GekkoScience BM1384 Project Development Discussion

[sidehack]

Thanks. I think I'm gonna toy with the regulator test PCB design today and see what pops out.

[philipma1957]

Thanks. I think I'm gonna toy with the regulator test PCB design today and see what pops out.

Let us know if you can get under .4 watts per gh with it.  thanks again.

[sidehack]

This is for the one-chip stick miner. Even given that the VRM will probably be at best 90% efficient, it'll probably be possible to get <0.4W/GH board-level at reasonable clock speeds. The stock setting I'm shooting for getting 8GH at 2.5W which is about 0.3W/GH board-level. I won't actually know how well the buck circuit performs until I've had a chance to assemble and test it, which will wait on parts so end of next week. And I won't know board-level expected W/GH until the week after when I have the BM1384 breakout board in hand and testing.

I'm verifying the PCB layout now. It's a lame-O 2-layer so not terribly dense but Novak'll be toner-transfer etching it and we're not quite magicians yet. Hopefully it works as expected when we get to testing it.

[sidehack]

Okay, 13 hours is enough for one day for PCB design. Regulator test board is done and Novak might be etching tomorrow, so that's cool. And I've got almost one square inch of the stickminer laid out, which means the entire comms and Vcore regulator. Still got a couple LDOs to tuck in the corner but there's plenty of room, and then the actual ASIC and its associated parts. I probably won't get back to it until Monday night or Tuesday, but I should have enough info to spec a heatsink after a couple more hours of work.

This is probably the most enjoyable project I've ever worked on. We gotta have a giant cheeseburger party or something when we get working stickminers done. It's very exciting.

[Unacceptable]

Okay, 13 hours is enough for one day for PCB design. Regulator test board is done and Novak might be etching tomorrow, so that's cool. And I've got almost one square inch of the stickminer laid out, which means the entire comms and Vcore regulator. Still got a couple LDOs to tuck in the corner but there's plenty of room, and then the actual ASIC and its associated parts. I probably won't get back to it until Monday night or Tuesday, but I should have enough info to spec a heatsink after a couple more hours of work.

This is probably the most enjoyable project I've ever worked on. We gotta have a giant cheeseburger party or something when we get working stickminers done. It's very exciting.

Curmudgeon on!!!!!!!!!! 

[sidehack]

I'm sorry, sir - but that comment was...

*puts on sunglasses*

...Unacceptable.

(nothing really meant by it, I've just been wanting to say that for a while because it's funny)

[Unacceptable]

I'm sorry, sir - but that comment was...

*puts on sunglasses*

...Unacceptable.

(nothing really meant by it, I've just been wanting to say that for a while because it's funny)

   Exactly what I was looking for!! 

I'm not thin skinned,like some folks,in fact if I don't get a "Fuck You" daily from them,they're NOT my friends!! 

[Finksy]

Nice work sidehack, going to be interesting to see the results. 

[valkir]

Okay, 13 hours is enough for one day for PCB design...

Nice work man! Keep it up and a cheeseburger party for sure! 

[sidehack]

I actually need to spend a couple days on other things. Unfortunately I've been shirking my duties in order to put in those long hours because it's actually really fun. But my house is a mess and I have hosting customers waiting on router configs and... um... gotta sleep sometime? I'm most satisfied at work when I have a lot of different things going on, so the past few weeks I've been swimmin' in it. A lot of the last year has been spent toying with projects we wouldn't have the money to build, or waiting on customers to get off their butts and TCB, or boring but essential manufacturing tasks. I'm really glad to have a complex and entertaining project, and it's even more satisfying to see other people getting behind it.

[daddyfatsax]

Sleep?? Who needs sleep? Apparently you don't. Posting at 2:40 am and then barely 8 hours later. I can't really convey a fake scolding tone over the internet, but just pretend.

[sidehack]

Pshaw. Novak's the king of not sleeping. He has some pretty great stories of consecutive all-nighters and such. Every time I come in and say "yeah I was up until 2AM" he just kinda shrugs and says "I was up until 5".

[valkir]

ahaha you guys are not human! You are a machine! 

[ManeBjorn]

I am looking forward to these. I want to review them as well for you.
Keep up the good work and I am glad it has been fun. 

Okay, 13 hours is enough for one day for PCB design. Regulator test board is done and Novak might be etching tomorrow, so that's cool. And I've got almost one square inch of the stickminer laid out, which means the entire comms and Vcore regulator. Still got a couple LDOs to tuck in the corner but there's plenty of room, and then the actual ASIC and its associated parts. I probably won't get back to it until Monday night or Tuesday, but I should have enough info to spec a heatsink after a couple more hours of work.

This is probably the most enjoyable project I've ever worked on. We gotta have a giant cheeseburger party or something when we get working stickminers done. It's very exciting.

[HerbPean]

Pshaw. Novak's the king of not sleeping. He has some pretty great stories of consecutive all-nighters and such. Every time I come in and say "yeah I was up until 2AM" he just kinda shrugs and says "I was up until 5".

Hahahahha

Great project, thanks for sharing !

[sidehack]

Also, Novak doesn't really trust the stability of a single-chip-wide string so I think we're gonna put a per-node overvolt checker on the Amita that will keep track of each chip's Vcore and if it goes over a safety threshold (0.8-0.85V probably) will kick off the regulator and kill power to the circuit. There should be enough capacitance to buffer out momentary jitters but that should prevent it from going Full Prisma™ and blowing up. If it works well enough I'll probably figure out how to stick it on the TypeZero boards. It adds complexity to the system, which we're generally not in favor of, but adding modular complexity in a way that increases overall reliability (instead of decreasing it, as complexity tends to do) is not a bad thing.

[EDIT]

First look at the Compac PCB layout. I'm probably gonna have to adjust a few things because I just realized I forgot to wrangle the logic level shifter outputs, but other than that it's the entire adjustable VCore regulator circuit, both LDOs for BM1384's logic and PLL voltages, the oscillator, USB communication and some status LEDs in one square inch. Holy balls I hope it works. The top half will be much simpler, and almost entirely under a heatsink.

[jekecoin]

First look at the Compac PCB layout. I'm probably gonna have to adjust a few things because I just realized I forgot to wrangle the logic level shifter outputs, but other than that it's the entire adjustable VCore regulator circuit, both LDOs for BM1384's logic and PLL voltages, the oscillator, USB communication and some status LEDs in one square inch. Holy balls I hope it works. The top half will be much simpler, and almost entirely under a heatsink.

Wow, that's great sire, I need it on my hand.

[notlist3d]

This is for the one-chip stick miner. Even given that the VRM will probably be at best 90% efficient, it'll probably be possible to get <0.4W/GH board-level at reasonable clock speeds. The stock setting I'm shooting for getting 8GH at 2.5W which is about 0.3W/GH board-level. I won't actually know how well the buck circuit performs until I've had a chance to assemble and test it, which will wait on parts so end of next week. And I won't know board-level expected W/GH until the week after when I have the BM1384 breakout board in hand and testing.

I'm verifying the PCB layout now. It's a lame-O 2-layer so not terribly dense but Novak'll be toner-transfer etching it and we're not quite magicians yet. Hopefully it works as expected when we get to testing it.

I would love to see you squeeze that number in efficiency, would be a accomplishment.

Still enjoying keeping up in thread.  Can see lots of work going on.  Very fun to look at as most seem so secretive on development.

[sidehack]

Yeah I probably won't be quite as active with sharing fine details on the TypeZero (there's a few aspects of it we're gonna keep in our pockets until the official release) but we'll definitely keep folks posted on the project development as it develops. Without the community there's no reason to do any of this anyway. One of my goals, in general, is to run a business without being a greedy bastard and prove to greedy bastards that it can be done successfully. One primary aspect of not being a greedy bastard is deliberately not ripping people off, and deliberately avoiding scenarios where even accidentally ripping people off is possible or likely, so being simultaneously self-sufficient and fairly transparent is conducive to all of that.

[RadekG]

Also, Novak doesn't really trust the stability of a single-chip-wide string so I think we're gonna put a per-node overvolt checker on the Amita that will keep track of each chip's Vcore and if it goes over a safety threshold (0.8-0.85V probably) will kick off the regulator and kill power to the circuit. There should be enough capacitance to buffer out momentary jitters but that should prevent it from going Full Prisma™ and blowing up. If it works well enough I'll probably figure out how to stick it on the TypeZero boards. It adds complexity to the system, which we're generally not in favor of, but adding modular complexity in a way that increases overall reliability (instead of decreasing it, as complexity tends to do) is not a bad thing.

What about a passive way? Just connect either two parallel schottky or single universal diode. ES2(A,F or J suffix) has Ifsm=50A and can be choosed for right Vmax for each chip. ES2A has 40mA at 0.6V = 0.024W per chip additional loss, while at 1V (still safe) it reaches 5A. It is not 100% protection, but it should perfectly balance each chip's voltage. Only disadvantage is that treshold voltage depends on temperature, so cold and hot Vmax will be different a bit. I think if you look around you can find diode with better V/I and V/Temp characteristic.