Intel Optane Persistent Memory: Architecture, Programming Model, and Development Kit Overview

Intel® Optane™ Persistent Memory (PMem) combines large‑capacity, cost‑effective memory with data persistence, offering performance higher than NAND and close to DRAM, making it suitable for AI, big data, and cloud workloads.

PMem can be used in two transparent ways: as a storage device accessed via standard file APIs, or as memory‑mode where it behaves like volatile RAM; each mode has trade‑offs in latency, bandwidth, and persistence.

To fully exploit PMem, the AppDirect mode allows applications to access persistent memory directly from user space, providing both high performance and on‑site data persistence with minimal code changes.

The industry‑standard SNIA NVM programming model defines a common API for persistent memory; Intel and Linux/Windows vendors have implemented this model, enabling direct access (DAX) and persistent‑memory‑aware file systems such as ext4, XFS, and NTFS.

Intel’s Persistent Memory Development Kit (PMDK) implements the SNIA model and offers a suite of libraries—including libmemkind, libvmemcache, pmemkv, libpmemobj, libpmemobj‑cpp, and libpmem—that simplify memory allocation, transaction management, key‑value storage, and low‑level memory operations on PMem.

These libraries provide both volatile and persistent APIs, support byte‑addressable data structures, and integrate with cloud‑native environments, allowing developers to build scalable, fault‑tolerant applications without extensive kernel modifications.

Additional resources such as Intel’s documentation, SNIA specifications, and open‑source repositories (e.g., GitHub) are listed for further learning and implementation.