7nm miner thread

[NotFuzzyWarm]

God forbid you actually READ this thread and understand that several folks here are involved in the semiconductor industry. When we say BULLSHIT we mean it.
Read the damn thread, explore for yourself where the chip industry is at and what the like of Intel, IBM, GloFo, Samsung and TSMC say about where the nodes sizes will be in a REALISTIC time frame.

The fact that GMO is releasing another crap-altcoin (that IS what 'Tokens' are) to rape investards scrape up capital just adds to the stench coming from the pile of their PR's to-date.

[ConnerM]

Do you have any specific info to back this allegation?

He doesn't really need any. Anyone familiar with the semiconductor industry immediately knows what GMO is claiming is utterly ridiculous. Pure PR and nothing more.

[Stedsm]

7nm chips will be coming out in early 2018 (or so they say).   Let us use this as the placeholder for discussions.    

One company has talked about it lightly in their presser - Canaan.   I assume they will be restrictive and not disclose much but we should watch for news coming from TSMC.  

With that said... Who is going to continue to ramp up operations with Avalon or S9 given these are coming out?   I would figure units like that go obsolete within the first release of 7nm.

It's obvious that everyone will try to grab their hands-on the latest 7nm chips which is less than half the current 16nm ones by bitmain. Avalon or S9s may go complementary when these will be out but after they enter the market, mining will no longer be a piece of cake for common people as those Giants in this industry are now here to stack up their wallets with the best hardwares and there will be Monopoly for sure for such entities where their money will speak for them.

[NotFuzzyWarm]

<snip>

It's obvious that everyone will try to grab their hands-on the latest 7nm chips which is less than half the current 16nm ones by bitmain. Avalon or S9s may go complementary when these will be out but after they enter the market, mining will no longer be a piece of cake for common people as those Giants in this industry are now here to stack up their wallets with the best hardwares and there will be Monopoly for sure for such entities where their money will speak for them.

What vital parts of this thread have you not read/understood?
The technology behind the low node chips - INCLUDING the current 16/14nm nodes - is geared for low power (typically meaning mobile) applications. A large part of the problems with 16/14nm being unstable is because of the miner makers pushing >10w into chips with over 2k cores in them using process technology that is optimized for far lower power. As in less than 1W sustained and average of less than 500mw. The physical structures are being pushed to the max right now.

10nm and lower - assuming Bitmain et al continue their historic design philosophy of packing in as many cores as possible in into these massively expensive chips in turn pushing thermals to the max, I foresee terrible troubles...

Will they finally take a more conservative approach? They really have to but if they do it will be accompanied by much kicking and screaming because to them -

[Stedsm]

<snip>

It's obvious that everyone will try to grab their hands-on the latest 7nm chips which is less than half the current 16nm ones by bitmain. Avalon or S9s may go complementary when these will be out but after they enter the market, mining will no longer be a piece of cake for common people as those Giants in this industry are now here to stack up their wallets with the best hardwares and there will be Monopoly for sure for such entities where their money will speak for them.

What vital parts of this thread have you not read/understood?
The technology behind the low node chips - INCLUDING the current 16/14nm nodes - is geared for low power (typically meaning mobile) applications. A large part of the problems with 16/14nm being unstable is because of the miner makers pushing >10w into chips with over 2k cores in them using process technology that is optimized for far lower power. As in less than 1W sustained and average of less than 500mw. The physical structures are being pushed to the max right now.

10nm and lower - assuming Bitmain et al continue their historic design philosophy of packing in as many cores as possible in into these massively expensive chips in turn pushing thermals to the max, I foresee terrible troubles...

Will they finally take a more conservative approach? They really have to but if they do it will be accompanied by much kicking and screaming because to them -

It's all about profits in the end and that's the fact. Aren't these 7nm chips meant to consume the least possible electricity and give max returns in turn? I have read everything in the thread and I know that the current ongoing 16nm miners are doing their best, but if that's so, why are these people coming up with their own miners? That's basically monopoly and they are trying to capture the mining market by going completely "pervert".
Don't believe? Check out Russia's decision over mining.

[fanatic26]

Aren't these 7nm chips meant to consume the least possible electricity and give max returns in turn?

These mythical 7nm chips that only appear in press releases?

[QuintLeo]

Folks that are making their own mining chips on the 14/16nm node are few.

 Bitfury
 Bitmain
 BW.Com
 Caanan
 EBang

 One "huge farm" I am aware of that decided it was better to "roll their own" than to try to deal with the limits of Bitfury production and Bitfury pricing (nobody else is selling SHA256 chips AT ALL on the current 14/16nm node).

 I am not aware of anyone else on this list.


 I suspect that process variation is more of a problem than the "process isn't designed for high power" - consider that AMD and Nvidia routinely build 200+ watt chips on some of these same process, though they might not be quite as high a utilization percentage as a cryptocoin mining chip they aren't all THAT much lower on their GPUs.
 As the node sizes get smaller, process variation is going to get worse - the closer you get to "one electron = turn on or off", the worse it's going to get unless the industry moves to some method of nanofabrication where it can PLACE single atoms where they are needed.

 (Yes, that CAN be done - IBM did it years ago to create an atomic-scale "IBM logo" - but it's not even close to being anything but a laboratory stunt at this point).

[alh]

One other thing worth noting. While it's true that Intel, AMD, Nvidia and so forth produce 100+ Watt parts on the 14nm process, they have much bigger packages with thousands of pins and lot's bigger dies inside. Bitmain, and I expect all the others (e.g. Cannan et. al.) have gone with small packages as well. I expect that feeds into the power supply design, board layout, and cooling.

The last folks that I saw with a "large package" mining chip was Spondoolies. They were obviously willing to make the investment in supporting a larger chip. History might suggest that their choice was not a good one in terms of cost competitiveness, though they had reliable and configurable hardware.

Looks to me like Bitmain has decided the best way to deal with "variation" is to mix and match blades to get to a repeatable hashrate for the whole miner, without strictly consistent parts.

[QuintLeo]

Spondoolies wasn't competative with Bitmain because they build their miners to be configurable, NOT because their chips were too expensive or not comparable.

 If they had been SMART, they would have settled for fewer sales and NOT tried to compete on price - the same route Caanan is taking with the Avalon - except that the SP20 when set for the same hashrate as the S5 was almost identical efficiency level and ran a hair cooler.

 The SP20 compared to the S5 was a night and day difference in many respects - the SP20 could get down semi-close to the high end of S7 efficiency range (at a cost in hashrate), or it could be tuned to do 50% or so more hashrate than the S5 (at a cost in efficiency), and that tuning also allowed it to handle high temps a LOT better than the S5 could.
 You could also STACK SP20s, they were a lot more compact design in actual usage (longer, but lower in height and no cables/lack of a top made the EFFECTIVE height quite a bit lower, and almost identical width).
 The SP20 was also overall a much more PROFESSIONAL design - it's STILL a more professional design than the S7 or S9 or even the Avalon 741.

 Configurability and professional design came at a cost though.

[spazzdla]

Unless they are sub 1000 watts they will be useless to me ..

[promojo]

Spondoolies wasn't competative with Bitmain because they build their miners to be configurable, NOT because their chips were too expensive or not comparable.

 If they had been SMART, they would have settled for fewer sales and NOT tried to compete on price - the same route Caanan is taking with the Avalon - except that the SP20 when set for the same hashrate as the S5 was almost identical efficiency level and ran a hair cooler.

 The SP20 compared to the S5 was a night and day difference in many respects - the SP20 could get down semi-close to the high end of S7 efficiency range (at a cost in hashrate), or it could be tuned to do 50% or so more hashrate than the S5 (at a cost in efficiency), and that tuning also allowed it to handle high temps a LOT better than the S5 could.
 You could also STACK SP20s, they were a lot more compact design in actual usage (longer, but lower in height and no cables/lack of a top made the EFFECTIVE height quite a bit lower, and almost identical width).
 The SP20 was also overall a much more PROFESSIONAL design - it's STILL a more professional design than the S7 or S9 or even the Avalon 741.

 Configurability and professional design came at a cost though.

RIP Spondoolies... I loved them!  Great support and product too.