GekkoScience BM1384 Project Development Discussion

[ZenFr]

As soon as people stop asking that question.

:-)

[1713482044]

1713482044

[1713482044]

1713482044

[1713482044]

1713482044

[1713482044]

1713482044

[Rabinovitch]

Nice.

[PlanetCrypto]

As soon as people stop asking that question.

:-)

"Are we there yet?" 

[Biodom]

My first choice would be on a mountain in Wyoming, but I was just this evening discussing Alaska with some folks.

Also, I definitely did no real work this weekend. I mean I finally fixed the windows in my Jeep, but I didn't really work on miners at all. Tomorrow I may do some more testing on noise limits or something on the Compac, and work on the L-board for two-chip base Amita testing. Or I might just sit around and talk about awesome stuff with Novak all day. Who knows.

I've been to alaska once as a tourist-it is great in the summer, but probabbly cold as heck during the rest of the seasons.
Wyoming looks incredibly beautiful, though, even outside of Yellowstone.
to get to 15C=57F you'll have to climb above 9000 ft in Wyoming, and even then average max temp is ~70F.
You'll have to make a cabin just below Ganett's peak (13000 ft) to get that cold (57F).
http://www.wrds.uwyo.edu/sco/wyoclimate.html

[AJRGale]

As soon as people stop asking that question.

:-)

"Are we there yet?" 

one day before release, im going to ask this question, and i laugh evilly as everyone squirms and it gets pushed back about 2 weeks

[sidehack]

I just this evening lit up a dual-chip board - with two chips in parallel, not strung. It's definitely pulling the amount of power I'd expect for two chips actually hashing, but cgminer's only reporting single-chip speeds. Might be a comms issue with the second chip. The good news is I'm getting a lot better at sticking those things down. I'll play with it more tomorrow. Right now it's about time for me to take care of some other work so I can go home and get lunch before 10PM. Mondays are usually fairly administrative so we blew a lot of time earlier today talking about budgets and tool purchases and annoying people and stuff.

No pictures yet. Maybe tomorrow.


Okay, so the problem was pretty obvious when I actually looked for it. Com lines to the second chip were solder bridged. Apparently now it's cooking; I have 150MHz about 17GH reporting. It should settle out to 16.5GH overnight. If it keeps working well, tomorrow I'll rig one up with a second chip in series and see if it does anything better. If it works, I'll design the Amita PCB in parallel with the Compac retool and try to get prototypes of both at the same time. Hopefully that's in the works by the end of this week.

[philipma1957]

I just this evening lit up a dual-chip board - with two chips in parallel, not strung. It's definitely pulling the amount of power I'd expect for two chips actually hashing, but cgminer's only reporting single-chip speeds. Might be a comms issue with the second chip. The good news is I'm getting a lot better at sticking those things down. I'll play with it more tomorrow. Right now it's about time for me to take care of some other work so I can go home and get lunch before 10PM. Mondays are usually fairly administrative so we blew a lot of time earlier today talking about budgets and tool purchases and annoying people and stuff.

No pictures yet. Maybe tomorrow.


Okay, so the problem was pretty obvious when I actually looked for it. Com lines to the second chip were solder bridged. Apparently now it's cooking; I have 150MHz about 17GH reporting. It should settle out to 16.5GH overnight. If it keeps working well, tomorrow I'll rig one up with a second chip in series and see if it does anything better. If it works, I'll design the Amita PCB in parallel with the Compac retool and try to get prototypes of both at the same time. Hopefully that's in the works by the end of this week.

16.5gh   or 17gh are both very nice.

[sidehack]

Well, the error rate was stupid high again and poolside looks like garbage so I'm gonna call that experiment a failure.

[cavaliersrus]

well hack keep up the good work man im enjoying seeing the updates and it was very cool to see your inital miner in person cant wait to see some of the new changes one of these days

[sloopy]

My first choice would be on a mountain in Wyoming, but I was just this evening discussing Alaska with some folks.

Also, I definitely did no real work this weekend. I mean I finally fixed the windows in my Jeep, but I didn't really work on miners at all. Tomorrow I may do some more testing on noise limits or something on the Compac, and work on the L-board for two-chip base Amita testing. Or I might just sit around and talk about awesome stuff with Novak all day. Who knows.

I have told my kids and anyone who will listen how beautiful Wyoming is. I spent a 4th of July weekend tromping through the Grand Tetons. It could easily be spiritual.

I lived in Livingston, Montana for about 3 months. Too cold. I like it in the 60s and it is an amazing thing to walk out your door and see land in that way.

I live in Oklahoma, and it has been the most unusual place I have ever lived. Tornadoes don't bother me, I grew up with them in another state so I am used to them, but the weather changes are extremely unusual. I was told today we have also overtaken California as the earthquake capitol of the US. I have an app which shows the earthquakes for a variable amount of time, say you want to see the significant geological activity for a week displayed by some sweet graphics, Oklahoma looks like throwing rocks in water.

Sidehack you and Novac sound like you have some of the best jobs int he world. I used to say I would like to work in a beta datacenter for a huge tech company, but I was wrong. You guys have the best jobs.

If you guys decide to start keeping an order list please put me on it. I want at least 2 of each USBminers and anything bigger. If you sell boards, I want those as well if possible.

I will not ask the question that other guy asked either regarding lengths of time  
 

 

[AJRGale]

Well, the error rate was stupid high again and poolside looks like garbage so I'm gonna call that experiment a failure.

Hey, thats a good start, i wouldn't be surprised that's its another simple little thing again, maybe the com lines crossed? didn't tweak your tong the right way when you soldered them down?

[sidehack]

I like to think we have the best jobs. I was just asked last night, if I could do the same job for a big private company and get paid a heck of a lot more, would I? Heck no. Right now I don't get paid for crap but my bills are covered so it's enough. The real benefit is I get to work on things that I think are awesome, instead of things that some old fat guy in a suit gets told is awesome by people with too much money. (And then he gets paid 500 times more than I do to make those decisions - seriously, median US CEO wage is approximately median US household daily income per minute)

The lousy part of the job is not having the resources to do things bigger, or faster, or sometimes (as the case with some upgrades and new designs for our PSU boards), at all. But that's coming. We're free, and the future is bright.

Also, I don't think "too cold" is really a thing. This past winter was pretty mild by my standards, but the winter before it got down to about -10 for a week or so. The average winter where I live now is actually less cold (and a lot less snowy) than I'd prefer. The summers here are way too hot (100F and 90+% humidity isn't uncommon). I'm physically optimized for cooler weather. Novak, however, is definitely built for more tropical climes. Grumble grumble compromise...

[ZenFr]

I have 150MHz about 17GH reporting. It should settle out to 16.5GH overnight. If it keeps working well, tomorrow I'll rig one up with a second chip in series and see if it does anything better. If it works, I'll design the Amita PCB in parallel with the Compac retool and try to get prototypes of both at the same time. Hopefully that's in the works by the end of this week.

The stick with 2 chips is a very nice project : 17 GH for 6 W, less :-), more :-( ?
Good job :-).

[sidehack]

Mostly sadface. Check the speed (http://eligius.st/~wizkid057/newstats/userstats.php/1BURGERAXHH6Yi6LRybRJK7ybEm5m5HwTr) - it's running the same as a single chip. I don't know if it's a hardware or software issue yet (I need to log the cgminer debug and see if Novak wants to look at it) but it looks like the second chip's output is pretty much entirely errors. I'm hoping it's a cgminer thing and not a hardware thing.

Oh, maybe I should see if I can't run the testboard off an S5 controller with the proper code for those chips. Right now it enumerates as an S3. I don't know if they made any protocol changes from one chip to the next that might alter how upstream shares are packaged. Be pretty cool if that works.

[sidehack]

So, we did some more checking with the L-board (two chip board) and it looks like the comically high HW errors being reported is because of duplicate shares cgminer tallies as errors. I was working on getting the thing to light up off an S5 controller to see if it might be software-related, but was having no luck. I can see a brief TX signal sent to the L-board, and a flash on my LED indicates it briefly responds, but I see no evidence of an attempted initialization. Could be a wiring issue or who knows what else. I see zero chips displayed on the stats page.

I did, however, figure out some ways to get my CP2102 RX and TX down to the BM1384 logic levels better and using fewer parts than are currently on the Compac. I need to work on cleaning up the regulator output and redesign the PCB with regulator changes and I'll toss in those comms changes as well - they're currently implemented on the L-board and hashing away.

Hopefully we can figure out if it's a hardware issue or software issue causing the duplicate shares. The two chips on the board should be addressed differently, and it enumerates in cgminer as a U3 which is a 4-chip device, so unless Bitmain changed how work is divided amongst the chips between BM1382 and BM1384 protocols, I don't really know what's up. I'm really hoping it's a software thing, but until we get it figured out that stalls Amita and TypeZero dev.

[ZenFr]

We hope too that's a software issue.
Good luck.

[HolgerDansk]

I've been offline for a few weeks (by my choice) and am thrilled to see the progress.

Thank you Sidehack & Novak for continuing to work through the development of such a great concept!

[chiguireitor]

Hopefully we can figure out if it's a hardware issue or software issue causing the duplicate shares. The two chips on the board should be addressed differently, and it enumerates in cgminer as a U3 which is a 4-chip device, so unless Bitmain changed how work is divided amongst the chips between BM1382 and BM1384 protocols, I don't really know what's up. I'm really hoping it's a software thing, but until we get it figured out that stalls Amita and TypeZero dev.

I think i read somewhere that the string design needs you to put them "stringed" so the data bus serves as a kind of push register, so you put one chip after another, and push two works to one chip and it just takes the first work and pushes it downstream to the next chip.

Also, i could be like totally wrong.

[sidehack]

Looks like a protocol issue. I currently have my L-board working off an S5 controller, spitting out 16.5GH at 150MHz with, so far (17 minutes) zero HW errors and a butt-ton of accepted shares on a low-diff pool. I'll shift it over to the 1Burger Eligius account here in a bit and get some good stats visible.

I'm not sure what the U3 code set the baudrate to, but I know the S5 runs 115200. The pulses coming back off my chips for submitted shares aren't triggering the LED flash for near as long, so depending on what we do for a final cgminer driver I might have to tweak that circuit a bit.

Novak's looking at code now to see what the differences are between S5 (BM1384) driving and U3 (BM1382) driving. I'll continue dev on the L-board using an S5 controller because it appears to work. I need to work up another L-board with the second chip strung instead of parallel, and if it works we basically have the layout for the Amita data-wise. I'll also need to do some more tweaking to the Compac regulator design and see if I can't knock out some of the noise.

[sidehack]

Surprisingly enough, this works perfectly.

What we have is our L-board, with some modifications - 5V is now tied directly to 3.3V, reset is now controller-driven and the whole UART level shifting is jerryrigged. The thing is powered off a hacked-up BTCGarden AM-V1 VRM, which is also forwarding 12VDC into an S5 controller board. I've had to hack up the cable a bit to make it think there's a proper board plugged in, but the thing no mines away just fine in standalone.

I've got to assemble a couple more L-boards (one for Novak to use for software testing, and one with series chips so I can get that ironed out) but for the time being I'll just leave this guy mining. First fully successful test of a two-chip mining board. If I didn't care about the extra couple percent efficiency boost of running series chips instead of parallel (or the necessary dev step of mastering string'd comms and power) this could become an Amita. But the point of the Amita is to demonstrate string topology.

Tell you guys what, if I wasn't doing stickminers and went straight for the TypeZero it'd probably be done already. I never expected to have lost so much time figuring out particulars for that daggum Compac. Hopefully I have a final design done for that soon.