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).
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).