Summary
Supermicro has launched a new JBOF (Just a Bunch of Flash) storage system powered by NVIDIA BlueField data processing units (DPUs). This system is designed to optimize AI storage by providing high-performance, low-latency storage solutions for AI training and inference workloads. The integration of NVIDIA BlueField DPUs enables hardware-accelerated data processing, reducing latency and improving efficiency.
Revolutionizing AI Storage with NVIDIA BlueField-Powered JBOF
The world of artificial intelligence (AI) is rapidly evolving, and with it, the need for high-performance storage solutions. Supermicro, a leading provider of IT solutions, has recently launched a new JBOF storage system powered by NVIDIA BlueField data processing units (DPUs). This system is specifically designed to optimize AI storage, providing the high-performance, low-latency storage solutions needed for AI training and inference workloads.
The Challenge of AI Storage
AI training and inference workloads present unique storage challenges. Training requires high throughput for large datasets and write-intensive operations, while inference demands exceptional read performance and low latency for real-time responsiveness. Traditional storage systems often struggle to meet these demands, leading to bottlenecks and inefficiencies.
The Solution: NVIDIA BlueField-Powered JBOF
Supermicro’s new JBOF storage system addresses these challenges by integrating NVIDIA BlueField DPUs. These DPUs are designed to optimize workloads for training and inference by offering hardware-accelerated data processing. This means that data can move directly between SSDs and high-speed network ports without needing to be processed by an external CPU or cross a separate PCIe bus multiple times.
Key Features and Benefits
- High-Performance Storage: The system supports up to 400 Gb/s of network traffic and can accelerate NVMe over Fabrics (NVMe-oF) storage protocol and other RDMA-based storage traffic.
- Low Latency: Testing has shown that the BlueField-based JBOF has 13% lower latency for small-block (4 KB) random read workloads compared to traditional X86-based JBOF.
- Power Efficiency: The use of a single DPU card to replace the CPU, memory, networking card, and BMC results in significant power savings, up to 50% for the non-SSD subsystem or 10-15% power savings for the entire JBOF.
- Scalability: The dense design is ideal for both scale-up and scale-out file and object storage, supporting the needs of AI training and HPC workloads.
Technical Specifications
| Specification | Detail |
|---|---|
| DPU | NVIDIA BlueField-3 |
| Networking | 400Gb Ethernet or InfiniBand |
| SSD Support | 24 or 36 SSDs, E3.S and U.2 form-factor |
| Raw Capacity | Up to 1.105PB using 30.71TB SSDs |
| Architecture | Active-active clustering for high availability |
Partnerships and Ecosystem
Supermicro has collaborated with NVIDIA to create a new JBOF ecosystem, including partnerships with Hammerspace and Cloudian. This ecosystem enables storage infrastructure software platforms to run natively on the NVIDIA BlueField-3 DPU in the Supermicro JBOF, further enhancing the efficiency and security of the solution.
Industry Impact
The launch of Supermicro’s NVIDIA BlueField-powered JBOF marks a significant advancement in AI and HPC infrastructure. By providing a highly integrated storage approach for large-scale AI training and analytics storage workloads, this system offers a more power-efficient solution than CPU-based storage servers and lower read latency due to its direct data path architecture.
Conclusion
In conclusion, Supermicro’s new JBOF storage system powered by NVIDIA BlueField DPUs is a game-changer for AI storage. By providing high-performance, low-latency storage solutions, this system is poised to revolutionize the way AI training and inference workloads are handled. With its dense design, high networking throughput, and significant power savings, this system is an essential tool for anyone looking to optimize their AI storage infrastructure.