Topic: HashFast Miner Protection Program Discussion (Read 10997 times)

HashFast's Golden Nonce GN ASIC successfully taped-out yesterday, Wednesday the 28th, and has been released for 28nm fabrication to a well-known, leading-edge foundry.  More details will follow in next week's joint press release.

-John

Dear John --

Yesterday, when you told me about VERY BIG NEWS TODAY, i promised myself not to pee until i heard from you, so's not to jinx it.
My pants are wet and cold, i feel foolish.  WHY WOULD YOU DO THIS TO ME?!

VERY BIG NEWS COMING TODAY!

Also, We have released more inventory for first batch due to a number of wire transfers that failed to come through, to complete the sale of the first batch of Baby Jets.

I have posted details about new order chain rules, and the release of inventory here:
https://bitcointalksearch.org/topic/m.3029016

-John

You have to admit though, it is a pretty fair plan.
I was hesitant for so long because I thought the MPP would only be in USD, but it being priced in btc makes all the difference.

Judging by the chip render on the HasFast site, the chip will be an Intel CPU with matte marble texture added & heat spreader pried off.  Hope this helps.

Cypherdoc, since you're being paid to say this, why not ask John to return the compliment?  Something along the lines of "Why, that's very nice of you to say so, cypherdoc, your kindness is on par with your honesty!"  That way, you can keep the exchange going & people will not notice that the only thing new in HashFast's announcement is limiting miner protection to just the first batch.
If you found my reply helpful, please donate to the BTC address conspicuously absent from my sig. 

Thanks!

Details are going to have to wait until after tapeout. (Engineering is busy right now) 
However, we want to encourage others to design rigs for our chips and we are excited about what this community will do with them!

-John

As a Group Buy Coordinator of 1 co-op owned HashFast BabyJet, I am pleased with the terms and conditions of your MPP. I like the transparency and level of detail.

It should serve as a model for future ASIC rig pre-orders IMO. Anything you can add about the Motherboard specs/availability and institutional pricing for said Mommy boards? [IT Birds N Bees - "See son: "Baby computers are made when a Daddy board and a Mommy board fall in love and..."]

Could be an entrepreneurial opportunity there for someone.  

That is incredibly generous.

Glad you're limiting it to just batch 1.

Hi Everyone,

We have unveiled details of our Miner Protection Program here:  hashfast.com/miner-protection-program/

-John

That's interesting. I'm not an electrical engineer, but I've actually been trying to figure out things like gate delay for various technologies and node sizes but haven't been able to find much except for formulas involving transistor load capacitance - which would be fine if I could find that information easily.  Is there anywhere you can look this stuff up?  I did find the beta and Vtn figures for CMP's BiCMOS process here, but I still don't know how to calculate the load capacitance. 

I can can come up with digital logic circuits (I designed a working CPU core in a class once. That was fun) , but I don't know much about the analog stuff underneath - we did everything with FPGAs.

I've actually been working on a low gate-depth iterative hasher circuit design - I'm curious to know how fast I could clock it for a given circuit depth.

So what? Bitcoin miner is an LSI circuit. Thus far people used VLSI design methodologies simply because the existing IC layout toolchains couldn't cope with thousands of copies of identical circuit and either failed to converge or converged very slowly. So they simply unrolled the rounds of SHA-2.

On the other hand alternative semiconductors offer radically lower leakage, radically lower parasitics and radically faster clocks, e.g. 250 GHz. Also it so happens that the characteristic impedance of a metal trace (as a transmission line) better matches both input and output impedances of the transistors. So instead of fighting with reflections and noise in the interconnect the circuit is using the supplied power to do an usefull work: toggle the gate/flip-flop.

If you know how do a simple experiment: simulate a paralled 32-bit adder with all the attendant carry-look-ahead logic. Then simulate a simple serial bit-wise adder at a clock speed 32 times faster. Then compare the total energy used by those two. The technology node doesn't matter so long as it is the same in both cases.

That's true, but you can't necessarily get 28nm chips with those materials.  I remember hearing that GaAs's can be made at around 180nm.

By the way, does anyone know what the wafer costs are for standard silicon chips at various resolutions? I would think the cost to the fab would be about the same, but there's probably a lot more demand for 28nm wafers.

If 28nm wafers are more then 4x as expensive as 55nm wafers, then 55nm chips would actually be cheaper to produce, per feature. (ignoring higher speeds)

I wouldn't be that pesimistic. There is a lot of possible avenues for improvement in a dedicated hashing circuit on the back-end analog side of the design.

When Intel design their "digital logic gate" they make certain assumptions about the probability of the circuit making a transient fault. I don't want to spend time digging numbers, they published some of that in a HotChips presentation about Itanium RAS (reliability, availability, serviceability). A CPU of the complexity of the Itanium is rated somewhere around 10^-20 for the probability of undetected machine check fault.

On the other hand the experience of the FPGA & ASIC miners is that the optimum hardware error rate for miner is in the order of 10^-2, that is single percentage points. And the complexity of even a fully-unrolled double SHA-256 hasher is nearly nothing compared to a modern 32/64-bit CPU.

So there is an ample room to use the curcuit design methodologies that are less digital and more analog. I don't think that any of the current vendors of the Bitcoin mining ASICs has any experience designing this type of circuits. Bitfury did a little bit of exploration in this direction by checking how low he can go with the supply voltage, but he had used pretty much standard push-pull static CMOS logic gates. There are many other ways of implementing logic gates than static push-pull (a.k.a. bang-bang). Edit: There are many other semiconductor materials suitable for the manufacture of integrated circuits: GaAs, SiGe, etc. There are many other IC design companies that are not very well known, but specialize in other areas of circuit design than the well-known CPU/GPU/DRAM/Flash in silicon e.g. http://en.wikipedia.org/wiki/Vitesse_Semiconductor .

Wait.  I'm not arguing about the chip being delivered.  I'm pointing out that the chip is being designed by a team of unknowns, and use of 28nm fab means only that they settled on 28nm -- they could have chosen to work with 65, 110, transistors or vacuum tubes.  Their chip doesn't begin to describe the limitations of 28nm fab process -- unless you've done some heavy research & concluded that 500GH/s is, indeed, the lightspeed limit.  Is it?


Absolutely.  I'm glad you're beginning to come around to my way of thinking.  Can't blame me for arguing in the alternative when it's so darn inviting though, can you?

Please stick to one claim at a time.  You claimed the chip would be obsolete due to superior tech.  Obviously it doesn't matter how superior future tech is if no chips are ever delivered.   So HF delivers a chip but the chips are worthless because some as of yet superior tech renders it obsolete in the near future.  Claiming superior (>10GH/W) tech is right around the corner is an extraordinary claim regardless of who's is making the chips.  The same "obsolete" argument would apply to every ASIC in production right now regardless of who is making it.

Will HF deliver a chip?  Well that is an entirely different argument.  It seems kinda stupid to be debating the merits of the miner protection plan if you believe no chip will be delivered at all.  Right?

Let's slow down a bit.  HashFast is the one making extraordinary claims -- a team of unknowns bringing an unusual 28nm ASIC to market in record time.  With no previous ASIC experience.  Nothing but the founder's love of bitcoin for credibility.  And you want *me* to substantiate my claims?

Edit:  I'm still trying to wrap my head around what you just wrote...  A guy on the interwebz promises to make a 400GH/s 28nm ASIC, and you're fine with that, but *I* have to prove that it might be obsolete in seven months?!  (and yes, that's seven from now -- start counting from the refund date: Jan 1st, 2014)

Well I think the burden of proof would be on the one making the outlanding claims.  Can you beat 1GH/W by a factor of 10x?  Can anyone?  No other competitor has, no other competitor even indicates that is possible.  There is nothing to indicate that a 1,000% increase in performance is possible at 28nm.  If it isn't then these chips would be "obsolete" anytime soon.  This is not to say that the price of new hardware won't continue to decline, they will .  Still given the financial rewards for designing a superior chip the fact that nobody (not just HF but all competitors) has done so would mean claiming the chip will be "obsolete" requires a little more than a empty statement.

You're tossing around 28nm as some sort of hallmark of quality, a standard by which a chip is measured.  That's ridiculous.  My cat can cut & paste some cores & have them implemented in 28nm process.  Or a team of experienced engineers could go through multiple iterations of design/prototype/redesign to come up with a slightly better chip using the same 28nm tech.  Do you suppose a good design house could beat my cat by a factor of 10?  My cat don't think so, but i worry.  He's ordered himself a dev board & picked up the lingo.  He's pretty good at getting what he wants, so we'll see.

How would a 28nm chip be "old tech" in 90 days?  See anyone releasing a chip magnitudes higher efficiency?  The price of this chip (and all other ASIC) chips is almost guaranteed to decline but that doesn't make it obsolete.