Cuckoo spends only 1/3 of its time doing hash computations, which take little power compared to DRAM switching rows.
So the possible gain by ASIC hashing is less than 33%.
Sorry not trying to make more work for you, but would you happen to have a reference for that claim? I am curious how you measured that? I remember you told smooth and I that you don't have a Kill-A-Watt meter to measure GPU power consumption.
I didn't measure that. I estimated that doing 90 64-bit operations (one siphash call) is cheaper than operating the (typically 16k)
sense amplifiers to read a new bitline. But I could be wrong about that...
I'm assuming a Cuckoo size much larger than 16k of course.
What about multiple processors sharing the same memory to amortize the memory electrical cost?
If they share the same memory bank then they likely cause more row switches and there is negligible amortization.
I thought Wagner's algorithm sorted. I'll have to reread the paper more carefully.
That puzzled me too. I don't see a need for sorting, just for binning to find collisions on the next chunk of n/(k+1) bits...,
and I will only believe sorting is more performant when I see it (i.e. when optimised code for that is made public).
If lookup is in the static RAM cache, then ratio of computation to memory power consumption should increase.
Their stated goal is a 1GB memory footprint, so we can rule out fitting in cache.