XCENA and SK hynix Show CXL Memory Options for AI Workloads

XCENA showed two demonstrations using MX1 CXL computational memory: sharing a key-value cache across inference servers through a CXL pool, and running vector search for retrieval-augmented generation inside the memory device. The source summary also says SK hynix unveiled a CXL Hybrid Memory product, but gives no specifications. The examples show efforts to use CXL to expand memory capacity for inference; no comparative performance figures appear in the excerpt.