Samsung and AMD demo CXL memory pooling at OCP 2026: 38% lower TTFT

Samsung and AMD showed a CXL memory pool built on CMM-D modules for agentic AI at OCP Global Summit 2026 in San Jose. Samsung claims elastic allocation delivers up to 38% lower time to first token and up to 3.3x higher goodput versus static allocation.

• Demo runs an AI agent workload on a GPU server against a DCD-based CXL pool with elastic allocation across LLM instances
• Samsung claims up to 38% lower TTFT and up to 3.3x higher goodput vs. static allocation
• Internal AI Memory Node with CMM-D delivered 5x to 7x KV cache utilization via an L3.5 tier
• Booth also shows HBM4E, LPDDR5X-PIM, PM1763 PCIe Gen6 SSD and the 256TB BM1773