The growth of AI is driving exponential growth in computing power and a doubling of networking speeds every few years. Less well-known is that it’s also putting…
Overview
Supermicro has launched a new JBOF powered by the NVIDIA BlueField data processing unit (DPU) to meet the growing demands of AI storage. This innovative design replaces traditional storage components with a single DPU, enhancing efficiency and performance for AI workloads.
What You'll Learn
1
How to optimize AI storage solutions using NVIDIA BlueField DPUs
2
Why object storage is preferred for AI workloads over traditional file storage
3
When to implement a JBOF design for high-performance computing tasks
Key Questions Answered
What are the benefits of using NVIDIA BlueField DPUs in storage solutions?
NVIDIA BlueField DPUs enhance storage solutions by offloading and accelerating networking, storage, and security functions. This integration allows for low-latency access and improved efficiency in terms of price, performance, and power consumption, making it ideal for AI workloads.
How does the new Supermicro JBOF improve storage performance?
The new Supermicro JBOF can support up to 400 Gb/s of network traffic and utilizes BlueField DPUs to manage SSDs and accelerate NVMe over Fabrics. This results in lower latency and higher throughput, crucial for AI training and high-performance computing workloads.
What storage capacities does the new Supermicro JBOF support?
The Supermicro JBOF supports either 36 E3.S SSDs or 24 U.2 SSDs, with a raw capacity of up to 1.44 PB, and can later support up to 2 PB with newer 60-TB SSDs, making it suitable for large-scale data storage needs.
What power savings can be expected with the new JBOF design?
The new JBOF design using a single DPU card can result in power savings of up to 50% for the non-SSD subsystem and 10-15% for the entire JBOF, which can lead to significant energy savings in large-scale deployments.
Key Statistics & Figures
Network traffic support
up to 400 Gb/s
This is the maximum network traffic the BlueField DPU storage controller card can handle.
Latency for small-block random read workload
86 µs
This latency is achieved by the BlueField-based JBOF compared to 100 µs for traditional X86-based JBOFs.
Power savings
up to 50%
This savings applies to the non-SSD subsystem when using the new JBOF design.
Technologies & Tools
Hardware
Nvidia Bluefield
Used as the data processing unit to enhance storage performance and efficiency.
Hardware
Supermicro Jbof
A new storage solution designed to optimize AI workloads.
Protocol
Nvme Over Fabrics
Accelerated storage protocol supported by the BlueField DPU.
Key Actionable Insights
1Consider integrating NVIDIA BlueField DPUs into your storage architecture to enhance performance and efficiency.This approach is particularly beneficial for organizations dealing with large AI datasets, as it reduces latency and improves throughput.
2Utilize the new Supermicro JBOF for scalable storage solutions in cloud environments.The flexible design and high capacity make it ideal for businesses looking to optimize their storage for AI and HPC workloads.
3Leverage partnerships with software providers like Cloudian and Hammerspace to maximize the capabilities of your storage solutions.These partnerships enhance the functionality of the JBOF, providing optimized object storage and data orchestration services.
Common Pitfalls
1
Underestimating the importance of low-latency access in AI workloads can lead to performance bottlenecks.
AI applications often require rapid data retrieval, and traditional storage solutions may not meet these demands, resulting in slower processing times.
Related Concepts
AI Storage Optimization Techniques
Dpu Architecture And Benefits
Scalable Storage Solutions For Cloud Environments