so looks like they made there 400ghash mark. You sure its only 4 dies?
Yes it is 4 dies per "chip". 96 double nonce cores per die or 768 hashes per clock.
This is also clearly visible on the PCB photo.
The big brown square in the middle is the substrate which forms the base of the entire chip package. The 4 squares on the substrate are the 4 dies. Just about everything else on the green PCB is what is required to funnel that much power into that small chip. I have to give them credit it is a pretty clean design. There is a 6 pin PCIe power connector at each end of the board currently not populated in this photo. Each half of the board essentially supplies power for two of the cores. The heat sink mounting holes are standard (looks like Intel hole spacing) so any off the shelf cooling component could be used without any custom work needed.
Quote
lets see if the keep the 500ghs overclockable. Would be nice to have a surprise and get 500ghs out of the box and 600 overclockable
As pointed out IF it is ~400 GH/s at 0.74V well that is incredibly low voltage. 0.85V would be 15% higher voltage, pushing it to 0.90V would be 22% higher. Clock rate should scale at least linearly with voltage increase (probably super linearly) albeit at lower efficiency (J/GH). I am not making a prediction on how much they can push it. Remember power usage in semiconductors increases by the square of the voltage increase and the chip already uses a lot of power in a small package. At some point you run into limits on how much power the PCB can deliver and how much power (heat) the cooling system can remove.
Still as tacotime pointed out the chip at 0.74V is a good sign. If they were struggling to get 400 GH/s at 0.90V that would be a bad sign as they would already be pushing the limits of what the chip can handle.
) 
