Re: Bitcoin puzzle transaction ~32 BTC prize to who solves it
I wasn't aware of such limitations of tensor cores now it makes sense why they haven't been used for key enumeration yet. I'm curious, if it were possible to leverage them at least for auxiliary inversion, what theoretical speedup could be expected?
If the inversion step is commented out of the code (basically making it a no-op and ignoring that the addition results are incorrect), the speed of EC public key addition, on a GPU, doubles, more or less. So the maximum speedup of the EC workload would be at most 100%, but this assumes an ideal condition that inversion is offloaded entirely to Tensor cores, which is unrealistic. No one can know until an implementation actually can take advantage of those cores, but we need an algorithm to exist first
I ignored everything related to the hashing workload here.