Topic: Bitfury: "16nm... sales to public start shortly" - page 23. (Read 108494 times)

Their version of "shortly" is a bit too long isn't it?

my lot of 5 s1s just showed up and ready to be converted to the new Sidehack miner. Look forward to seeing the progress and eventual for sale thread.

Exiting times, again!

Whatever is happening with these elusive chips, it doesn't appear to be having much of a positive effect on Bitfury's mining. I see they are down to around 5% of total hashrate today.
Even if you include the MLM BitClub Network who claim
it's still only barely 10% in total.
Chinese pools now account for more than 80% of the total, nothing to worry about.........

 Innosilicon is still around, though they've been mighty quiet about status on the A3 and A4 the last couple months. I suspect that one or both of those could be "any day now" given how little in advance they've tended to announce when they're releasing their chips/gear in the past. They also have a substantial existing chip design business outside the cryptocoin world after all, so they can afford to not "crowdadvancesalefund" their development, plus whatever they made from the A1 and A2 (LOTS).

 BW.com (LKetc) is still around, though status of when (if) they're going to start selling the B11 is unknown, they DID announce the next-gen stuff that should be able to compete pretty closely against the Bitfury stuff and it's pretty obvious that the B11 is actually in production.

 Avalon ... meh, their announced stance on the 14/16nm generation might very well manage to kill them as they're WAY far behind the development curve at this point.

 I don't see Bitmain getting most of the pie for much longer.


 Do remember that hope was the last thing out of Pandora's Box.

Apparently for the original Bitfury chips the numerical value of "suckage" was such that it was still worthwhile to buy/sell them.

IIRC the even the original Bitfury chip has some sort of bit vector that enabled/disabled individual hashing cores. I don't think that they would've removed it from their newest design. It is fairly simple design-wise to take care of clock-disabling and powering-off "cores shot".

I think many people in this thread are in their mind designing miner according to some ideal "specification-in-the-sky" or "spec-sheet-of-their-dreams". I'm all for doing idle speculation as a mind exercise when there's not much else to do. But if somebody wants to do non-idle stuff then it is better to learn some basics of engineering instead of incessantly placing bets on various pies-in-the-sky.

 

Were the first-gen chips outputting half speed because half the cores were shot, or was it a power limitation or similar issue?

If the chips with half the cores shot are still drawing full power for half the hashrate, efficiency sucks. If the chips with half the cores shot are drawing half power for half hashrate, efficiency is preserved but not power, so without extensive grading for balance they're useless for a string design (which Bitfury favors heavily). In either case, they blow the designed specs. Bitfury may find a use for chips with half the cores shot, but one way or another it's going to suck. Unless the design specs are assuming half the cores are shot, in which case they'd want to bin chips anyway and keep the fully-functional ones for themselves to build some legendary machines out of.

By now I probably should have my standard debunking of the "low yield" argument assigned to a single-key shortcut.
Bitcoin mining chips are too repetitive to apply the industry standard measures of yield.

The big names mentioned like Intel,Apple,etc. order extremely complex digital chips with very little redundancy. In particular the industry standard testing framework (JTAG) requires that all (or nearly all) flip-flops on the whole chip are threaded onto a single JTAG-chain for testing. Any break or short in JTAG-chain will make chip faulty even if the all non-testing logic is correct.

On the other hand Bitcoin mining chip consist of ridiculous amount of redundancy and SHA256D is self testing. therefore the standard JTAG chain would be a complete waste of space.

In case of Bitcoin mining chip a "50% yield" would mean than on the average chip about half of the hashing engines in the sea-of-hashers are working correctly. Such chips would be still commercially viable and sellable.

In addition to the above Bitcoin mining chip is nearly 100% self-love, it doesn't have to interface or be compatible with any external standard like DRAM or WiFi. Nearly all functional and timing violation could be worked-around at the driver level.

To refresh the history: the original Bitfury chip had yield of 0%: it was supposed to produce 5GH/s but only achieved 2-3GH/s. Additionally there was some scramble/permutation in the output logic that required inverse permutation in the mining software. Yet those chips sold quite well. Not only they sold, they also made money for buyers.

In summary: if you want to make non-absurdist conclusions about Bitcoin mining chips you'll need to use the yield measures used in the analog & mixed-signal fields, not the ones used in the digital logic field.

Along with clicking 'Ignore' ditto.

Back on topic, my thoughts re: chip delay.
Disclaimer: The following is purely an educated guess by me.
I have no 'in' with information on the hold up of BitFury's chip. My business is not directly involved in the actual production of physical silicon wafers/dies but is directly involved in what happens to the wafers full of dies (not BitFury's) needing to be tested and packaged into functional chips. Since current Node-size directly impacts what our end has to deal with I DO closely follow what is happening at the Foundies be they TSMC, Samsung, GloFo or others.
-End disclaimer

That said, I have a good idea on probably what is up. Right? Wrong? Time will tell.

As I have been saying since oh, around Nov. of last year when noise began popping up about 16/14nm mining ASIC's coming out: Great. Yes there are huge advantages to be had. Once the production technology become viable enough for boutique chips vs micro-processors for Apple, AMD, Samsung, HTC and Cisco (network 'fabric' switches/buffers).

Guess what? To meet a reasonable final price-per-chip needed for 16/14nm miner ASIC's the yields per wafer just ain't there yet. Period.

Per statements by TSMC earlier this year Apple, AMD, and Cisco will take >80% of their capacity at the 16nm node until around June when 30% (my note - hopefully) more overall capacity there comes on line. Per a few articles I've read pertaining to yields at the 16/14nm node from all foundries is just barely over 50% viable dies per run. Acceptable for Apple, AMD, et al as they can write off scrap costs as part of the dev costs for 16/14nm because they and the other companies I mentioned are literally funding all development at that node and have been for years. Yields like that are devastating to our mining chip needs.

That means that only 20% of capacity is available to other companies, eg  BitFury with no certainty of how many good dies per wafer they will get. They along with others have to wait their turn in line, probably set once a month or at best every 2 weeks, to use that capacity to its fullest.

If based on development runs there are tweaks that need to be made to the silicon the delays snowball as Engineering sample wafers make their way through the initial testing at the Foundry, then onto a packaging house to be again probed before dicing from wafers into individual dies and finally packaged into actual chips, then final (hopefully at full speed/power) testing. Only then are the said Engineering samples sent to BitFury and from them to integrators for design testing.

The one shortcut there is that I suspect BitFury have their own packaging house (for what they need it ain't rocket science) so no scheduling conflicts with other customers there. *If* they do own a packaging house, a huge plus to BitFury's bottom line would be that packaging & testing would move from an unavoidable cost to instead being a profit center

Anywho, I'd venture given Punin's acknowledgement of Kilo17 winning their bet that some design issues have arisen requiring respins to address. BFL (all their ASIC's), Bitmine.ch/Innosilicon (A1), and others all come to mind on what can happen to dramatically delay full production of chips.

Will these lower node chips from Bitfury (and no doubt Bitmain ) eventually reach full production mode and yields get better? I see no reason to say no. As for when, anyones guess on that.

Much to the credit of BitFury at least they have not taken the "Promise the moon for specs and pre-order now for delivery in ,!" route with the public!

Time to get off of the stump.

I hear you. I have had similar thoughts. It just takes too much effort IMHO.

sidehack's advice generally requires a lot less effort.

AleScamHole  is not the type of name that would do it politely.

I am thinking of an alter ego name for my self

BlowMeBitches
MoronsAllofYou
FoolsWhyBotherWithBTC

All come to mind. For a nasty poster name.  Just post really nasty stuff.  Somedays I think I could do this not most.

   


As for bitfury, bitmaintech and or Avalon.

  I prefer bitfury to do the right thing  and I believe they will

Step zero - don't be an ass
Step one - display proof in a rational manner
Step two - discourse civilly

Did Bitfury ever build their own stickminer? All the ones I know of were independent projects.

And if Bitmain becomes our only hope, we're screwed. In the past year they've really only sold overpriced gear with mediocre performance and reliability and, since the market puts up with it, will probably continue on that trend. I'd like to see Spondoolies come out ahead; their stuff was overly expensive but at least it was because of complexity instead of greed, and they didn't intentionally abuse their customer base.

alh - I have signed no agreements limiting my ability to do or say whatever I want. That said, I handed off chip acquisition and the details thereof to kilo17, since he's more administrative and has more money to wave around, so I don't really know any details. I have not received chips or documentation; I have been asking for documentation for over a month. I'm still working on reverse-engineered Avalon A3218 stuff for concept testing.

Are you "unbound" enough that you can disclose your state of any BitFury centric work?

I assume you have received neither documentation, nor actual chips from BitFury?

Chips can be late for a variety of reasons outside of BitFury's control. Documentation though should be available entirely at their discretion and interest in putting forth the effort. That's kinda worrisome.

I assume there's some private miners out there with bitfury's new chip, but they must be silently replacing their current miners in datacenters to keep it secret. We don't know crap for sure now, the global hashrate could be 0.01% 16nm bitfury or 10%, nobody's sure. I liked the past with bitfury building their own miners, but sadly those days are over and the old stickminers by bitfury are gone. Who knows, they might never release this chip and hog it themselves so that difficulty drives skyward and bitmain will become our only hope. Purely speculation now.

Hey retard, nobody scammed me. HashFash was just an example which I smelled as a scam long before most of the forum users. Go back to your cave!


And I will be reporting your ass to the mods all day long!

Thanks for taking the time to answer, appreciated.

That's the general assumption. Bitfury's good at being quiet about their stuff, so there's no telling for sure what exists and what doesn't, but from what we've heard there is no mass deployment of their new chips yet. They probably won't be building miners that see retail sales directly, but were planning on getting chips to third-party developers. I don't know that that has happened yet.

^^
Thanks, skimmed through the last pages. Not interested in buying myself, wanted to know if commercial mining farms are using the chips & guess at what percentage of the hashrate they are responsible for.

From skimming this thread, I'm guessing no & none?

Scroll up a bit.