Post
Topic
Board Hardware
Re: Who was first with 28nm?
by
thoughtcourier
on 05/12/2013, 16:31:32 UTC
I was with Sam from KnC a few days ago
KnC's inability to tell a consistent story was one of the reasons I happily chose to not to business with them. They've clearly stated both before and after product production that it was a structured asic run, and their power results support it. Ultimately it doesn't matter if they used crackerjack boxes to make their masks, at the end what matters is the specs and they're a mixed story. I mean, sure, feel free to not care.  But a 2x increase in operating cost, and thermal load is not "a few watts", especially for those of us not interested in a high risk gamble involving mining for a few months and then throwing the hardware out.

By all means, be happy with their product— they shipped a working device to many people mostly on time, better than a lot of other vendors, and many of those customers will be happy enough with a product that goes to the landfill before the bitfury and bitmain devices. Though in terms of feature size I don't see a reason to brag about 28nm when it doesn't achieve substantially better hashrate per U or hashrate per watt even close to multiple competing 55nm designs.

I am very happy with my KNC hardware, but I do have some questions for those who have been keeping up with KnC so I can properly evaluate their promises:

I assume the extra hashrate over the promise came from overclocking (400 -> 550) and overvolting (550 -> 650). Power consumption likewise had to increase at least linearly, but likely exponentially (Huh -> 600 -> 850).

So if this is correct, what's the news on underclocking/undervolting firmware or tuning API?

Would you have egg on your face if after some min-maxing, machines can run and say... 800 MH/J? Or rather, at what efficiency would you have egg on your face?