Bitcoin Forum

We did a 15 bitfury miner design using the proprietary
'string' feature this ASIC has. In a string, all chips are
arranged in a series configuration and power is applied
accordingly. So instead of a low voltage/high current
power supply, a less exotic power source can be used,
driving down cost considerably and taking up less board
area.

The board measures 70 x 100 mm, is powered with 12 Volt
and has a USB interface. As no DC/DC converter is needed,
BOM costs can be well below $ 10 (ex. PCB and cooling).

Hashrate can be compared to conventional designs and
overclocking can be achieved by just increasing the supply
voltage within certain limits. Reported hashrate is with 12V8
power supply and with a 2A6 current.

Cooling for the prototype was done using a low-cost
CPU heatsink. Temperatures measured on the PCB were
in the 50..70 oC range.

https://i.imgur.com/Zh1VmzD.jpg

https://i.imgur.com/87kwQav.jpg

https://i.imgur.com/G0lN2zi.gif

intron & c-scape

Note: Some images were retouched to protect our IP.
Hope you understand.

- reserved -

How much?

Well done Intron and c-scape !

Selling a miner based on this design is planned ?

We don't sell miners, we only design miners.

The design can be purchased so other parties can
produce them. We have not set definitive prices
or conditions yet, just probing interrest for now.

intron

looks nice guys! But if the BF chip situation stays the same..

with hashfast announcing 500 GH / chip .. this is ... dust in the wind ... all respect for your work tho !

If the bitfury chips become available again, the higher difficulty requires a cheap board. A design like this has minimal BOM cost, so it will be a good solution.

What's their total cost for the chip, board, power regulators and cooling system ?

The other side of the board has 15 different ground planes. The radiator can short them. Do you use a thermal pad between PCB and a heatsink? Or all ground covered with protective mask?

PS: Well done.

didnt a similar design implemented as an H-board have a problem where if a chip shut off, all other chips became overvolted and shut off or melted?

Prices for bitfury chips might need to go down too. Can't have those ridiculous prices anymore.

Smart guy:) It's Sil-Pad here.

https://i.imgur.com/lNT8oMs.jpg

intron

Yes, you are right. We did that design, but we
didn't operate the board, no idea why it went
down in flames. Our strings board is running
solid now for days and days on end.

intron

As a side note: this string configuration of chips is beyond brilliant,
I think we must honour Valery for this. I have never seen anyone
doing this before. It makes board design so much simpler, elegant
and less costly.

intron

Very nice. Are you using your own custom miner software?

+1

BF is the most integrated chip design of all SHA-256 hashers. It's like Intel Haswell with included clock, Vreg, and logic. What's missing is auto-overclock and p-states ;) just jk.

btw which MCU are you using? Thanks!

all i can say is: thank god this design does not rely on a Raspberry Pi controller. Somehow my bitfury rig went its first 20 shutdown cycles without any issue, but since then it seems everytime i need to adjust it, the card is damaged. Just today i had to format it, run for 10 min, make the new chainminer, then after a few min of 0.000 results i used sudo reboot and the SD is dead again

A tiny NXP ARM, LPC11U24FHI33/301.

intron

The USB interface protocol (http://c-scape.nl/bi-fury/bifury_protocol.pdf) is the same as for bi*fury, so the latest cgminer/bfgminer version can be used.

Note that the design is not dependent on the exact type of microcontroller. As long as it has a built-in USB interface, some I/O pins, and a few kB of RAM, it can be used in the design. We picked the LPC because we had already used it on the bi*fury.

neat and amazing job... are you guys planning to do the same for the latest avalon chips?

I don't think Avalon chips have a String feature.

I really like this design :)

You can have it:-)

It can be done with any chips, but requires extra components (opamp + mosfet)

So when will we see this hitting the streets?

now get this to hash scrypt!....nice design.

Hand me some chips, we'll make you a miner:)

Good job guys. I really like new design ideas!

This is the kind of design you need for a massive yet more cost effective mining farm. If had a bunch of bitfury chip reels guess what I'd be doing....

Congrats to intro and team for innovative board designs.

Intron - Is this design based on the same idea that Tytus had:

(in case the images above don't work here are the two images again: http://imgur.com/D7yH26n and http://imgur.com/gqVOtA5 )

and where you later commented that it was an experiment that went out of control -

So basically - is it the same design? If so - what has been done (and was there anything done) to avoid the same outcome as tytus' ?

And just out of curiosity - are you guys going to be publishing the design schematics?

this was my concern too, but intron's design has its own controller, so perhaps (hopefully) it is capable of detecting a lost chip and acting (restart the mining process or turn off) so that we dont see a bunch of home fires and molten PCBs

The microcontroller is pretty much irrelevant for that issue.

The issue that Tytus had is the same one that you get with any Christmas lights string - if one light shorts the rest get some extra voltage. If two short, the remaining lights get even more overpowered. And it gets worse and worse even faster until they all melt and/or the fuse blows.
Or if one light burns - they all die.

In tytus' case he was getting the first option - when chips overheat they tend to use more power and then they start overheating even further even faster until they hit the limit and the melting point. At which point you get either a short in one of the chips (and then everything dies very quickly), or the chip cuts the string and again everything dies. Actually my preference would be for the second option - at least the rest of the chips may survive :)
Unfortunately (or fortunately?) bitfury's chip has a very solid design, and it's much easier to melt the silicon inside than to burn all of those power wires.

Bitfury said at some point that he gave up on that design (quoting various reasons) but as time goes people may have found a workaround - and if that's the case - lots of respect for intron and co!

And I guess that's the root of my question - is it some novel approach?

We have three One String boards hashing now,
looks like performance is rock solid.

Board temperature well below 45 oC. Hottest
ASIC about 70 oC.

https://i.imgur.com/n3xjh1l.gif
https://i.imgur.com/5aL4hGD.gif

intron

PS: The little blip in the green trace was
cgminer going down and needed restarting:)

Ok, so the selling point on these isn't so much the hashrate as the lower cost of the power supply? You are losing me there. can you give an example of a typical power supply used and a lower cost alternative? I'm not sure what we are talking about.

This design uses 12V to power the chips directly. Other designs, such as the H-CARD have an on-board 12V -> 0.8V power regulator with 30-50Amp capability.

In both cases, you need a 12V supply, but for our new board, you avoid the additional cost of the 0.8V regulator. Also, the 0.8V regulator is typically only about 90% efficient, so it creates 10% more heat for the same hashrate. The powerful switching regulator also produces lots of electrical interference that can degrade reliability of the overall system, and requires additional components (capacitors/inductors) to help reduce the noise.

Oh, so the cost savings is in the parts list, not the actual power supply. The only reason i asked is that i'm getting tired of miners requiring ATX power supplies. Unless you get a modular one you have to deal with a bunch of unneeded cables. also they take up more room than a power brick. Unless you use some sort of case to put your miner in, your work space can become quite messy/unwieldy. This is one of the reasons i won't buy ASICminer Cubes, 1-2 Cubes per powersupply can make for a lot of power supplies laying around. I'm a Hobbyist.

But this design has a very big disadvantage. If the first chip in the series connection got damaged the other chips doesnt work anymore, too.

That depends on the exact nature of the damage. If the hashing core is damaged (which makes up the biggest part of the chip), the chip may still provide a working connection to the other chips.

It's not much different than low-voltage bitfury designs, such as the H-CARD. In those designs the communication between the chips also forms a chain. Actually, the M-BOARD has 4 H-CARDS with total of 64 chips in each chain.

Correct. You still need a 12V power supply. Each card draws about 2-2.5 Amps, depending on the exact voltage.

Just as an example, take a look at the HashFast board: almost
50% is taken up by the DC/DC converting circuitery providing
for the low voltage/high current the chip needs.

In a string design this DC/DC conversion is no longer needed,
so taking up no PCB area and not adding to the Bill of Materials.

intron

And it's not just the cost of the components, but also the availability of some of the more exotic ones.

Another factor is the circuit board design. The fast switching DC/DC regulators typically require at least a 4-layer board for good performance. The One String design uses a cheap 2-layer board.

Might be a more interesting design if the chips were priced anywhere near the marginal cost of manufacturer, but with them priced like they were solid gold— the cost of the additional electronics and the reliability concerns may outweigh the benefits.

These boards don't have additional electronics. On the contrary, a lot of the expensive components have been removed, and the 2-layer board is cheaper too.

Yes, I know. I was attempting and failing to say that the because the chips are so costly the additional electronics are not that substantial, and considering the reliability concerns the efficiencies gained may not be worth it.  E.g. you save $x on support electronics but increase the failure rate by y% then if its a win or not depends on the price of the chips.

E.g. if chips cost a million dollars a piece and you save $20/chip on electronics at the cost of a 0.01% increase in failures its not a win.

So far, we have no indication of increased failure rate. The prototypes that we have are rock solid with constant temperature and constant hashing speed. In a couple of weeks, we'll make a few dozen more boards, and we'll see what happens to the failure rate.

That will change. The chips are going down to $5/chip.

Either they will go to $5/chip and lower, or they will not be sold at all. In the first case, you need a cheap, low-overhead, mining board. In the second case, you don't need a mining board at all.

Fair enough.

So when, where and how much?

I would like to know that for the 5$/ bitfury chip, because if you are asking about boards, then find out that they don't sell boards. Only the design.

i'm still not sure if my concern was fully answered; what makes this board design work when the H-board string design ended in molten plastic and scorched PCB?

We have a better design :)

The problem with this design is that it requires the 12V power supply to be very well regulated (and adjustable if you want to overclock). Normal ATX PSUs are not that well regulated (ATX spec provides for up to +-5% tolerance on the 12V line (older spec allowed 10%).

On the other hand, I could buy a big well regulated (and adjustable, or at least modify it to be adjustable) 12V PSU and connect a lot of the boards to it. Then it comes to what is cheaper - A standard ATX PSU (I can sometimes get them cheap by taking failed ones and repairing them) and regulators in each board or an expensive PSU and no regulators. I guess the "no regulators" design wins over for large numbers of boards, while the traditional design is better for a few boards.

The on-board 0.8V, 30A regulator would add about $15-$20 to the design. For 10-15 cards, the money saved would buy a good PSU with voltage trim option.

Either way, the cards work fine if the voltage isn't perfectly regulated. They just run a bit cooler under 12V, and a bit faster/hotter over 12V.

So, now it's time for you to design board with parallel/serial connection of chips ;)
PowerMatrix ;)

We did that already. It's called X-HASH:)

intron

and how did that work?

It takes so many bitfury chips to test it, we were
reluctant to try it. That's why we started with the
smaller strings.

intron

In the next day or two I should have prototype information available for an interface board designed to break out server supplies for use with miners; these can typically be found for much cheaper than an equivalent output/efficiency ATX supply and usually exhibit better regulation. Screw terminals so you only use what wires you want, and additional output filtering. They're also designed to be run in parallel with load sharing and daisy-chain for automatic power-on in groups.



If enough people want it, I'll look into adding a trimmer to tweak output voltages.


Been watching this thread; series power dropping chips (if it can be done reliably) is such a darn cool idea and if someone ends up making these boards affordably (and chips are available for not a bajillion dollars) I'll definitely be in the market. Good work, guys.

are there any photographs available of said x-hash or was one ever fabbed?

No, the board was only designed, never made (yet).
As I mentioned, we didn't want to risk a lot of bitfury
chips for just one experiment.

intron

I have repaired a couple of power supplies used in Supermicro servers (Ablecom forgot the part number). While I still do not know the full pinout of the connector, I found which two traces to short inside to turn it on. Aslong as the power supply is not fully digital, it should not be difficult to use it. Still, the regular ATX PSUs are simpler and easier to fix :)
Traditional (not digital) power supplies can be modified to adjust he output voltage just by controlling the feedback. However, if you want the adjustment to be external to the power supply (so you don't need to open the PSU), then youwill have to use a powerful regulator and the total cost will probably be higher than an adjustable PSU.

Had five Ones String boards hashing for some
time, performance looks nice and stable and
no thermal runaways.

https://i.imgur.com/2TykbVp.gif

intron

I don't think it will work better than several strings in paralel. The chip seems to like constant current more than constant voltage, but the built in regulator will try to equalize the voltage across paraleled chips and if they are not equal, they will not perform at their best ... and you save nothing more than few resistors on the SPI. I have my strings running on 24V better than on 12V, probably because the current is more stable

I hope the chips will be cheap and freely available in volume soon - my idea is to run the string directly on 220V, which also risks a lot of chips for the experiment, but will save additional 10-20% of power efficiency ... just not sure if RPi will be capable of running 350-400 chips in a string and also there is the lower SPI speed with the increased string length

Did you try placing chips in parallel? I'm also not
sure whether to make seperate strings or connect
all nodes with the same voltage level together.
Biggest challenge is the huge amount of chips you
have to sacrifice for a single experiment that might
run haywire:)

And this 220V string is af course the optimum you
can get in cost efficency:) Looked into a rectifier
for this, it's not easy, mainly due to the current that
is needed.

Didn't even tried that. During my tests i have discovered that each chip requires a bit different voltage for optimal speed, but always the same amps, so it's like the chip 'wants' to be serially connected :)
You need 2-2.5A for the string ... let's make them 3A = 600uF for 0.1V @50Hz it's not that much and expensive, but still, the main ripple regulator should be the current limiter for the string and then each chip for itself (via it's current mirror)

One of the problems is that the grid voltage is not so stable, you need protection from the (kA) spikes, so you will definitely need an UPS (and if it's on-line model with 220V to 110V conversion, you can go with shorter strings), but then you'll loose in efficiency again :(
The second one is, the string length will reduce the SPI speed and the third one is that RPi (or dedicated MCU) may not be able to drive that much chips

EDIT: Ooops it's 300mF not 600uF, so you are right it will be expensive

Well done folks !

Hi

Are you able to make me this miner if I get the chips

Cheers

https://bitcointalk.org/index.php?topic=370932.msg3963420#msg3963420

Fastest USB miner at this moment:

https://i.imgur.com/nFqPO7d.jpg

https://i.imgur.com/MjCPOg4.gif

Doing 15+ GH sustained, hex•fury is running
from a 5.5 Volts supply.

intron

How much?

PM-ed you.

So it's not powered from the USB, but from an external power supply ?

Even at the minimum 0.6W/Gh the BF chips are capable of - 15Gh is 9W, which is not possible from a single USB port be it even 3.0
With 2x USB 3.0 or 4x USB 2.0 it is doable

nice USb, but i dont understand why - it draws far too much power for a USB port, so for the effort of getting it to work it seems like a much easier idea to go with the one-string or other design that uses 12V and holds more chips

please pm the price for hex fury , thanks

It takes about 2.5 Amps, you can power it from
an USB hub. And it also has a seperate power
input that feeds the bitfury ASICs. When this
option is chosen then the USB has to power
only the processor.

And indeed, we were just playing around a bit...;)

Hi intron, please PM me price for hex-fury. 

Extremely promising design. Congrats :) what is the controller chip that you are using? Thanks!

awesome. someone actually made a 6 chip version. i'd thought about making a 6 chip stick when the sample chips first came out (but our first design failed and we gave up)...I think this is the best bang for the buck you can get out of bitfury chips in a USB stick form factor.

Yes, that was the idea. Total BOM cost is about $4, excluding bitfury ASICs.

How much (PM please  ;))

Hi guys,

it's really nice work you've done there. Interested in getting one of them! :)

Greetings,
Silv0r

please pm with prices :D please and thank you

Are the chips available now ? If so a roundabout price ?

No they won't be back in stock until February

I really like the One String miner. Wish i had some bitfury chips right bout now. Is this what Dave was carrying with him at Inside Bitcoins Las Vegas?

Don't think so:)

Intron, Cscape,
         Very nice work! Just when i thought I've seen it all, you come a long and surprise me.
in your opinion, what are the possibilities this will become a standard with new ASIC chip designs from others?
Hopefully other ASIC competitors will take notice.
No DCDC converter with only two layer PCBs is a huge win in price reduction.

Many thanks to bitfury - https://bitcointalk.org/index.php?topic=228677.msg2411923#msg2411923 (https://bitcointalk.org/index.php?topic=228677.msg2411923#msg2411923) he designed the chip with that in mind ... so far no one picked up that trend and we are still far from the best design of a chip, but if he fixes the few small problems for V2 (which i am sure he is working on and wish him a luck with that) - that will be the best of both worlds: high hashrate and cheap board.
Many thanks to Intron and Cscape for their work too - they have showed to the public that it's possible and hopefully more chip designers will at least think about the possibility.

To be able to make a string design, the chip must provide
for a method to shunt the current that is carried by the
string when it is not in full operation. When you got that
one right and have some means of level shifting of the
signals going up and down the string it should be possible
whith any chip.

And Valery got that right:)

Also interested in the hex*fury...  Please PM for the price.  Thanks.

how much for any of ur products ?

I have a few quick question regarding string configuration.

Q1 ) Can I assume each bitfury chip acts like a diode, and a string is just a series connection of multiple chips?

Q2 ) Do the chips have an internal voltage regulator or some kind of impedance adjustment?

Thanks in advance.

Not quite, as it has it's own consumption, not a fixed voltage drop like diodes, but the second part is correct.

Yes there is current mirror inside the chip, which does the job.
Detailed explanation is here (https://bitcointalk.org/index.php?topic=183368.msg2460608#msg2460608) and here (https://bitcointalk.org/index.php?topic=242745.msg2576406#msg2576406) is a picture of what's inside the chip (to replace Russian explanation), but the external links of the schematic are wrong

Where do we get these to test out??  That looks like a little beast with a big punch.

I'D like to buy one can I get directions where to buy from? :)

We are working with others to get both the 15-chip One String Miner, as well as the 6-chip hex*fury produced.

I await updates on this :)

someone needs to compete with Bitmain. its shocking that that company came out of no-where with a developed product, and has sold from stock for months now at the best price on the market and some of the best design (could be a little more compact or a bit more power efficient, but those are little details)

I am definitely interested in them when they come out.  Looks great so far.
If you need someone to test any units for you let me know.

I am definitely interested in them when they come out.  Looks great so far.
If you need someone to test any units for you let me know.
--
Thank you for the offer, but we are 'testing' the hell out
of these things already. And for more than Two Weeks (tm) ;)

intron

Very cool.
I can't wait to see them hit the market.

Only if there are chips first :( Waiting since November to build some more strings.

One strange thing i have noticed, is that the hashrate is increasing with the time without any other changes - started with 245Gh and now at 265Gh/s ... ~7% increase it's like with the time the clock stabilizes and increases or the chips get in some kind of resonance between themselves.
@intron do you have similar experience with strings?

The hashrate of a strings design is heavily dependent
of the supply voltage. You are sure this didn't change?

The PSU is 24V +- 1% not 7% and the power (edit: at the wall) is the same, while an increased voltage should have been increased the consumption too.

My strings are of 28 chips or 14 per 12V instead of 15 as in your case. The last chip is replaced from a diode to get the IOVDD (wasn't sure if IOVDD = VDD will work, so played safe) - this is the only part that may affect the overall per chip voltage ... in case the voltage drop on the diode decreases with the time, but then the effect of the temperature would be much more, which is not the case, so my (current) bet is on either CMQ (+-) resitors change in value (= less unused current for balancing) or some kind of resonance of the chip clocks

We've got them  :)
https://bitcointalk.org/index.php?topic=495536.0

Intron - please PM me details about the hex miner, VERY interested xD

PM-ed you.

Same here please, want to buy at least one, awesome design!

Me too please.  I had already PM'd you though.  ;D
Thanks

Any ETA on these? Pumped for 15 gh/s usb sticks xD