Yeah, it is fairly flat. The upperbound on clocking is going to depend on a lot on how easy it is to keep the chips cool. If they're like AM's chips of the past though, they won't mind high temperatures too much, but in a 5x5mm package (if I'm remembering right, it's several pages back) that's a real question. I know BE100 chips, in I believe a 6x6mm package, it took some rigging to make them run reliably at 4W dissipation. That was a PCB-cooled chip with not-the-best PCB-heatsink contact though (at least on Blades), so maybe a top-cooled chip will have a better overall thermal resistance between case and sink?
5x5mm is a 30% reduction in surface area which is a huge hit. Getting that heatsink right is going to be mighty important as to the formfactor of the device. Nice thin blades like in the Cube at this point are really required to get any sort of 'next generation' density out of these things.
I'd like to see about a 30-chip board that can be software volt/clocked between 100 and 300GH with a quiet fan and can run off a DC brick. That'd make a really good entry-level miner, Jalapeno or U3 market sector.
Maybe I should just go design it. But that's about three projects away. But I think it's a good idea.
Yeah, this chips scales quiet good. Bitmain's chips don't do well in that matter. Going to double hashrate means decreasing efficiency almost twice.
With AM chips double hashrate = only 25% worse efficiency.
3.6GH/s | 0.2095W/G
7.2GH/s | 0.2495W/G
That's not really a fair comparison when the BE300 is running / being tested in a range far far lower than the BM1384 is running at. If the graph continued back towards 0 it would probably look very, very similar to that of the AM chip.