interesting project ...
did you have a working prototype ... ? to test heat, power drain and hashrate ...
I'm unsure but from my experience, already getting 2 GH from one chip drains a lot more power and generates a lot more heat, then what they are supposed to use in there normal 1.5 GH running mod.
Did you calculated with that? ... Just looking at my own boards ... 1.6 more power drain .... fully cooled the pcb temp. went from 41° to 49° ... having front and back fans with 1200 rpm .... and most of the chips are still running under 2 GH ...
To be quite honest I'm not concerned as much about the temperature. If we have to add a small heatsink - so be it. They're cheap. The board is already designed for one, so it will be just a matter of choice. The goal here is to get the most hashes for the $. If adding a $5 heatsink will get us 0.5GH more then that might be worth the effort. If adding it will give us 0.1-0.2GH then maybe not as much.
There were also plenty of notes on bitfury's FREE MONEY thread with feedback and also a lot on the Russian thread. According to this:
Latest test was done to maximize performance, while
keeping chip temperature below 50 degrees C. A small heatsink was mounted on the bottom of the PCB, and cooled with a fan to control the temperature.
Vcore at 0.835 Volts, resulting in 2.5A current (2.1W). Clock frequency of 230 MHz. Instead of bitfury's test vectors, a test was done with 2048 more or less random vectors. Based on 756 cores, a yield of 1512 valid hashes would be expected, but 1511 were found, so COP is very close to 1.
Net hash rate would be almost 2.7 GH/sec @ 0.8W/GH.
http://i.imgur.com/g7UTw6V.jpgIn our case VCore will be lower - 0.800V-0.815V at least initially. There are also various options for the clock settings - instead of 54 bits we're looking at 53.
Also
bitfury's test results using the same board - roughly translated via google-translate:
(...)
Measurements of the errors and hash rate values were done using the first 100 test vectors , the board is not actively cooled (impressive - running on a passive 2-layer board made by Rene - intron - by the way big thanks to him) (...)
So the results : (...)
0.596 V, 45 Mhz, 0.39 Amps, 0.23 W, 100 tasks sent, 145 solutions got, COP is 1.0, Ideal 0.52 Gh / s, Real 0.52 Gh / s, 0.44 W / Gh / s
0.596 V, 96 Mhz, 0.725 Amps, 0.43 W, 100 tasks sent, 143 solutions got, COP is 0.986, Ideal 1.12 Gh / s, Real 1.1 Gh / s, 0.39 W / Gh / s
Please note that the 96 Mhz looks very close to the best solution ... After I started to raise the voltage without changing the settings of the oscillator - it was on purpose not voltage stabilized and its oscillation frequency should follow the supply voltage. What is most interesting that, when the oscillation frequency increases , the error-rate remained virtually unchanged, thanks to matching the characteristics of the oscillator transistor and transistor logic within a single crystal.
0.7 V 140 Mhz 1.2 Amps 0.84 W 100 tasks, 143 solutions got, COP is 0.986, Ideal 1.63 Gh / s, Real 1.6 Gh / s, 0.52 W / GH / s
0.8 V 180 Mhz 1.87 Amps 1.49 W 100 tasks, 139 solutions got, COP is 0.958, Ideal 2.093 Gh / s, Real 2 Gh / s, 0.75 W / GH / s
0.9 V 214 Mhz 2.62 Amps 2.36 W 100 tasks, 132 solutions got, COP is 0.91, Ideal 2.48 Gh / s, Real 2.26 Gh / s, 1.04 W / Gh / s.
Did not test any further - does not seem to have enough cooling.
So getting the 2GH is not the question. It is if the passive cooling will be sufficient. If we get better bang for the buck at 1.8GH then that may be the "stock" hashrate. The primary goal here is simplicity, even if that means that we may have to sacrifice a few hashes.
Have in mind that those chips are quite "overclockable" - there is nothing stopping you from trying to get 3.2GH/s. The board would likely still work just fine except that you'll then have to add some active cooling.
). We'll see how that goes. We may end up shooting them first anyways. 
