Klondike - 16 chip ASIC Open Source Board - Preliminary

[bitcoiner49er]

Not sure what the point of that would be, the ASICMINER chips are different from the Avalon chips...

Ah yes, stoopid me. Was thinking Erupter was Avalon.

[neslekkim]

The connector area is cut away . On the projected view you see the radius on the inner edge, that is needed to avoid additional side operation on CNC mill

Aha, you will use cnc mill on these?, then it should be possible to mill out areas where one would use throughole components, and thus don't worry about finding proper smt connectors?
Or even, mill out so the 16 pads for the chips are raised? (ok, i'm proably blabbering here..)

[sensei]

That looks good. I checked the hole and cutaway positions and they are correct. The width and length seem to be a bit less visually than I'd expect. I'm not sure the fins should go sideways. I guess that would depend on how someone plans to install them. That seems like a lot of machining. Is it easier to cutaway the connector area rather than drill some countersunk holes?

I myself was also thinking that the fins should go in the other direction.

[marto74]

That looks good. I checked the hole and cutaway positions and they are correct. The width and length seem to be a bit less visually than I'd expect. I'm not sure the fins should go sideways. I guess that would depend on how someone plans to install them. That seems like a lot of machining. Is it easier to cutaway the connector area rather than drill some countersunk holes?

I myself was also thinking that the fins should go in the other direction.

This is no problem .
We can do it both ways

[marto74]

Here is the updated hest sink
http://imageshack.us/photo/my-images/694/fan3i.jpg/

[ecliptic]

I know the board is still WIP and shouldn't be ordered for anything except very initial testing [probably w/o the chips themselves even], but has anyone contacted PCB assembly places to get a rough quote on what the boards will cost?  I plan to do that this week, at least to give them a heads up.

[marto74]

look at my thread in the signature

[ecliptic]

Crossposting this from raginazn's order thread.

..

I am an EE with access to a lot of nice test equipment and plan on having my Klondike-16 boards assembled by a local professional PCB Assembly (P&P + Reflow with Pb63Sn37 solder) company I've worked with and can get boards made very fast.  I'd like to get one or two of the sample chips to test and publish for everyone's knowledge the results of these tests before the full batches ship in July:

* The Klondike-16 design as-is (P&P files, Solder Paste locations -- especially the QFN pad, Solder paste stencil thickness, etc) as populated by an outsourced assembly house
* The standard functionality of the boards (with a single chip, my understanding is the design will work with any number of chips, might need to tweak the firmware).  Hash rate, power draw.  Power draw of ASIC at various clock and hash rates.
* Thermal analysis of the board.  With various heatsink, without a heatsink, with various heatsinks with poor mechanical connection (I'll watch the temp closely so it doesn't overheat and damage the chip), with and without forced airflow.
* Overclocking.  I can modify the design, change out clock sources, modify the settings on the PLL on the chip(s).  Am very curious to see what these chips are capable of with proper cooling.  I would not be surprised to see 20, maybe 30%+ better performance.  I do not think the current 282MH/sec is a thermal limit, but I can't know for sure until i can play with it.  Similar to thermal analysis, I'll keep the chip within acceptable operating range.  I don't want to blow anything up until we have a lot more information.  Especially with only the one chip.
* Affects of the above on the power requirements (increased current on the 1.2V line, perhaps above 2A/chip.  Possibly need for additional decoupling capacitance on the chips.  Oscope measurements, Spectrum analyzer (depending on if there is harmonics or noise coupling onto the Vcc plane(s)).
* After the above, some modifications to the design to support overclocking.  -- While at this time I believe there is a good amount of headroom on these chips, the current design, both Avalon's official reference and Klondikes, provides exactly the max current spec for the chips (i.e. 32A @ 1.2V for 16 chips, each rated for 2A max.)  I expect that OC'ing will push this beyond, something i'll confirm with power/current measurements when OC'ing.  (Will report Clock freq, Hash rate, Current draw, Temperature, Heatsinking methods used, etc.)


If i happen to get more than 1 (i have some chips in other orders), I will probably have multiple PCB assemblies built up, each with 1 chip.  This will help to verify a tiny bit better the yields of the assembly, which in theory should be very high.  But there would be nothing worse than suddenly realizing that 50% of the boards are scrap.  If i get say, 4, i will have 2x boards with 1 chip and 1x board with 2 chips made.

Of course we want to check to make sure that a board with say, 16x chips works of course. But i expect that BkkCoins will be getting several dozen chips from various people, so he should be able to take care of that.

Assuming i get chip(s) from my orders or elsewhere I will test and report my results on these, but otherwise food for thought for anyone who will be getting sample chips.

[BkkCoins]

Note that while the docs state a max of 2A per chip, the reported value for the Avalon is stated to be 6.6W/GH/s or about 1.55A per chip. I went with the 16A reg. anticipating that over clocking would probably push this 1.55 up closer to 2A. I would not expect 2A to be normal for stock clock.

[ecliptic]

Hmm so i am looking at the gerbers in altium at the moment, since KiCad can't find a project file to open anything..

It looks like the inner copper layer1 has the 5x5 thermal vias under each chip (and nothing shorting out), but layer2 appears to have solid copper, which would short out all the vias

Maybe it makes sense in the regular PCB design in Kicad but from the gerbers i'm confused

Also, do you have any solder mask between the pads of the QFN?  It looks like a solid block where it's removed.  Probably not a huge deal, but it can help when trying to reflow that especially for DIY type work

[ecliptic]

Ok, i got the avalon reference files opened (finally).. [i know i'm late on this]

They connected the thermal vias (a lot too, maybe 7x7 per chip?, but marginal advantage over 5x5, diminishing returns and all that) to a solid ground plane on layer 2, and a small ground island on layer 3

Layer 1 : 1.2V Plane [top]
Layer 2 : Ground plane
Layer 3 : Signal routing.  The rectgular pad for each chip is a solid square of copper.  Because it's not very big, it does not contribute massively to heat dissipation, but perhaps helps a bit.
Layer 4 : Ground plane [bottom]

This is also a good design as it gives you pretty good isolation for noisy or RF/controlled impedance (reasonably cheap and low requirement controlled impedence) on layer 3 sandwitched between grounds.

The note on the silkscreen calls out for resin filling on the vias.  Filling them with a thermally/electrically conductive material like copper would make sense.  Resin is a bit suprising, but it's probably thermally conductive resin, probably cheaper to fill the vias with that instead of a metal.  This also gets more reliable/better solder paste and reflow for the pads -- prevents solder paste from sometimes going into the vias, resulting in more variation and failure between batches.  definitely something worth quoting to see how much more it costs.

looks pretty good when compared to the gerbers -- looking forward to the release of the klondike design files for the pcb and schematic

[Bicknellski]

It terms of the mining software / firmware BKKcoins basically you are going to provide that for the Klondike correct as open source as well to the community and I hope we can help develop and revise, improve this as a community of Klondike users? Basically we shouldn't need to give "free" Klondikes to those who code for CGminer or other software etc do we? I find that counter productive to the whole spirit of the open source aspect of this project... very parasitic actually right? I think we need to NIP that right in the bud so no one is held hostage by miner developers.

I would prefer to add a small fee, which is sent to you as a donation for every download for the Klondike, anytime I provide the software / firmware to someone who buys a board from me. Of course giving you a royalty for each board sold, is a must and I will be putting that directly into my costs for those buying boards. We should all be supporting your work this way I hope. Of course some will leech off the project but really we should be giving what we can to you for all this work you are doing right?

[BkkCoins]

It terms of the mining software / firmware BKKcoins basically you are going to provide that for the Klondike correct as open source as well to the community and I hope we can help develop and revise, improve this as a community of Klondike users? Basically we shouldn't need to give "free" Klondikes to those who code for CGminer or other software etc do we? I find that counter productive to the whole spirit of the open source aspect of this project... very parasitic actually right? I think we need to NIP that right in the bud so no one is held hostage by miner developers.

I would prefer to add a small fee, which is sent to you as a donation for every download for the Klondike, anytime I provide the software / firmware to someone who buys a board from me. Of course giving you a royalty for each board sold to me is a must and I will be put that directly into my fees for those buying boards. We should all be supporting your work this way I hope. Of course some will leech off the project but really we should be giving what we can to you for all this work you are doing right?

I am going to release the firmware and driver as open source. But I don't intend to release the firmware "early". I guess I need to comment further as some few are going to jump on this.

My original intent with this project was to provide a design so that DIY hackers could make their own and do mining. But it's changed and many people have jumped in and want to make a profit off the project and I've increasingly asking myself why I'm working to put this out there when it's so obviously detrimental to my own mining.

IMO those who have contributed and supported my efforts all along, and are willing to continue with a small per unit fee, to support me and be first to market, will have access to either programmed PICs, or serialized locked binaries. Those who have been hovering, and looking for a free ride to a pot of gold, will need access to a competent programmer who works for free. I won't be providing technical support and free coding services for that. If they believe that offering a $2/unit cheaper product is worth the effort of writing their own firmware, or producing their own design or boards, then I'm not going to fight it. I don't have a problem with those who want to profit from my work. Just those who thought they'd profit by cutting me out and get to market first. Why would I want to help them do that?

Avalon makes a few $$ off each chip, and CKOlivas ought to make a few $$ off each miner (which I don't have the power to enforce), and I guess I should get a couple $$ for each Klondike. Eventually all my work will be free and open and anyone can use it to turn a profit mining, or selling mining rigs. And I'll move on to whatever's next.

edit: By eventually, I mean before too long. Like maybe a short while after the chips start to get to end users. That way DIYers can use it and make or buy boards, and mine.

[zefir]

It terms of the mining software / firmware BKKcoins basically you are going to provide that for the Klondike correct as open source as well to the community and I hope we can help develop and revise, improve this as a community of Klondike users? Basically we shouldn't need to give "free" Klondikes to those who code for CGminer or other software etc do we? I find that counter productive to the whole spirit of the open source aspect of this project... very parasitic actually right? I think we need to NIP that right in the bud so no one is held hostage by miner developers.

I would prefer to add a small fee, which is sent to you as a donation for every download for the Klondike, anytime I provide the software / firmware to someone who buys a board from me. Of course giving you a royalty for each board sold to me is a must and I will be put that directly into my fees for those buying boards. We should all be supporting your work this way I hope. Of course some will leech off the project but really we should be giving what we can to you for all this work you are doing right?

Adapting cgminer for Klondike should be way less challenging than the HW part and I assume there are lots of folks out there who could do it once the protocol interface is specified. Otherwise I can do that (did driver optimization for CM1 boards before). Since I plan to get a prototype assembled with my portion of sample chips, I will be able to test early.

As for the compensation for the work being done in this project: since Bkk is not going to produce and sell the boards he is designing (in contrast to e.g. burnin), the only way to get compensated for the hard work is through donations. That's Open Source, but since you can't pay your rent with the reputation you gain, I firmly hope end believe that everyone who will take the design to build Klondike clones won't forget to send a donation.


Edit: overlapped with Bkk's post, sorry.

[steamboat]

BkkCoins is receiving a percentage of every miner produced through me. I sincerely hope everyone who is planning on using the Klondike design understands the amount of talent, expertise, and labor involved in producing a mining solution from the ground up, and gives credit (and coin) where it is due.

We as a community would not be in the position we are if it were not for people like BkkCoins who are willing to contribute to the network.

Remember ladies and gentlemen: Those who serve best profit most.

[BkkCoins]

I was watching this scope on eBay but it finished while I slept last night. Ouch. Someone got a great deal on it - a relatively nice new $800 scope sold at $305.

http://www.ebay.com/itm/200924817394?ssPageName=STRK:MEWAX:IT&_trksid=p3984.m1438.l2649

I would have bid higher but it's dangerously close to being beyond what I could rationalize. Of course, I could have sold it off later and probably even profited at this price.

[Bicknellski]

BkkCoins is receiving a percentage of every miner produced through me. I sincerely hope everyone who is planning on using the Klondike design understands the amount of talent, expertise, and labor involved in producing a mining solution from the ground up, and gives credit (and coin) where it is due.

We as a community would not be in the position we are if it were not for people like BkkCoins who are willing to contribute to the network.

Remember ladies and gentlemen: Those who serve best profit most.

+1 Steamboat
+1 Zefir
+1 BKKCoins

Thanks guys appreciate the candor. I will emulate your kick back per board steamboat and donate for the software whenever possible for Zefir and BKKCoins. I don't really think it is necessary for anyone to give completed Klondikes to anyone other than people who are doing the work first and have the full intent to promote open source development. Thanks again gents so nice to see this project has backing from very reputable and forward thinking people. This is certainly an extension of what the Avalon team is doing and that to me is great to see.

[gill83]

PIC16LF1459-I/SS are out of stock everywhere. Soonest lead time is 5 weeks.

http://www.digikey.com/product-detail/en/PIC16LF1459-I%2FSS/PIC16LF1459-I%2FSS-ND/3671495?enterprise=32

How much would you sell programmed PICs for?

[BkkCoins]

PIC16LF1459-I/SS are out of stock everywhere. Soonest lead time is 5 weeks.

http://www.digikey.com/product-detail/en/PIC16LF1459-I%2FSS/PIC16LF1459-I%2FSS-ND/3671495?enterprise=32

How much would you sell programmed PICs for?

I've been watching their supply dwindle over the last few weeks. I don't know if its' just bad timing or they actually don't make many of them due to it being pretty new, and not used in many existing designs.

I've seriously thought about making a QFN variant of the board to take advantage of that package. It wouldn't be hard and could be done after prototyping. There are lots of them still available. Note also you can use the E/SS variant but even they seem to be vanishing now. Having an QFN board variant would open up about 1800 of LF1459-I/ML currently sitting at Microchip and 6200 of F1459-x/ML that work also.

I know some users contacted me to make sure I had some for testing before placing their orders.

If someone wants to start scooping up PIC16LF1459-I/ML, then let me know. I can produce a board that uses them. There's 1881 of those on hand at Microchip.

edit:
Programmed PICs - don't know yet but the idea is $2 / board. I'd order them pre-prog. from Microchip except judging by stock it's pretty risky waiting until the firmware is fully ready. Or maybe not as it only requires the bootloader and serial# to be ready.

I would prefer to work out a downloadable binary method instead.
Maybe it could just be a fixed fee to get a pre-release "early" firmware image - based on expected production run.

[gill83]

I've been watching their supply dwindle over the last few weeks. I don't know if its' just bad timing or they actually don't make many of them due to it being pretty new, and not used in many existing designs.

I've seriously thought about making a QFN variant of the board to take advantage of that package. It wouldn't be hard and could be done after prototyping. There are lots of them still available. Note also you can use the E/SS variant but even they seem to be vanishing now. Having an QFN board variant would open up about 1800 of LF1459-I/ML currently sitting at Microchip and 6200 of F1459-x/ML that work also.

I know some users contacted me to make sure I had some for testing before placing their orders.

If someone wants to start scooping up PIC16LF1459-I/ML, then let me know. I can produce a board that uses them. There's 1881 of those on hand at Microchip.

I think I/SS would be fine, as we don't expect Avalon to start shipping chips before 5 weeks anyways. Both Digikey and MicrochipDirect have a lead time of 5 weeks. So, personally I don't think you need to invest your time for using the QFN version. I was able to order 41 of I/SS, so let me know if you need a few for testing (for free of course).