Memory Layouts for GPU-Data Transfer Buffering in SPH

Collection of systems and HPC papers (GPU/host transfers, GPU Confidential Computing on Blackwell, RDMA/CXL memory pooling and hash-table designs, MicroVM snapshot serving, multi-LLM serving and KV-cache disaggregation, federated learning, concurrency/synchronization primitives, and GPU-aware bioinformatics structures). Key findings: (1) GPU-CC (Blackwell/RTX Pro/B300) preserves raw compute but serializes host↔device transfers via a confidential VM–GPU bridge, causing large throughput and latency regressions for LLM serving and KV-cache restore; scheduling/workarounds recover a portion of the损