The decreasing rate of performance jumps from 68nm to 16nm?

[CryptoPanda]

As far as I know with every jump to lower NM process the performance/power consumption ratio improves by significantly less, right?
for example, when we switched from 68nm BTF singles to 28nm antminers, we were humming with about 661%  more hashes per second for the same watts
(BTF single 68nm, was doing 60ghs with ~300W at the wall and antminer S3 28nm does 450ghs with ~340W at the wall,  so we have 5W per gigahash in the first case and 0.75W per gigahash in the second.  Which is 6.617647055 times more)

The 20np Neptune claims  0.57 watts per Ghs which is only 1.315789474 times more or 31% better performance

can someone speculate based on that what improvement we will have with the 16nm technology?

Other interesting points of speculation would be when we can expect 16nm (and further)


That info will be useful for some longer term planning

[Vortex20000]

No idea, but I'd be overjoyed if 16nm technology came out tomorrow.

[xstr8guy]

BFL's first ASIC was 65nm not 68nm. Avalon and ASICMiner were first with 110nm.

[CryptoPanda]

ok my bad
what do you guys think about the rest though?

[cryptcoin_de]

There are some rumours here that Globalfoundries is working at 16nm technology for Apple to add 16nm chips to the next generation IPhones.
For production level they need about 8 to 12 month from now on...
But rumours...

[CryptoPanda]

And  I suppose the efficiency improvement will be even lower than the last jump?

[Korbman]

what do you guys think about the rest though?

While I do lack a deep knowledge of chip architecture, I do know that shrinking from one die size to another doesn't automatically make the processor more efficient. That's something that's refined over time through updates to the chips, PCBs, and firmware / software. CoinTerra and KnC chips were 28nm, for example, but their wattage varied (~1.2-1.3W and 1-1.1W per GH, respectively). The newer KnC Neptune chips are built on a 20nm process, yet they use nearly half the energy as their previous 28nm chips.

As I think back (and someone correct me on this if I'm mistaken), I thought there was a time when ASICMiner's 110nm process ended up being just as, or more, efficient than BFL's 65nm (though that's not saying much I suppose).

Anyway, I'll just leave this here: http://arstechnica.com/gadgets/2014/08/broadwell-is-coming-a-look-at-intels-low-power-core-m-and-its-14nm-process/

[CryptoPanda]

From what I read seems like we will probably have just 20nm asics until the end of 2015 or so
Which means this time the equipment might become obsolete much slower