2 ms·
On top of that, Nvidia has provided a unified addressing abstraction over PCI for a looooong time via CUDA: https://developer.nvidia.com/blog/unified-memory-in-
by talldayo 3y ago
On top of that, Nvidia has provided a unified addressing abstraction over PCI for a looooong time via CUDA: https://developer.nvidia.com/blog/unified-memory-in-cuda-6/ https://developer.nvidia.com/blog/unified-memory-in-cuda-6/
Customers like Pixar could probably push this even further, with a more recent Nvidia rack and Mellanox networking. Networking a couple Mac Studios over Thunderbolt doesn't have a hope of competing, at that scale.