Klondike - 16 chip ASIC Open Source Board - Preliminary

[sensei]

i understand this right

30 board(about 5gh each) at $5 each?

so $150 for 150gh?

K16 with 16 Avalon chips = 16 x 282 MH/s = 4.5 GH/s
K64 with 64 Avalon chips = 64 x 282 MH/s = 18.05 GH/s

Many are overclocking

K16 with 16 Avalon chips = 16 x 300 MH/s = 4.8 GH/s
K64 with 64 Avalon chips = 64 x 300 MH/s = 19.2 GH/s

[dani]

i understand this right

30 board(about 5gh each) at $5 each?

so $150 for 150gh?

K16 with 16 Avalon chips = 16 x 282 MH/s = 4.5 GH/s
K64 with 64 Avalon chips = 64 x 282 MH/s = 72.2 GH/s

Many are overclocking

K16 with 16 Avalon chips = 16 x 300 MH/s = 4.8 GH/s
K64 with 64 Avalon chips = 64 x 300 MH/s = 76.8 GH/s

64 x282 = 18GH/s

dont know where you get 72..

[bassclef]

Yeah check that math.

I wonder how feasible overclocking will be. Obviously you could compensate with better cooling/heatsink design. Would one have to manually adjust the voltage, or is it software controlled?

[sensei]

Right. That's 16 of the K16's. We need only 4 to get the equivalent of a K16.

Sorry

[sensei]

Yeah check that math.

I wonder how feasible overclocking will be. Obviously you could compensate with better cooling/heatsink design. Would one have to manually adjust the voltage, or is it software controlled?

It looks to be a configuration word sent to the board. I think the ASIC has an internal oscillator.

I should also say that bkkcoins will likely correct me very soon.

[BkkCoins]

Yeah check that math.

I wonder how feasible overclocking will be. Obviously you could compensate with better cooling/heatsink design. Would one have to manually adjust the voltage, or is it software controlled?

It looks to be a configuration word sent to the board. I think the ASIC has an internal oscillator.

It appears the ASIC has a PLL (phase locked-loop) on board to multiply the clock frequency. It takes 32Mhz input and manages to generate a higher speed clock internally that is selectable by a config word.

The supply voltage is not dynamically settable but can be controlled by a suitable resistor change. The purpose of changing the voltage would be to reduce power use and heat, and with less heat more stability at a higher speed. This will have to be determined once a working board exists and optimal values are chosen. It may be feasible to add a digitally controlled resistor to make the voltage adjustable but I'm not designing that in initially as it would be yet another variable to factor into debugging/testing, and potentially due to switching power supply design issues could cause headaches. When there's time later I may look at testing the idea.

[SebastianJu]

psu no problem

ok for fans and heatsink also, they don't cost too much
overall it's still very cheap

$$$ for Avalon chips most expensive part... scroll back through the thread mentioned several times. Also list of BOM in his github in PDF format.
https://github.com/bkkcoins/klondike/blob/master/docs/Parts%20List.pdf


$20.02 per board for 100+ board run his costs.

Also he has design up for Klego... for a 256 set up for daisy chain connected boards. A 40 x 40 configuration. So people with larger orders might be interested in that solution.

https://github.com/bkkcoins/klondike/blob/master/docs/KLego.pdf

Thanks for the pdfs... i didnt check his github yet. The KLego... is it for 16 chip-pcbs or for 64 chip pcbs? I have a bigger number of chips ordered too and i would like to go the cheapest and less work route of course.

Scrolling through the deal I think would be this Sebastian

20 x 20 boards that have 4 x 16 = K64

Those can then be layered into a 40 x 40 and daisy chained into a K256 chip monster.

Then set up another layer and stack and stack and stack. Have to look back on what he posted on maximum but I have 540 chips so I will want two layers of 8 x K64's.

Ill add what BKKcoins posted in a sec.

I have a Project Manager project on mouser.com with all the parts in it. Apparently this can be shared but when I checked they require an email to send it to who you will share with. If someone wants that shared too them then PM me your email and I'll plug it in. and it can be accessed at this url:

http://www.mouser.com/ProjectManager/ProjectDetail.aspx?AccessID=b645e5b812

A K16 will use around 32W, so a K64 will need about 128W each (with 4 leads). They're designed to take a PCIe 6 pin connector much like a GPU. I would guess that you could run a K64 off each PCIe lead from the PSU with splitters to feed each of the 4 sections. I haven't gone to check the specs but they would  handle something like that. I wouldn't have a problem running 4x K64 off my 600W supply. I used to run 3 GPUs at 500W without any issues. I'd say you're better off going that route than using crappy low wattage power adapters for each board, or section of board.  Like with GPUs  stick with solid reliable PSU units like Corsair, preferably high efficiency Gold (90%) rated.

https://109.201.133.65/index.php?topic=190731.msg2060629#msg2060629

Note that the K64 wouldn't have internal pin hdrs. I would wire those one in the board either as tracks or actual wires, depending on how I panellize (in gerbers or in brd design).

https://109.201.133.65/index.php?topic=190731.msg2092711#msg2092711

Sounds great, because at one point i will have a good chunk of chips too.

What i wonder is, the K1 is similar to the Asicminer Erupter USB, the K16 is the standard setting and the K64 is 4 of these K16 put together. Whats the advantage of putting them together? Can parts be spared then?

I have read the recommendations about avalon miners and found this psu: Seasonic Platinum 1000
It seems to be a nice piece or work, you only need to put in the cables you need and it gets highest values in benchmarks. I think i will get one of these and hope it will be enough to run many or all diy-miners. Maybe, if needed, with y-cables. Unfortunately this psu never is sold cheap at ebay... i have an email alarm since a month without results...

What do you mean with layered into a 40x40? Double sided? So you put asics on both sides of the pcb? Wouldnt that mean a redesign is needed and more problems with the heat because both sides asics will heat up the counterpart too? Maybe you mean to put 4 of the K16 (each 20x20) put together and get one pcb that is 40x40? That would be the K64 then. but daisy chain? K256 would be 4 x K64 then? So 8x8 K16. I wonder if that really saves much parts. A K256 would mean 80x80 cm. Manageable probably. Maybe putting them in a rack like asicminer is doing it, putting some copper cooling units on the chips so that wind can flow freely through the copper and save space this way. I really would like to see such machine in my flat...

So at the end in reality there are only 2 models. The 1-chip miner and the 16 chip miner. Everything else is stacked together with the 16 chip miner. Then there must be a reasonable less needed parts when the K64 is $100 only while the K16 is $35 (4 x $35 = $140). Would be interesting to know if price drops for K256 or even a K576 with 3 x 3 K64 = 6x6 K16?

Now that i saw the pdf with the needed parts it looks like a good chunk of work again to create such miner. i hope my skills are sufficient.

[WynX]

Great work here BkkCoins, will definitely order when you have the DIY kit ready.

I've checked your repository and noticed something that made me wonder. Your order list is mentioning TDK C1005X7R1A104M (of which we'd need 174 pcs) in the 0402 package. When I check the TDK spec sheets, these are NOT 0.47uF as mentioned in the ordersheet, but 0.1uF. Is the spec off or do we need TDK C1005X6S1A474M? Note that in the latter case we'd move from thermal package X7R (-55 to +125ºC, ±15%) to X6S(-55 to 105ºC, ±22%).
Maybe I'm just off...hope that is the case as I'm obviously an EE amateur .

Edit: add spec link for your convenience: http://www.tdk.com/pdf/general_B11.pdf

[sensei]

I think bkkcoins is intending on creating a 40x40cm version of the k16, to be called the k64. It is essentially the 4 of the K16's, and will probably save on a few parts.

i.e. only one USB connecter, maybe no klego pins, maybe less fan controllers, one power supply instead of 4.

Again, bkkcoins will correct me if needed.

[sensei]

Great work here BkkCoins, will definitely order when you have the DIY kit ready.

I've checked your repository and noticed something that made me wonder. Your order list is mentioning TDK C1005X7R1A104M (of which we'd need 174 pcs) in the 0402 package. When I check the TDK spec sheets, these are NOT 0.47uF as mentioned in the ordersheet, but 0.1uF. Is the spec off or do we need TDK C1005X6S1A474M? Note that in the latter case we'd move from thermal package X7R (-55 to +125ºC, ±15%) to X6S(-55 to 105ºC, ±22%).
Maybe I'm just off...hope that is the case as I'm obviously an EE amateur .

Edit: add spec link for your convenience: http://www.tdk.com/pdf/general_B11.pdf

I noticed this also and substituted this one for it.

http://www.mouser.com/ProductDetail/TDK/C1005X7S1A474M/?qs=sGAEpiMZZMvQvaS66kI3TsXrxDUb6kFjGSZXVr8GtXc%3d

[WynX]

Great work here BkkCoins, will definitely order when you have the DIY kit ready.

I've checked your repository and noticed something that made me wonder. Your order list is mentioning TDK C1005X7R1A104M (of which we'd need 174 pcs) in the 0402 package. When I check the TDK spec sheets, these are NOT 0.47uF as mentioned in the ordersheet, but 0.1uF. Is the spec off or do we need TDK C1005X6S1A474M? Note that in the latter case we'd move from thermal package X7R (-55 to +125ºC, ±15%) to X6S(-55 to 105ºC, ±22%).
Maybe I'm just off...hope that is the case as I'm obviously an EE amateur .

Edit: add spec link for your convenience: http://www.tdk.com/pdf/general_B11.pdf

I noticed this also and substituted this one for it.

http://www.mouser.com/ProductDetail/TDK/C1005X7S1A474M/?qs=sGAEpiMZZMvQvaS66kI3TsXrxDUb6kFjGSZXVr8GtXc%3d

Ah yes thats better, nice. I've found also the C1005X7S1A474K, which is basically the same but with lower tolerances. Seems to be the same price though??

[KS]

Nice one on the Phoenix connector.

As for the laptop chargers: im unsure of what to use now. Many ATX PSU's I've had go high on the 12v when loaded really lightly and I don't exactly want to use one for a single board. What would you suggest?

The only supply I would feel good recommending is a high quality ATX one like a Corsair. Even their low-end CX series is good. The ATX should be outputting a 12V regulated voltage so it it's off by more than 5-10% then it's not doing it's job. The specs for the on board buck regs are that they can take up to 21V normally by design and 25V under worst case conditions. So unless you have something that's gone very wrong it probably is ok.

++

Better brands have tighter deviations. I think ATX calls for +/-5% (too lazy to Google it right now )

Are there not issues running at less than 10% capacity/load though? Always thought they needed at least 15% load to be stable.

http://en.wikipedia.org/wiki/ATX

Supply voltage is to be within 5% or you can't be certified (don't put it past the cheap chinese PSUs you can find on ebay to cheat with that).

I think you're mistaking efficiency with V-delta. Your 80+ Gold certified PSU won't be that golden if you run it at 5-10% load. Check the box, the Gold bit should be specified within a range of power output.

Fox ex., the Enermax Platimax 1500 box states "90-95% efficiency at 20-100% loads". So it's not Platinum rated below 20% load.

If you invest in some expensive equipment, don't go cheap on the PSU or you WILL regret it

[BkkCoins]

Great work here BkkCoins, will definitely order when you have the DIY kit ready.

I've checked your repository and noticed something that made me wonder. Your order list is mentioning TDK C1005X7R1A104M (of which we'd need 174 pcs) in the 0402 package. When I check the TDK spec sheets, these are NOT 0.47uF as mentioned in the ordersheet, but 0.1uF. Is the spec off or do we need TDK C1005X6S1A474M? Note that in the latter case we'd move from thermal package X7R (-55 to +125ºC, ±15%) to X6S(-55 to 105ºC, ±22%).
Maybe I'm just off...hope that is the case as I'm obviously an EE amateur .

Edit: add spec link for your convenience: http://www.tdk.com/pdf/general_B11.pdf

I noticed this also and substituted this one for it.

http://www.mouser.com/ProductDetail/TDK/C1005X7S1A474M/?qs=sGAEpiMZZMvQvaS66kI3TsXrxDUb6kFjGSZXVr8GtXc%3d

Ah yes thats better, nice. I've found also the C1005X7S1A474K, which is basically the same but with lower tolerances. Seems to be the same price though??

Sorry about that. Just sloppy when I updated the parts list. I changed the value but not the product code. I'll fix that now. Thank you for checking. Believe me, I'm not infallible.

[WynX]

Sorry about that. Just sloppy when I updated the parts list. I changed the value but not the product code. I'll fix that now. Thank you for checking. Believe me, I'm not infallible.

You're welcome, and that's exactly why open source = good

[btceic]

I have less than zero skills when it comes to DIY.

Any chance we can get a lego like kit sans the soldering, i.e. part A1 goes into part B33 type thing?

I have tried soldering, it didn't turn out so well..........

[torzsy]

Im in for 2 x K16 and 5-8 nanos DIY kits. If ready maybe need an assembly Steamboat offering.

 

[WynX]

I have less than zero skills when it comes to DIY.

Any chance we can get a lego like kit sans the soldering, i.e. part A1 goes into part B33 type thing?

I have tried soldering, it didn't turn out so well..........

As most parts are SMD (surface mount device), they will require soldering. Guess there's no way around it, however, many folks around the world know how to handle that kind of work. I'd recommend getting someone else to assemble them, either a skilled person for low volume (manual) assembly, or a specialized person/business who will be prepared for handling many of the same boards using some kind of (professional/automatic) assembly-line.

[villex]

I have less than zero skills when it comes to DIY.

Any chance we can get a lego like kit sans the soldering, i.e. part A1 goes into part B33 type thing?

I have tried soldering, it didn't turn out so well..........

As most parts are SMD (surface mount device), they will require soldering. Guess there's no way around it, however, many folks around the world know how to handle that kind of work. I'd recommend getting someone else to assemble them, either a skilled person for low volume (manual) assembly, or a specialized person/business who will be prepared for handling many of the same boards using some kind of (professional/automatic) assembly-line.

yeap, this is nothing like an IKEA kit. You need to know some advance electronics to do it.

I'm looking forward to purchase a kit and test my equipment.. 

[BkkCoins]

Sorry about that. Just sloppy when I updated the parts list. I changed the value but not the product code. I'll fix that now. Thank you for checking. Believe me, I'm not infallible.

You're welcome, and that's exactly why open source = good

Kinda sucks that these cost 4x as much as the X7R. It's possible to save a little with X5R, and I don't expect > 85C. Blah, better to spend $0.53 extra total and not worry about ceramic popcorn...

[WynX]

Kinda sucks that these cost 4x as much as the X7R. It's possible to save a little with X5R, and I don't expect > 85C. Blah, better to spend $0.53 extra total and not worry about ceramic popcorn...

Hmm yeah that does suck...but I agree on avoiding the popcorn . I'm not sure if the X7R/X7S components could also handle higher temperatures during assembly, because there will be a lot of different techniques applied to get them on the board (skillets, IR-ovens, heck maybe even manual soldering). If they do support higher temps, that would be an additional argument for picking them.

I've gone through the nice KiCad files, and noticed that in the stencil (for paste) you've added only a minor blob of solderpaste to the center of each ASIC. Are you sure this would result in adequate heat dissipation to the PCB? I know that in theory the ASIC and PCB should touch, however, there could be minimal spacing which can be resolved with some soldering...
Looking at the bigger picture, the ASICs are designed to dissipate their heat to the PCB and not be actively cooled by a heatsink. Applying a heatsink to the plastic cast of the chip might not take away much heat at all?
I'm sure you've been thinking about this, so I'm probably wrong on the workings of the cooling, but maybe you don't mind putting my mind at ease ?