> All of a sudden they're competitive with NVIDIA Apple has no interconnect technology comparable to what Nvidia ships to datacenters. The larger Nvidia cluste

> All of a sudden they're competitive with NVIDIA Apple has no interconnect technology comparable to what Nvidia ships to datacenters. The larger Nvidia clusters measure their addressable memory in terabytes, the value of Unified Memory at that scale is practically negligible (if not wasted bandwidth). You're making some pretty handwavy generalizations here without a solid grasp on why Nvidia dominates GPGPU compute.