GekkoScience BM1384 Project Development Discussion

[sidehack]

No, that was just the four sticks at 150MHz. Looks like the Pi reset itself sometime, because when I came in this morning my terminals were logged out but the hardware was up and running. Restarted cgminer and I've been working on figuring out problem sticks all morning.

The one that was drawing too much power, I swapped the chip and now it's working fine (starts at 150MHz at 615mV like it should, instead of 660). The other one that wasn't lighting up at all, well I swapped its chip and reseated a couple times and finally gave up and stole its parts to make a new stick. And then that one didn't work either. And then I realized the reset wasn't connected on the first board and I guaran-darn-tee it that was the problem. It's only about the third time I haven't noticed exactly that problem before. So now I have two sticks which almost work and hopefully they'll be functional within the hour. That'll actually give me 10 sticks since I have two prototypes still, which I can run at 200MHz and bust out 110GH off what, 45W with a fan? It's cool, I know you're jealous.

[1714213458]

1714213458

[1714213458]

1714213458

[1714213458]

1714213458

[1714213458]

1714213458

[1714213458]

1714213458

[valkir]

110gh 45w where to I sign! 

[Mikestang]

Yea, that's very impressive!  So long, Antminer U3, hello GekkoScience!

Plus how cool will it be when we're running gear built by those in our community and not some overseas corporation?!

[WBF1]

Yea, that's very impressive!  So long, Antminer U3, hello GekkoScience!

Plus how cool will it be when we're running gear built by those in our community and not some overseas corporation?!

On that note, is there any plan (even a twinkle in the eye) for some form factor here along the lines of a U3 or R-Box?

[sidehack]

Just to make Phil (and everyone else) jealous...



I now have eleven test sticks working. I bet you'll never guess what the problem ended up being with the last stick I put together? If you guessed "it was also the reset cap not connected proper" well, shut up. I found it sooner this time, at least.

So this here is a ten-port hub, got the Pi running off one port and the other nine are sticks. cgminer is set at 200MHz so this setup should be spitting out 99GH all told. It's pointed at the 1BURGER on Eligius, which I noticed it defaulted to 512 diff instead of 128 which sucks because I finally have a rig that could run clean on a 128 diff. Dernit.

The eight gold-sink units are the ones I'll be sending out for test/review/opinions probably on Monday. I have addresses for seven of my eight guys, and if I haven't heard back from the eighth yet in a couple days I'll probably send one to... well, I should probably get one to the German guy that wants to build 'em for Europe.

The final version heatsink will be green. I'm okay with that.

Regarding that thing WBF1 just said, no not really. At least not directly. The 18-chip board by itself, with a heatsink and quiet fan, would be a good direct replacement for the New R-Box niche. At this time we're not planning on manufacturing a setup like that, but we would provide boards to anyone wanting to make some kind of kit (unless we change our minds and/or have the resources to do it ourselves). The 18-board (TypeZero Spec1, to be precise) would run between about 50 and 180 watts, pushing 150 to 360 gigahash. The power and control interfacing are pretty much the same as the New R-Box - a single USB connection and your 12V power (though really it could probably run between 8V and 24V in if I design it like I intend to).

[chiguireitor]

Dayum sidehack! 1430 GH/s with the same wattage as the S5 sounds impressive!

I'm giggling here just thinking how much more performance will you get when the BM1386 gets out (crossing fingers for same pinout/configuration).

EDIT: Just to clarify:

590w / 45w = 13.1111
13 x 110GH/s = 1430 GH/s

[Dexter770221]

...
 If you guessed "it was also the reset cap not connected proper" well, shut up.
...

Learned that lesson some time ago.... With ATmega uC's...
Painful lesson...

[alh]

Is the whole setup above (Fan + Pi + Hub + Sticks) all off a single power supply?

If you get a chance, it might be interesting to get a "Kila-Watt" view of the power consumption at the wall. That's the best comparison to an S5 I would think.

Look like a great progress to me!

[philipma1957]

No, that was just the four sticks at 150MHz. Looks like the Pi reset itself sometime, because when I came in this morning my terminals were logged out but the hardware was up and running. Restarted cgminer and I've been working on figuring out problem sticks all morning.

The one that was drawing too much power, I swapped the chip and now it's working fine (starts at 150MHz at 615mV like it should, instead of 660). The other one that wasn't lighting up at all, well I swapped its chip and reseated a couple times and finally gave up and stole its parts to make a new stick. And then that one didn't work either. And then I realized the reset wasn't connected on the first board and I guaran-darn-tee it that was the problem. It's only about the third time I haven't noticed exactly that problem before. So now I have two sticks which almost work and hopefully they'll be functional within the hour. That'll actually give me 10 sticks since I have two prototypes still, which I can run at 200MHz and bust out 110GH off what, 45W with a fan? It's cool, I know you're jealous.

yeah I am jealous.  more of your abilities building them then running them.  But I am also a bit jealous of that.

[sobe-it]

No, that was just the four sticks at 150MHz. Looks like the Pi reset itself sometime, because when I came in this morning my terminals were logged out but the hardware was up and running. Restarted cgminer and I've been working on figuring out problem sticks all morning.

The one that was drawing too much power, I swapped the chip and now it's working fine (starts at 150MHz at 615mV like it should, instead of 660). The other one that wasn't lighting up at all, well I swapped its chip and reseated a couple times and finally gave up and stole its parts to make a new stick. And then that one didn't work either. And then I realized the reset wasn't connected on the first board and I guaran-darn-tee it that was the problem. It's only about the third time I haven't noticed exactly that problem before. So now I have two sticks which almost work and hopefully they'll be functional within the hour. That'll actually give me 10 sticks since I have two prototypes still, which I can run at 200MHz and bust out 110GH off what, 45W with a fan? It's cool, I know you're jealous.

yeah I am jealous.  more of your abilities building them then running them.  But I am also a bit jealous of that.

Yea, I'd love to build/design some too. Building isn't that hard, its kind of like a model car (testing /working/debug is the tricky part) I have messed around in pcb design programs but never too serious with it, more breadboard type stuff.

Edit:
Are those antminer u2 heat sinks? The leds look about as bright as the og B.E.'s

[sidehack]

Actually, building took a lot longer than it should have. When we have a good stencil system for solder paste and a precise machine for rapid and repeatable placement of parts, most of the issues I ran into with mine will be gone. I have a V0.3 stencil that's about 95% right for the V0.4 PCB (the big cap and maybe two other things were changed) but the jerryrigged block and my crappy home-made squeegee aren't really good enough for a production environment, which leads to problems like the reset cap not being connected (on at least two sticks of nine) and gates on low-side FETs not being connected (on two sticks of nine). The BM1384 appears to have some sort of failsafe shunt in it that keeps Vcore in the neighborhood of 1V, because when the low side gates were left floating on a couple sticks (whoops solder paste) the output voltages on the buck were 930mV and 1120mV on 'em until I pulled the ASICs, at which point it rapidly drifted to out of range on the scope (but presumably 5V, I didn't bother checking where it actually went). That's kinda nice, because I reckon it means these chips might survive even a severe overvolt from a shorted node in a string depending on how things jump around. Or it could have just been a fluke and the relatively low current provided by the buck didn't cause them to push any higher. But yeah, it should not have taken me two days to get nine working sticks assembled.

Those are not U2 heatsinks but they are about the same size. I think the LEDs are pretty sexy and it's really nice watching them dance when they're all lined up like that. I'll change it up a bit on the Amita though, show you something different. Also yes the picture was taken exactly as number 8 found a share, hence the different color of the blink.

Yes they're all off one PSU. I was going to whip out one of my spare 30A 5V supplies but I figured heck, it's really only going to need about 7A so I scrounged a crappy ATX out of the pile. Which was handy for a 12V fan also. I actually do not own a kill-a-watt, but I reckon I could build one with the stuff I have laying around. Except the line power meters I have are for 200A lines so there's no decimal resolution. Maybe I should just buy a kill-a-watt.

[sobe-it]

Actually, building took a lot longer than it should have. When we have a good stencil system for solder paste and a precise machine for rapid and repeatable placement of parts, most of the issues I ran into with mine will be gone. I have a V0.3 stencil that's about 95% right for the V0.4 PCB (the big cap and maybe two other things were changed) but the jerryrigged block and my crappy home-made squeegee aren't really good enough for a production environment, which leads to problems like the reset cap not being connected (on at least two sticks of nine) and gates on low-side FETs not being connected (on two sticks of nine). The BM1384 appears to have some sort of failsafe shunt in it that keeps Vcore in the neighborhood of 1V, because when the low side gates were left floating on a couple sticks (whoops solder paste) the output voltages on the buck were 930mV and 1120mV on 'em until I pulled the ASICs, at which point it rapidly drifted to out of range on the scope (but presumably 5V, I didn't bother checking where it actually went). That's kinda nice, because I reckon it means these chips might survive even a severe overvolt from a shorted node in a string depending on how things jump around. Or it could have just been a fluke and the relatively low current provided by the buck didn't cause them to push any higher. But yeah, it should not have taken me two days to get nine working sticks assembled.

Those are not U2 heatsinks but they are about the same size. I think the LEDs are pretty sexy and it's really nice watching them dance when they're all lined up like that. I'll change it up a bit on the Amita though, show you something different. Also yes the picture was taken exactly as number 8 found a share, hence the different color of the blink.

Oh I know its not easy hand placing them small components (especially 603/402). Stare at it long enough while placing them you'd swear you were cross-eyed. I like the leds, love watching my antminer usb's dancing around as i fall asleep. Still looking for red fury's, saw a video on youtube and decided I have to get a few if/when they pop up. Been keeping myself busy for now getting an s2 blade running on a s1 control board. Next project is s3 blades on a s1 chassis with out the controller adapter.

[cavaliersrus]

hack if you get one more i would love to help the project out here at the house sir from a adverage person perspective

that is if somebody dosent answer

[philipma1957]

Actually, building took a lot longer than it should have. When we have a good stencil system for solder paste and a precise machine for rapid and repeatable placement of parts, most of the issues I ran into with mine will be gone. I have a V0.3 stencil that's about 95% right for the V0.4 PCB (the big cap and maybe two other things were changed) but the jerryrigged block and my crappy home-made squeegee aren't really good enough for a production environment, which leads to problems like the reset cap not being connected (on at least two sticks of nine) and gates on low-side FETs not being connected (on two sticks of nine). The BM1384 appears to have some sort of failsafe shunt in it that keeps Vcore in the neighborhood of 1V, because when the low side gates were left floating on a couple sticks (whoops solder paste) the output voltages on the buck were 930mV and 1120mV on 'em until I pulled the ASICs, at which point it rapidly drifted to out of range on the scope (but presumably 5V, I didn't bother checking where it actually went). That's kinda nice, because I reckon it means these chips might survive even a severe overvolt from a shorted node in a string depending on how things jump around. Or it could have just been a fluke and the relatively low current provided by the buck didn't cause them to push any higher. But yeah, it should not have taken me two days to get nine working sticks assembled.

Those are not U2 heatsinks but they are about the same size. I think the LEDs are pretty sexy and it's really nice watching them dance when they're all lined up like that. I'll change it up a bit on the Amita though, show you something different. Also yes the picture was taken exactly as number 8 found a share, hence the different color of the blink.

Oh I know its not easy hand placing them small components (especially 603/402). Stare at it long enough while placing them you'd swear you were cross-eyed. I like the leds, love watching my antminer usb's dancing around as i fall asleep. Still looking for red fury's, saw a video on youtube and decided I have to get a few if/when they pop up. Been keeping myself busy for now getting an s2 blade running on a s1 control board. Next project is s3 blades on a s1 chassis with out the controller adapter.

yeah red furies and ice furies where very nice looking sticks.

I should have kept 1 ice fury 1 red fury and 1 blue fury.  I would have had a red white and blue  set of sticks.

[PlanetCrypto]

Sweet!! 

So is that a calibrated piece of dimensional spruce? LOL
Looks alot like our test bench.

Don't see a trimmer pot for Vcore, course then my eye sight ain't so good.
Doing that in S/W?

Think I'd send one on over to Germany, and pass the addiction on into Europe.

Can't wait to see what ya' do with the 18 chip version.

Still in for 2 of the sticks when you get a round-2-it.

[valkir]

If you need a review from Canada, I will be more than happy to help!
Cant wait to see the 18 chips board.   

[philipma1957]

Sweet!!  

So is that a calibrated piece of dimensional spruce? LOL
Looks alot like our test bench.

Don't see a trimmer pot for Vcore, course then my eye sight ain't so good.
Doing that in S/W?

Think I'd send one on over to Germany, and pass the addiction on into Europe.

Can't wait to see what ya' do with the 18 chip version.

Still in for 2 of the sticks when you get a round-2-it.

the bottom corner appears to have a pot like mine did:

https://bitcointalk.org/index.php?topic=1086011.msg11730704#msg11730704

[TheRealSteve]

Don't see a trimmer pot for Vcore, course then my eye sight ain't so good.
Doing that in S/W?

No, it is a trimpot.  Probably easier to see here (bottom right corner):

pot is tiny and delicate not looking to turn it much.

[philipma1957]

Don't see a trimmer pot for Vcore, course then my eye sight ain't so good.
Doing that in S/W?

No, it is a trimpot.  Probably easier to see here (bottom right corner):

pot is tiny and delicate not looking to turn it much.

beat you to it !!

LOL!

[TheRealSteve]

beat you to it !!  LOL!

Bah, only because you keep ninja-editing your posts!