so Novak could play with the code more readily) to verify it right now, but I am admittedly disappointed in the 3-hour average currently presented. It's been running at speed for right around 4 hours. Of course, 16GH on a min-diff 128 makes for some pretty gnarley variance. The 12-hour should look much better.

Hopefully tomorrow early I can get the string L working, swap a fresh chip on and get the 3.3V RX back up. If it ends up working like I'm pretty sure it should, all I need to do after that is iron out the Compac regulator noise issue, get some solid numbers on minimum node-level capacitance to maintain balance during a two-chip initialization (testing so far indicates it's somewhere between 200 and 2000uF) and lay into some PCB redesigns. Heck, if the Compac gets to working, the Amita should be a single revision because the chip extension is pretty trivial compared to the friggin' month I've lost to regulator issues.

would like to watch the stick mine is it on eligus with the burger address?

anyone with a link ?  

found it


http://eligius.st/~wizkid057/newstats/userstats.php/1BURGERAXHH6Yi6LRybRJK7ybEm5m5HwTr

looks like 10-13gh

Very cool, dude. Grats on getting your 2 chip design working ! Looking forward to your 'big' miner, and hoping Bitmain will be reasonable with chip pricing for you.

Well, my string L-board worked pretty well. The 3.3V RX for some reason decided to crap out pretty much without ever actually working, which becomes a chip problem. All that left me was a 1.8V RX, which was fine for the CP2102 adapter into cgminer, which so far doesn't work with dual chips. The S5 controller uses 3.3V logic, so I needed that 3.3V RX line. Feeding it from the 1.8V RX line gave it enough, apparently, to know there was a board attached with some chips and to start mining but I never saw any recorded hashes. And then the 1.8V RX started wigging out, possibly an internal high-tie resistor on the S5 controller. So, technically, the board worked - but now I have to replace the first ASIC in order to make it work for real.

I refit the schfifty-three regulator to run 0.8-1.3V and tested balance on the board. It was pretty good. And then when the chips start to initialize, the bottom chip kicked up first and drowned itself out with the 200uF node-level caps I had. So I slapped a couple 1800uF 6.3V cans on there and that gave it enough buffer to fully initialize. But that's gonna be seriously annoying if I have to do that on the Amita.

Anyway, it's about lunchtime so I'm gonna head out. The parallel L-board is hooked up on the S5 now, back how it was this afternoon, and running at the 1BURGER if anyone's paying attention to that. I had hoped the string L would work and I could leave it going, but it won't be worth messing with until I get that chip replaced and figure out what went wrong with the RF line hosing up my 3.3V RX.

Oh, also, when it was working briefly off cgminer and the USB adapter, I was seeing about 800mA off the 5V line into the regulator and should have been getting 16.5GH so that's pretty balls. That 800mA is surely not accurate (20 year old analog gauge), but it was in the neighborhood.

Looking forward to this very much.
Nice work on these.

The how wasn't too difficult, with a shear and an hour with the soldering iron. The why is because it's a simple regulator that'll take in 5-12V (so I can test with USB or 12V power) and output 0.6-0.8V at 30A, which is really handy for testing miner chips.

I'd love to know more about that VRM and how/why you managed it.  I have one of those units sitting on a shelf turned off right now, and with <$200 resale value it seems

See above, but just as a reminder.. they do accept burgers;

Congratulation Sidehack and Novak : very happy for you... and for us :-).

When I have a final manufacturable design that's independently proven to work as advertised, and when I have secured a chip source. Not before both of those conditions are met.

Congratulations!  When will you accept our money?

That will absolutely not work, because these chips do not use SPI protocol. Or whatever it is you're asking - a string of controllers? I don't know what that means.

Also, when you say "that resistor on the L board", there are currently ten resistors on the L board but I don't think any of them are doing what you're asking about.