GMP used during 3 and a half years to solve MIT's LCS35

[Laurent Desnogues]

On Tue, Apr 30, 2019 at 11:00 AM Hans Åberg <haberg-1 at telia.com> wrote:
>
>
> > On 29 Apr 2019, at 23:26, Bernard Fabrot <bfabrot at gmail.com> wrote:
> >
> > Hi all,
> >
> > that'd be me:
> >
> > https://www.csail.mit.edu/news/programmers-solve-mits-20-year-old-cryptographic-puzzle
> >
> > I wanted to thank you a lot for GMP and let you know that GMP did the 79
> > 685 186 856 218 modulo/squarings without failing once.
>
> It says in the description that it is designed to not be parallelizable, but cannot that be done for the individual iterations? Posit each each iteration is done in one cycle, then with a 4 GHz clock, it solved is 6.5 hours.

I doubt one can design hardware that could compute 2048-bit squaring
mod 2048-bit in 1 cycle @4 GHz :-)  But I might be wrong. The team
doing it with FPGA reached 70ns per squaring (cf.
http://www.cryptophage.com/).

Bernard, impressive result!  I'm not surprised GMP could do that
reliably, what I find surprising is that no HW fault occurred over
such a long period.  Did you do anything special to check intermediate
results?  And if so how often did you have to remake part of the
computation?

Laurent