2K 93% Exploring Distributed Caching for Faster GPU Training with NVMe, GDS, and RDMA – Hope Wang & Bin Fan