for example puzzle 129 bit can be reducted to 2**90 bit -
by one key?
That's what I'd like to know as well - it does not seem feasible to condense a 129-bit range into a 90-bit range without accounting for all of the divisors and hence all the keys it will result.
you need about 1,5 TB ram.
That's basically enterprise server kind of hardware, right?
Do those even support GPUs at all? I mean, maybe if it had enough PCIe lanes, you could fit a few NVIDIA gpus in there and get to work on CUDA. I guess it would have to be assembled manually and it would be very large.