Building a new super fast computer

[Dann Corbit]

I think that the best results will come from a hybrid approach.
Have a giant pile of threads doing clever alpha-beta.
Perhaps 64 threads are doing Lazy SMP shared hash only and 64 threads are doing Monte Carlo and 64 threads are doing ABDADA search and 64 threads are doing tactical search
Have a box full of GPUs doing clever learned intelligent approach.
Maybe 4 GPU cards are doing Matthew Lai's learned approach.
Maybe 4 GPU cards are doing a proof number mate search using Ankan's move generator.
Stuff all results into a shared hash table that each approach can use to refine its own efforts.