Topic: New Official AMT Thread - page 81. (Read 149472 times)

Zefir knows how to do it better then us. Also we havn't had too much success because our manufacturer may have sourced a different trimpot than the one which was due to be used. Tony, i'll send you an nda as well, that way you'll have the bomb and other design aspects as Fuzy and Isamhim.

Once I see some examples of what was done wrong, assuming the various connections are correct then at worst it will involve just making sure the right parts are where they belong, are the correct value vs what they say (or I feel they should be) and are soldered properly. Especially the A1 chips for both power feed and thermal reasons. Green mask errors can be addressed several ways, just need to make sure the A1 thermal contact is maintained. If there are signal/power noise/ripple issues they usually can be negated. Safe bet that rework will be needed.

As for cladding thickness and problems arising from it - we'll see wassup there and what can be done. Maybe run in eco-mode but provide another blade to make up for the lower hash rate? Would probably be a better alternative vs scrapping them.

... nice timing on this blog post. Deals with design verification while under the gun... http://www.edn.com/electronics-blogs/tales-from-the-cube/4430374/The-need-to-succeed?elq=ff788031dff34ef49b5ab56ebef3a41b&elqCampaignId=17206

@AMT

If you have any other details or documentation on the drivers that could be of use? Or resources, I am looking at the thread posted by zefir to figure out the HWreset. I did suggest another method that MIGHT work and that would be converting each SPI into its own USB channel SPI/USB. This MIGHT get around the problem. Its a dirty workaround. basically create dedicated channels for each card that extends using USB to SPI connectors. This will address the potenital backplane issue and possibly allow for the cards to work properly. Of course there is still the driver issue with HWresets. I am looking at that now to see if I can figure it out. Based on various conversations I had I started to suspect this was the issue for a number of boards. I just have to learn the needed info to actually fix it.

Bitmine and AMT gave a reference design and its free as well. Its a two chips board which does in fact work.  We also have 200 bare boards if anyone would like to purchase these bare boards and or chips from us, and we can cut a deal that will beat out cointerra any day.

Furthermore, its not that the miners dont work, its that the manufacture we chose to build to miners didnt do the job well. Because bitmine produces machines that work, and using their same model of bulk purchasing in order cut costs, we chose to do the same.And we trusted a manufacturer which didn't/couldn't get the job right,we're in the position. sitting around boxes of bad boards trying to figure out how to get them to work again. Lets take some pictures..

Thank you for assistance zefir and help. Zefir is of the few good hearted individuals within the bitcoin community that wanted to see bitcoin become something substantial for the world to use, rather than a 3 month roi on one's investment. He is an inspiration to this community and to us at AMT. Again thanks.

You are obviously not including paying anyone a salary or any other overhead. For a commercial (paid) endeavor sure I could if I had any interest in it. However, I *have* a career and it is not in making miners.

Years ago I would have said that any decent electronics design engineer could do it. Sadly, that no longer seems to be the case as more and more are not being taught *why* things work the way they do, just what to do with them. Understanding the 'why' gives insight into what to do when over-simplified models do not produce working physical designs.

Give the Cointerra reference design... you think you can come up with a reliable layout?

I am sure someone would be extremely interested considering price point... I mean... $100K can get you to almost 200 THs.   That can't be beat!

Again, a reference design is just that: a basic how-to example to make something work and test it. It is a only a starting point. It is NOT an optimized final design much less something ready to be stuffed into a complete miner eco-system . They will give you a list of things to look out for but not their precise solutions to it.

It is not just a matter of taking a bunch of boards and stuffing them in a case, sprinkling pixie dust on it and Volia' - it works. BMch/AMT, & HashFast tried that approach and we see where that led.

Well Cointerra sells their chips and will give you the reference design.  

It is actually a compelling deal... for $105,840  you get 525 chips.  Each chip is at 500GH/s nominal.

So if you were to build a farm.... $105K gets you  262 THs!   (Not including board manufacturing and power supplies)

Holy sh*t!   What they are saying is that you can build a 1THS for under $1,000.  

At $200 a chip in bulk.... they've considerably undercut Innosilicon at $45 per chip.   

Won't disagree at all on the commodity aspect and I am at a loss to explain the poor designs. What, did they (Bitmine.ch) hire some interns with not even 1st year design experience to make these rigs?

"If AMT had instead focused on prompt delivery and solid support,  then maybe they would not be in this mess"
Delivery of what? The earlier non-working Bitmine-based designs? Again, read the A1 dev thread. Look at when ANYBODY 1st had chips to experiment with. Look at when the 1st test boards were made. Now consider that BMch et-al had promised full blown systems - not test kits - to be in customer hands months before. There simply was literally zero time left for the commercial entities to make/properly test a salable system.

IP does matter. It matters very very much. IP is what will make one design work when others trying to use the same core parts don't work. Right now AMT is competing with HF, Dragon and to some extent Bitmine.ch to get a decent A1 design out the door. Competing against other chip systems is not exactly an issue yet compared with getting things to work and be suitable for large production runs.

Bitmaintech will provide design guidelines if you want to use their chips but they will NOT give you their sales product designs. Same with Cointerra. Even Technobit licensed their board design to Bitmine/AMT. They did not just give it away and it is a damn safe bet that if Technobit sold systems vs modules/kits they would not have helped at all.

Swing and a miss.
The NDA will be because specific product IP is needed for me and ISA to do the voodoo we do so well. One can only go so far with visuals and conjecture based on the 1 & 2 chip reference boards and schematics. Obviously many things by necessity have to be different with the 8-chip setups. I've read the dev and DIY threads from start to finish before. Those are what pointed me at the various things I've been bringing up here as problems being seen sound a lot like what those threads covered. Somewhere between a reasonably good start and scaling to production systems seems that a lot of A1 company designers fell victim to the NIH attitude and tried to re-write the rules. Even the Technobit designs are still being tinkered with and discovered (and nice job on the power/speed work in their part of the Forum Losh...) but then that seems like what they were made for: stand-alone miners and tinkering toys, not piled in a common box as systems.

As with Hashfast, Black Arrow and others in the A1 game things for them have moved past the community-only aspect. Business is business and you can't sell systems unless you can differentiate your product from the others and that takes IP. Or at least systems that work...

Thanks mdiggy, that is very nice of you... I actually ordered some from where notfuzzywarm suggested. I have ordered the double sided 10 pin connector if anyone needs those.

Likely stepped on the lawyers toes. At this point I just hope it all gets worked out for everyone. I am trying for my part to hopefully salvage things in favor of the people who got something broken. And maybe in the process try to make sure this does not happen again to others. I do think they tried to make this work, but just failed on alot of levels. I get a bit pissed sometimes thinking about it, because its my 11k that got pissed away. So I like others have that axe to grind. I am sure AMT is keenly aware of that. Hopefully soon we hear about what will be done to make this all right for us.

Please read my related post: https://bitcointalksearch.org/topic/m.6536840


Aside from that, if you are struggling with technical problems or want to discuss them, you should visit the above A1 DIY thread with better chances to meet other developers and get your issues resolved.


Cheers,
zefir

Can't disagree there. But... Community & business rarely make good bedfellows.

I design very very expensive industrial systems for a living.
 A good part of their market is for chip making. Because the product is in front of the bleeding edge I am often brought in nearly from day-one of our customers customer asking, "Can we do this...?". (Usually after informing them of what it will entail - yes. I love challenges).

The folks who commissioned the ASIC's be they Bitmine/InnoSilicon, Cointerra, BFL or whoever aside from Bitmaintech seem to have gone to the Underpants Gnomes school of business. Look it up. It's actually an oft used term

In this case it goes:
step-1: Make chips for miners
step-2: ? ? ?
step-3: Profit!

The step-2 that is missing is design what *uses* the chip.

Point is I am damn certain of what will need to be addressed for my equipment & process to work 24x7x365 for years and often decades. Why am I certain? Because I and my company have been constantly researching future needs and at our expense already done all the groundwork. My customers are telling me they want to spend several million dollars a year bringing in our systems and I need to be certain the design will work. Period.

Yes chips delay hurt massively but more to the point those selling chips did nothing to actually *use* the chips in-house. After that release true engineering samples along with test boards. IF I have the timeline for the A1 right it went from a 1 to 2 chip reference pretty much conceived of/gotten to work in the dev & DIY threads in about a month (Jan-Feb) and then straight to 'production' with 8 chips. Of course the 1st versions have problems. eg. the multiple SPI slave chain matter. The entire design was never characterized to begin with. Just get it running... I understand the pressures that the pre-order syndrome brought on but hell, they threw a baby into battle!

At least it seems InnoSilicon/Dragon & Technobit saw what they actually had to deal with and addressed it better then those who relied on the BitMine design team.

I will stay busy reading all those new links just posted, lots of good stuff therein,,,,
Did find the reason all are struggling, noone wanted to share info, till lately.
with a 6month lifespan on high end equiptment, they have no time to act like that.
Community shared it would have gone much faster.

Thanks. All are  8-chip boards. Will be checking all levels I can including downstream SPI signals. Hopefully the A1 buffers things between the 2 side of it.

I take it option #1 is no longer possible seeing how another 2 weeks have passed since that "end of this week" deadline.

Quote zefir:
Step-by-Step Bring-up Process
1) Physical
Most is depicted in the above figure, this is the prose version:

    chip is 1.8V only => use level shifter for all signals from/to host SPI interface
    VDD needs to be ~820-850mV with a max. ripple of 70mV (pilot run chips do not support undervolting)
    AVDD needs to be 1.8V with a max. ripple of 200mV
    power-up PLL settings are based on 12MHz reference clock; if you use a higher value do not start hashing without reducing system clock via PLL or you risk bricking the chip by overclocking it.
    if you have a multi-chip board, use a clock distribution device to drive them with a single oscillator
    heat-sinks on both sides of the chip needed, monitor and ensure surface temperature does not exceed 50°C
    HW reset is mandatory; RSTN needs to be pulled low for at least one second; ensure it was released for at least one second before the first command is issued