Supermicro JBOF All-Flash Expansion SSG-136B-NE384R

The Supermicro SSG-136B-NE384R is a 1U Just a Bunch of Flash (JBOF) expansion enclosure purpose-built for scale-out AI and HPC storage architectures. Accommodating up to 32 E3.S 1T form-factor NVMe SSDs in a compact 1U chassis, the unit achieves raw capacities exceeding 1.5PB in a single rack unit when populated with high-density E3.S drives, making it an exceptionally space-efficient solution for data centers constrained by floor space and power budgets. The architecture leverages PCIe Gen 5 fabric connectivity, enabling aggregate sequential read throughput in excess of 200 GB/s per enclosure—directly addressing the insatiable bandwidth demands of large language model training, distributed inference pipelines, and checkpoint-intensive deep learning workflows feeding modern GPU clusters.

The SSG-136B-NE384R connects to host servers via dual NVMe-oF (NVMe over Fabrics) capable controllers using PCIe or RDMA Ethernet interfaces, providing low-latency, high-throughput access that eliminates traditional storage bottlenecks between compute and persistent flash tiers. The enclosure integrates Supermicro's intelligent management subsystem, exposing drive health, thermal telemetry, and enclosure status through IPMI and Redfish-compliant APIs, enabling seamless integration into enterprise monitoring and orchestration platforms. Redundant 2000W platinum-efficiency power supply units and hot-swap fan modules ensure continuous operation under sustained full-bandwidth workloads.

Launched in Q4 2025, the SSG-136B-NE384R is positioned as a direct-attach or fabric-attached flash tier for GPU server clusters built around platforms such as Supermicro's HGX H100 and H200 systems. Its dense E3.S 1T slot layout, PCIe Gen 5 backplane, and tool-less drive replacement capability make it operationally suitable for hyperscale AI infrastructure teams requiring rapid drive serviceability at scale. The enclosure is fully compatible with open standards including NVMe 2.0 and NVMe-oF specifications, ensuring broad interoperability across host operating systems and storage software stacks such as VAST Data, WekaIO, and open-source Lustre or GPFS deployments.

Ideal for

• Scale-out flash storage tier feeding GPU clusters during large language model (LLM) training, providing sustained 200+ GB/s bandwidth for dataset ingestion and gradient checkpointing
• AI inference serving infrastructure requiring ultra-low latency model weight loading from persistent NVMe storage directly attached via PCIe Gen 5 fabric
• High-performance computing (HPC) scratch storage for simulation and computational fluid dynamics workloads demanding parallel I/O across thousands of concurrent processes
• Enterprise data lakehouse acceleration where columnar analytics engines such as Apache Spark or Trino require burst-read throughput beyond what SATA/SAS arrays can deliver
• Checkpoint and restart storage for distributed training frameworks (PyTorch FSDP, JAX), minimizing recovery time from node failures in multi-thousand GPU clusters
• Media and entertainment render farm shared storage for 8K RAW asset streaming and visual effects pipelines requiring consistent low-latency random and sequential access

Technical specifications

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