I can't be the only person that started scratching my head when it's announced that 10.488TH/s takes up 2 42u racks according to Ken.
Yes but.... Let's do the math with 28nm chips...
The numbers are:
30gh/chip
24 chips per card
230 cards per "system"
2 x 42U racks per "System"
Someone estimated it would take 190 racks to house 1ph of 55nm cards - so nearer 250 racks to host the 1.3PH ken suggested would come online.
This would be 125 "systems"
In 55nm with chips at 1.9gh/s that would be 10.5TH/s or 5.25TH/s per rack.
With 28nm chips it's 165.6TH/system or 82.8TH/rack
250 racks of 28nm chips is.... 20.7PH/s - this would only need 700,000 chips or around approx 200 x 28nm wafers. Or about $2mil-$5mil for NRE and wafer cost. For 3% TNHR if this came online in august...
So yes - the density of 55nm (1.9gh/s) chips might seem a little low, but the same space could eventually hold over 20PH of hashing power - that is a LOT of hashing power. Also the cost price of getting that hash power online is freakin tiny compared to current options.
So yes, we didn't start hashing when we'd hoped. Yes ken is pissing a lot of people off with shares (not getting trading sorted and dumping a tonne of shares at super low prices) - but things really are ONLY bleak if NOTHING ever happens. I'm of the feeling and understanding that with the engineering team and the tapeout of 55nm, the chance of "nothing" happening is... nil
Vince,
While I like your direction; I think you are extrapolating information that we just don't know yet.
- Die per board - Needed to estimate Board GH/s
- Die Size - Needed to estimate wafer cost
- Estimated power of 28nm
Honestly 30GH/s @ 28nm is better than I could have hoped for (assuming the die stays the same size as the 55nm or shrinks).
For comparison sake...
Cointerra 500GH per chip is 4 die... each die is 10x10 and produces 125GH.
A cointerera wafer would yield around 520 functional die per wafer. At a total hash estimated at 65,000GH.
Based on our 55nm asic; we are shooting for around 4.5nm x 4.5nm (Ken please correct me). Which as Ken indicated yields 6800 die per wafer or 12,920GH.
If Ken can increase the hash rate to 30GH without increasing the die size... at 6800 die per wafer you are looking at 204,000GH compared to Cointerra's 65,000GH.
if our die size increases to 10x10 (which is our package size) then we could be looking at 15,600GH per wafer.
Now this isn't exactly apples to apples... if Ken would tell us the die size of the 28nm we can better judge.
*** EDIT to correct numbers
6800 Yield would mean a 10sqmm die size or 3.16mm x 3.16mm
die per wafer calculator
http://anysilicon.com/die-per-wafer-formula-free-calculators/
Show Posts
