some of it sounds almost too good to be true,
Yep, that's a biggie, I know bitcoin ASICs in general have a bit of a way to go before they're fully bleeding edge state of the art, but this seems like an exponential jump that's larger than might be expected. As if we were all using 8088s at 6mhz (PC/XT class) and then all of a sudden there's a 600Mhz Pentium III (Or if you like 486sx33s to 3.3Ghz i5s)
First of all, a 600Mhz PIII is actually a lot more then 100x as powerful as an 8088. The PIII can crunch through multiple instructions per clock, while the 8088 would take multiple clocks to do one instruction.
Also, keep in mind how old some of these processes are. 130nm chips first came out in 2001.
That's 12 years agoSo for example, 12 years is a bit less then the difference between the first Pentium chip (1993) and the Core 2 (2006). It's also about the difference between the 286 (1982) and the Pentium Pro (1995), as well as the difference between the P4 and today.
Now, you'll notice that the jump from the P4 to today is way less then the jump between the first Pentium chip and the 286. That's because thermal issues are a big deal now compared to when the first Pentium came out.
There's also the die size issue. HashFast may simply be making really large chips. That's what KnC plans to do, and they're planning on at least 100Ghash per chip, compared to 0.3 for the Avalon.