AMD ROCm 10.1 adds tools for moving AI data between storage and GPUs

AMD released ROCm 10.1 on October 5 with changes aimed at reducing friction between storage and GPU workloads, according to StorageReview. The release builds on AMD Infinity Storage and includes a hipFILE fast path, NUMA memory allocation, and LLVM 24. The source notes that AI checkpoints, model weights, and key-value caches can exceed accelerator memory, making data movement an important part of system performance. No benchmark results are provided in the supplied summary.