> What is it exactly that an Intel-based "supercomputer" can do that a hangar full of NVidia GPU's can't? 1. Crunch problems that are parallel but not "embarra

> What is it exactly that an Intel-based "supercomputer" can do that a hangar full of NVidia GPU's can't? 1. Crunch problems that are parallel but not "embarrassingly parallel", thus requiring a fast network to communicate between nodes. 2. Crunch problems that have not been ported from CPU code the GPU code.