Topic: Cointerra Mining ASIC coming soon - page 17. (Read 35558 times)

Awesome. All I need now is some videos.

Bitfury was supposed to ship their chips in July.  So far I haven't heard anyone complain about not getting their chips.

First of all, there's no way a "legit" American company is going to sell shares on a virtual stock exchange at this point in time.  There are tons of laws about selling securities in the U.S. It's not going to happen.

Secondly, whether or not they can take pre-orders depends on how much they charge. If they're chips are cheap enough they could potentially still sell chips. It all depends on the $/Ghash. 

If they're $1/Gh/s, for example I'm sure they'll have plenty of people willing to buy from them. That would be like selling an Avalon for $80. Do you think you'd have trouble selling an Avalon for that price in November or December?

Not saying they'll hit that price, though. But the way things are shaping up people are probably going to be terrified to pre-order at anything but the lowest prices in Oct/Nov.

Which in turn means the network hashrate will likely start to top out.

At this point, though trying to make any kind of prediction impossible.

Well, I just checked Lockhart on wunderground dot com, and it's showing 96F / 108 heat index.  See ya in September.

And, you know I was joshin' about real Texans v Austin, right?  But every state has a town which many citizens are fond of prefacing with "The People's Republic of..."  Even many who, to their Father Confessor, will admit to being secretly proud of the town and its accomplishments.

Which side of the cellular? In the pocket or on the tower?

If you can answer the above question you can deduct my answer based on my previous post.

Uhm... I'm as REAL a Texan as it gets. With the possible exception of the politics, I'm pretty darn proud of Austin. No, I don't live there (I'm a little further south), but's it's a helluva fun place to visit! I'm always puzzled when people badmouth Austin...

BTW, it's sure been a mild summer so far...

So, what's your company USP (Unique Selling Point) compared with others?!
[/quote]

Our USP is highest performance for lowest cost and lowest power; $/GH and GH/W.

Hmmm, do you think that implies some advantage towards KNC, considering ORSoC is supposed to have cellular experience?

I don't like these "coming soon" announcements with no technical details of products coming out.

Having said that, if these guys are in Austin, I have a few trips planned there before the New Year, driving back and forth from Dallas, and would be happy to report on any demos or product these guys are developing if they are entertaining visitors.

I'm not going to fight with your strawmen about Nvidia/ATI. Just quoting for the future reference.

There's no need to bullshit here about "optimising datapaths". SHA-256 is basically just a pair of 32-bit-wide shift registers with some cobinatorial logic thrown in the feedback loops. The cryptographers at NIST/NSA/etc. worked really hard to make sure that this logic is not minimisable in any meaningfull way because that would make it susceptible to cryptoanalysis. The "architectural" tricks would've already been exploited by the cryptoanalysts. There isn't any way to optimize power by e.g. not clocking parts of the circuit when not in productive use, which is where the most of modern CPUs and GPUs save power. So please no further low-power bullshit unless you can tell us how your low-power strategy applies to a circuit with 50% signal toggle probability. Nobody's going to run a pocket bitmine on a battery power.

There are some implementation tricks possible that minimize the critical timing paths, but they are all already published in an open literature or at most behind the ACM/IEEE paywalls. Other designers already took advantage of them, bitfury even sort-of republished the information behind the paywall.

Anyone who implemented SHA-256, even in software, knows that it is a self-testing logic, so please no further bullshit about design for testability.

The last remaining avenue for the significant gains is by somehow exploiting the high toggle rate in the circuit. The gates and flip/flops change state at the rate very close to the maximum possible, which normally happens only in a test structure called ring oscillator. My personal bet is that the progress will come from designers that take advantage of that and instead of using the bang-bang static logic use something out-of-ordinary: maybe some relatively obscure dynamic logic or some low-noise logic current-mode logic (a.k.a. source-coupled logic). Or the combination of the above: DyCML. Or something which I haven't even heard of.

Anyway, if anyone of you is going to visit Cointerra: ask them about their BSIM4 models, which of the 5 process corners they've simulated thus far, pay close attention if they have anyone working in analog simulators to optimize metal thickness and optimize the transistor/gate geometry to facilitate the best power noise bypass.

Bitfury's 5GH chip currently hashes slower primarily because nobody had spend any time on the problem of heath-sinking the QFN package. Also QFN is wire-bonded, which basically is a collection of half-turn induction coils. Those require very careful analog resonant pin/pad connection design, which again nobody had spend time on.

The way I see it, the progress will come from another company, which has designers experienced in analog/mixed signal ICs and who did high-power designs like cellular tower radios or synthetic aperture radars.

Edits: spelling and underlining.

Sorry to break in, a quick question:  Is optimising layout to reduce noise & timing errors considered digital or analog domain? 

Are you saying there's no more gains to be had by optimising datapaths, efficient implementation and/or scaling up the number of hasher units?

and presumably you're saying that every time NVidia or ATI bring out a faster GPU that pushes faster math or higher polygon counts, that they have resorted to analogue optimisation to make it happen?  i really doubt that many people resort to analogue asic design.. it just takes too long and the result isn't always predictable nor accurately simulateable.  bitfury is a very notable exception (and not the rule).. but even bitfury expected a faster chip than they ended up with... (the labels on the chip say 5GH, which is at least double what they ended up with).   Im not knocking the bitfury chip nor its designer, who I think is awesome!  im just saying that expecting every asic designer to go into the analogue domain is unlikely and impractical.  theres plenty that can be done in the digital domain.

2112, how do you perceive Simon Barber and his technical crew?

This is the first paragraph that actually contained technical information.

Seems like they are a front-end house with crew of front-end designers. To make a big gain in hashing performance they will need back-end expertise in analog design, not anything related to CPU architecture, cryptography or design for testability.

From the content-free press release that says "today our CEO put the pants on starting with his left leg, the same way as he did it the past" one can clearly infer that they are scared shitless and witless. The usual ass-kising PR doesn't work well when the ass-to-kiss is distributed, like in Bitcoin.

I wouldn't expect any breakthroughs from them. CAD-monkeys they aren't, but they also are sailing for a territory that is new for them. I expect another safe static CMOS logic design like from every one else thus far. Don't expect them to explore the frontier of Bitcoin hashing. They gave themselves no time for it.

Obviously, there is a small possibility that the insipid PR is just a cover-up for the ace-in-a-sleeve of a designer who already did several iterations of the hasher design on a side while working at the previous job.

Just saying hi from Austin. It's about time. Let me know if you need any help.

Unwatching this thread because it is clearly going downhill.

You have a talented team, with reputation and a million in seed money and you yutzes want to know the price already.

Yep, even six weeks ago, you can find incontrovertible, no-uncertain-terms claims in posts (see the KnC thread) that anyone who believed in 28nm - based devices was believing in Unicorns.

Now, beyond the incredibly impressive bitfury "garage build," there are three groups of professional designers at 28nm.  Moves fast, this biness.

This is a bit OT, and probably obvious to most, but short & worth saying:

To sort the wheat from the chaff, use the Tilde Test.

The Tilde Test is simplicity itself:  If negating a statement (sticking a "~" in front of it) makes it unenlightening, the original statement is also unenlightening.
In other words, if ~A tells you nothing, then A *also* tells you nothing.

As applied to the standard PR copy above:
"Founder Dr. Naveed Sherwani's idea was to select best-in-class worst-in-class technology from the open market and apply it through an engineering process focused on three goals: low high cost, high low schedule predictability, and high low reliability."

The Tilde Test, AKA the Duh Test.  

Edit:  Just sick of reading through all the boilerplate PR, just realized above sounds patronizing.  Just a vent.

When I last looked deeply into 28nm early in the spring, it did have costs that seemed inconceivable for BTCers to support...... however it appears now that 28nm fab capacity has been significantly overbuilt, so the costs have been falling like a rock.

Some of course could be scammers, and some of course could be incompetent from either a business or engineering standpoint, but seems recent flurry of 28nm activity is not in and of itself suspicious for aforementioned reason.

So far None, Yes None of this guys have delivered on time. Butterfly, Avalon, will knc and bitfury delivers on time?