Doing file I/O at 20 GB/s would be quite nice for loading large AI models.
Could we create library bindings for the file I/O and the RDMA transport separately, so that you can read and write files at high speed, in addition to sending them across the network.
Doing file I/O at 20 GB/s would be quite nice for loading large AI models.
Could we create library bindings for the file I/O and the RDMA transport separately, so that you can read and write files at high speed, in addition to sending them across the network.