Samsung PM9D3 U.2 NVMe SSD 30.72TB

The Samsung PM9D3 is a sixth-generation enterprise NVMe SSD built on PCIe 5.0 x4 (NVMe 2.0) architecture and housed in the industry-standard 2.5-inch U.2 (SFF-8639) form factor. Leveraging Samsung's latest V-NAND technology paired with an advanced in-house NVMe controller, the PM9D3 achieves a dramatic uplift in sequential read bandwidth compared to the PM9A3 generation, making it purpose-built for the throughput-hungry demands of large language model (LLM) serving, AI training data pipelines, and hyperscale cloud storage infrastructure. The 30.72TB capacity point addresses the growing need to consolidate massive datasets into fewer drive bays, reducing rack footprint, power consumption per terabyte, and interconnect complexity in scale-out deployments.

The PM9D3 is optimized for mixed enterprise workloads that blend sequential streaming with randomized metadata operations. Its end-to-end data path protection, including power-loss protection (PLP) with onboard capacitors, advanced error correction (LDPC), and T10 DIF (PI) support, ensures data integrity in always-on enterprise environments. The drive also incorporates AES 256-bit self-encrypting drive (SED) capability with TCG Opal 2.0 and TCG Enterprise compliance, enabling hardware-accelerated encryption without impacting user-accessible performance.

Targeted at hyperscale cloud operators, AI/ML infrastructure teams, and enterprise storage architects, the PM9D3 is validated for deployment in NVMe-oF (NVMe over Fabrics) topologies, direct-attach server configurations, and high-density JBOF (Just a Bunch of Flash) enclosures. Its U.2 form factor ensures broad compatibility with existing enterprise server platforms from major brands, and its PCIe 5.0 interface doubles the available bandwidth over PCIe 4.0 predecessors, unlocking new throughput ceilings for storage-bound AI and analytics pipelines.

Ideal for

• AI/ML model serving and LLM inference where fast sequential dataset streaming from flash storage directly accelerates GPU utilization and reduces time-to-first-token latency
• Hyperscale object storage nodes requiring maximum capacity density per U.2 bay to store exabyte-scale unstructured data with minimal rack space
• High-performance data lakes and analytics platforms (e.g., Apache Spark, Presto) that demand sustained sequential read throughput for parallel query execution across petabyte datasets
• NVMe-oF storage fabric deployments where multiple hosts share high-capacity flash over RDMA networks, benefiting from the drive's low-latency PCIe 5.0 backend
• Video surveillance and media archive tiers requiring write-intensive, high-capacity NVMe storage with enterprise-grade data integrity and power-loss protection
• Containerized cloud-native workloads and Kubernetes persistent volume backends in hyperscale data centers that need high QoS consistency across thousands of concurrent I/O streams

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.