Bitcoin Forum

Have you guys seen this? https://www.kncminer.com/products/knc-28nm-sha256-processor

I'm surprised there is not a group buy up for it yet.  They're selling the chips for dirt cheap.  The only thing I don't like is they're not supplying the documentation before the purchase, I emailed them to see if I could get the documentation ahead of time.

Seems like the perfect chip for an open-sourced board/solution

They're technically very, very, difficult chips to design for. There's also likely a reason they've been sitting on KNC's shelves for a while.

Why do you say that?

Low yield rate, BGA package with high ball count, difficult DC/DC.

It will be better to inform the buyers on the PCB manufacturer that made their original boards, whit that manufacturer we could be 100% that will get working board, otherwise, even whit the gerber files from KnN its still a huge risk to order boards just to find out that there not mining.

The gerber files from HashFast Evo boards are free from months and yet by this point only peppermining manage the make a working board for the GoldenNonce 700Gh Chip.

And BitCrane (http://goo.gl/skfx67).

Did you just make this stuff up?   You should probably stick to Mechanics.

No. You going to disagree with what I said or just generally disagree.

Low yield rate has nothing to do with the difficulty to "design for".  You design for a fully functional chip, and in the event that a chip bins less than A grade, it's not an issue.

High BGA count really doesn't mean anything.  You either construct the component on your own, or use a pre-existing model.  This process is only time intensive and really does not reflect on "difficult to design for".  Otherwise, the bulk of the BGA pads are for Vdd/Vcc and ground, all of which are just tied to their respective plane(s).  The few IO pins that do exist can be adequately routed through a multi-layer PCB.

Multiphase DC/DC regulation isn't rocket science.  In fact, most MP controller documentation makes recommendations as to what MOSFETs are ideal as well as the desired pad configuration.  Further, KnC didn't even use their own implementation of a multiphase design.  They used complete design solutions from GE and Ericsson.

I agree that there are some challenges in designing a PCB for these chips, but the design challenges don't have much to do with what you've stated, if at all.

is it large cost to resolder/replace broken chips at kncminer boards?

maybe this is a way to use thoose chips...

or create 1:1 clone of knc boards

is there a chance they will give enough information to create a clone?

thanks

My auto refresher destroyed my reply to this 3x, so I've given up typing it out. But yeah, I'll take a QFN64 chip over a 2800 ball BGA chip any day.

Unless they have a different board in the production versions than is shown on their website, the Bitcrane board is exactly the same as the Evo board, with a Bitcrane logo added to the silkscreen layer.

That being said, while KnC hasn't gotten back to me with information on the chips yet, I don't see why they should be a challenge to design for. They should certainly be easier to work with than the HF chips in terms of board layout.

There are differents!  Bitcrane paid Dogie toll of sending miner.

Another one for the ignore list... Its reference with minor tweaks.

.... Why would you buy this chip anyway, the maximum you could get out of it is probably 1W/GH.

Very cheap cost for chip. The issue is the cost required for the manufacturing if the boards. To make purchasing the chips and manufacturing them onto boards viable it would need to be done immediately and at the cheapest possible manufacturing price. You would also need cheap power.

got a reply back from them about getting documentation:

Is that documentation for the chip or documentation for a complete pcb?

I was interested too, but it looks like you can't even have 1 sample and datasheet or anything to test a design before ordering.
Does anyone know of the typical power draw of the chips?
I've read somewhere that they aren't that efficient, so if we need downclocking them to half their hashrate to have 1W/GH, we are effectively paying 0.3$/GH
Not that interesting compared to other one where you get the datasheet and everyhting you need before ordering the chips.

I did the same thing asking of only documentation on pinout and protocol

And received an almost word for word reply.

They replied to me that "more information will become available from the website later today"

Maybe they got tired of getting emails asking for the datasheets.

I also got an email saying docs would be uploaded.

BGA - fun to solder at home.
No open source board design.
No docs.

What happens if you put them in power chain? 13 chips should feed nicely from 12V PSU - I calculated 1877A of current @ 0.87V - can they handle this?
So far I saw only 1 successful design with high current chip - MrTeal's HF board!

I also got a reply saying do a go public shortly.

If we go by their November boards with 8 DCDC modules @ 40A

That's 320A@0.8-0.9V or around 300W. Per board of one chip, that's only 25-50% of the power density of MrTeal's Hoppin Habs. But if it's worth doing, overdo it by 50+% for them overclocking people

So say 480A@0.8-0.9V or about 430W

I meant to say it should be possible to chain them like BFY chips - the only concern is - could the chips handle that much current?
The idea is to keep the BOM as low as possible and not blow things up.

For 180 GH, unless you don't pay electricity, don't expect any ROI.
I hope they can run at something near 1W/GH or they are useless for most of us.
The point is what hashrate will we have @ 1W/GH

True. Only time will tell.

but i can for sure say that the Line Item cost for the VRM can be significantly reduced. A quick look on digikey yielded these numbers:

the VRM used on Jupiter boards is MDT040A0X3-SRPHZ (http://www.digikey.com/product-detail/en/MDT040A0X3-SRPHZ/555-1303-1-ND/3131780)
with

• a cost @QTY1 of 31.55+shipping
• an extrapolated effiency of between 82-85% at maximum load of 40A

A better system can be yielded by replacing the 8 Ericsson modules with:

• 12-16 IR3550 (http://www.digikey.com/product-detail/en/IR3550MTRPBF/IR3550MTRPBFCT-ND/3060832) (2 IR3550 to replace each module, or 3 IR3550 to replace 2 modules)
• 2 UCD9248 (http://www.digikey.com/product-detail/en/UCD9248PFC/296-25804-ND/2230953) to monitor and control each individual phase/rail
• supporting electronics: R,C,L

This would yield a system both lower in parts cost and Greater system effeciency, in the neighborhood of 87-92% efficient, thus lowering energy consumption, lowering the ripple currents & voltages to the asic(will help reduce HW errors, and Less stress on the components) and allowing for some great overclocking capabilities if the dies are able to (as was the case for the HF ASIC which was just starved for power until 'Teal got hold of it and fed it.)


this would reduce the BOM for the VRM from ~ 250USD to about 140USD

Yeah, that's a pass (not that I was going to anyways). Seriously though, why don't they do what the rest of the world does and offer at least partial documentation contingent on a proper NDA being in place?

!!!ALERT!!!

KnC "documentation" has been publicized. and by documentation i mean some grainy 600ish pixel wide photos

https://www.kncminer.com/pictures/product/big/5507_big.jpg
https://www.kncminer.com/pictures/product/big/5509_big.jpg
https://www.kncminer.com/pictures/product/big/5510_big.jpg

Is that it?
No documentation on the pcb design they have?
No in depth chip information?
Just those three pictures?

If so that is woeful.

they've publicly said only documentation on the chip will be made available.

but to be honest i think this is only interim docs while they scramble to prepare the formal ones.

Hope that Mr. Teal find out more about the chips and the original KnC board, build his own version and see some 400Gh mining board at 200 USD :) ( like he did whit the HF Evo Boards)

I actually asked him earlier today on advice  for the power stages since he has more experience with the IR3550

But at 1.5 to 3W/GH, I don't think the market would appreciate even $0.5/GH. Assuming these can get turned around in a month, the rest of the market will be closing up on that price.

I'm not going to read through the whole KnC thread, but my understanding is the FW is closed and no voltage adjustments are possible.
Is any hacked FW available, and has anyone made posted any information on max frequency and power draw at various voltages?

Are you saying that KNC likes to pay more? Edit, ok, it's because KNC likes to over engineer it.

I've never found them publicly. One doesn't really need to be a genius to take the blueprints from HF and make their boards. They are only expensive, but still almost competitive.

I'm also sure that HF would be happy to sell everyone chips at $0.2/GH.

Got my October Jupiter running at 535GH/s@550W (at the wall) for the last 2 months

Is that with four chips?

Yes (can't be otherwise). 133GH/chip, if every chip requires $150-250 in VRMs alone it's out of the market already.

That's a lot of Amps. Is that the reason why it isn't being done? (or is it feasible?).

Which is why the chips have been sitting on the shelf and are being sold (https://bitcointalk.org/index.php?topic=743258.msg8402885#msg8402885). KNC have the scale, all the design done and just need to press the button to have them made into more Jupiters, but they're not.

You could probably find some cheaper design alternatives at those loads however, remember they only used half the VRMs for most of the October units. Then again why would anyone bother trying to get these on boards even if they were given the chips for free, considering the hardware will be obsolete by the time you have something working. Wouldn't mind one for a key chain though!

If it takes downclocking to 133GH/s per chip to get to 1J/GH, it likely isn't worthwhile to pursue.

However if the chip can do 190GH and it was downclocked to 133GH in the 2013 deliveries, while being shipped with both cooling and VRMs with room to spare, does that mean that KNC used them in their DC "overcooking" them? (KNC would have unlocked the firmware). That way they had 190GH out of each chip and not the 133GH they shipped to their customers...

If that was true... Well, it's interesting.

Well, I got this reply from Liam when asking if anything other than the low res ball map would be published.

Strange I got the same answer from both cointerra and hash fast 3 months back.
A lot of people are mad @ Avalon , but in terms of chip info the thing were always ok.
Black Arrow and bitmine.ch doc were good written and available before chip release.
Bitfury was ok.
And spondo documentation and engineering support is outstanding.
All this from board design perspective

So we have to lay down 5.5K just to even be able to design for them.

That blows. If a GB gets started  it only needs to be bought once to be in the wild

Its worth it is you know how to build your own ASIC, but don't risk buying if you don't even know where to start..

The documents aren't for fabricating the chip they are for fabricating the pcb board itself.

the "docs" to make the chip itself cost knc >$1million USD. and will never see the light of day outside of the vault where they're stored

These chips aren't expensive.. How much you could get out of it is probably ?  I mean W/GH

You"ll be doing in the future some boards for the SP RockerBox??

What is the bottom line?  MrTeal (or someone who has done this before), what would it cost to get this going?  What is the final product going to cost?

Bottom line is that it would cost $5,400 to get the process started, and it's hard to say what the final cost would be since until you spend the $5,400 KnC will not release any information on their system.

Hmmm.... It sounds a little tricky... What do you think, if you had the chips in your hand, what would your guess be as to how long until there is a finished product ready to mine?  What would the ballpark cost be for the finished product?  Is it not really possible to know as there is not enough info?

All I know that $.15 / GH/s really caught my attention.  For me, I don't care about electrical costs, the cost per GH/s is basically all I care about.

Probably somewhere in the range of $1/GH/s or so, though it would be lower if you did a large volume. Leadtime really depends on what they release. It could be a couple months if the documentation is sparse and they're hard to get a hold of.

How low can it go?  Say with 200 chips, 1,000 chips, 10,000 chips?  What is possible?

Found at the bottom of page.

* Due to yield rate the number of usable cores varies. Statistically at least 70% of our production at this level has contained chips with 100% usable cores, and up to 90% of all the chips have 187 or more usable cores, with the rest of the chips varying below that number.

That 30% of partly unusable cores should be scary to any buyer. So if you make a $5400 investment, technically you run the risk of having at least $1620 worth of bad chips.

In a practical sense it's only 10% that the end-user needs to be concerned about.  >90% of the chips allege to be >97% functional.  As for the remaining 10%, 'bad' is relative.  They likely still work, just not as well as the others.

++1

Would you be willing/want to head up a group buy for this?
Say a minimum buy in of 0.5BTC per share, to get a 20BTC fund to purchase 200 of these chips and start the research process?

I for one would find this a very attractive proposition, far more than any other pre-order I've ever been involved with.

If anyone wants to get involved does so on the basis that "I invest what I am willing to lose", is this something you would be willing to take on?

i'd be up for it cuz 'Teal all the way!!!!

to post the docs they send with chips would be the only stipulation, since they said they would release them and have gone back on that.

I have opened a thread looking for expressions of interest for a community driven project for ordering KNC chips. The chips will come with a $4 addition per chip above what KNC are selling them for but this is so that chips can be given for FREE to developers such as MrTeal and others who are willing to design boards for these chips. The group buy will use an escrow system i am currently looking for a reputable escrow provider for this. I welcome feedback in the thread. I feel that these chips could be great value if boards can be designed for them.

Thread is here: https://bitcointalk.org/index.php?topic=758408

I'm not looking to acquire the chips themselves, but would be interested in adding funds to acquire the chips for the potential development of a rig(s)/group buy mini pool (could be about 40TH??)
Just a suggestion, but something like a minimum of 10 shares (2BTC each) and a maximum of 40 shares (0.5BTC each) with a cap per investor of 2 BTC.
The developer (MrTeal?) would get 10% of the final hashrate of anything that gets produced (if at all).
The caveat would be, that anyone investing does so on the basis that they are investing what they are willing to lose.
For me, this is still a much more attractive proposition than all pre-orders I've been involved in.

Anyway, just a suggestion, could be a pretty interesting project for the right person :)

Then why won't people let this thread die? Power consumption is 2x the market, price to build is estimated at 50% over TODAY'S price and lead time could be 8-10 weeks out.

And where is the fun in that  ;)

Chip datasheet is exist? Or more readable pins image.  ???

KnC's canned response has been order the hardware, and they'll share the datasheet.