They seem to be planning something like http://www.computermachines.org/joe/publications/pdfs/hpca2017_exascale_apu.pdf http://www.computermachines.org/joe/publ

They seem to be planning something like http://www.computermachines.org/joe/publications/pdfs/hpca2017_exascale_apu.pdf http://www.computermachines.org/joe/publications/pdfs/hpca20... Something like that could be a monster for HPC & ML type tasks. Provided AMD gets their software and tooling up to par; Nvidia doesn't rule HPC/ML because their GPU's are leaps and bounds ahead of AMD, but because CUDA is.. If so, might explain why they went with a (relatively) gimped vector FPU for Zen compared to Intel offerings (particularly the soon arriving Skylake Xeons with 512-bit wide SIMD).