TL;DR:
- Supermicro introduces a comprehensive AI storage solution, optimizing data pipelines from collection to delivery.
- The solution focuses on AI training and can handle massive petascale data volumes, enhancing time-to-value.
- It offers multi-petabyte data capabilities for AIOps and MLOps in production environments.
- Charles Liang, CEO of Supermicro, highlights the solution’s impressive capabilities, including high-performance flash storage and GPU server optimization.
- Supermicro’s extensive portfolio of storage servers supports low-latency, high-bandwidth GPU productivity.
- Key components include All-Flash Petascale Storage Servers, 8U GPU Servers, and object tier storage servers.
- The solution is complemented by software like the WEKA Data Platform and Quantum ActiveScale object storage.
- Jonathan Martin, President at WEKA, emphasizes the synergy between Supermicro’s servers and WEKA’s AI-native data platform.
Main AI News:
Supermicro, Inc., a leading Total IT Solution Manufacturer specializing in AI, Cloud, Storage, and 5G/Edge technologies, has introduced a comprehensive storage solution designed to revolutionize AI and ML data pipelines. This groundbreaking offering spans the entire spectrum, from data collection to high-performance data delivery, setting a new standard for AI time-to-value by ensuring that GPU data pipelines remain consistently optimized. With a focus on AI training, this solution can seamlessly handle massive volumes of raw data at petascale capacities, transforming them into an organization’s AI workflow pipeline. This multi-tiered Supermicro solution has demonstrated its capability to provide multi-petabyte data for AIOps and MLOps in real-world production environments. Supermicro’s holistic approach aims to mitigate implementation risks, expedite model training, and enable rapid utilization of resultant data for AI inference.
Charles Liang, President, and CEO of Supermicro, elaborates on the solution’s capabilities: “With 20 PB per rack of high-performance flash storage, powering either four application-optimized NVIDIA HGX H100 8-GPU air-cooled servers or eight NVIDIA HGX H100 8-GPU liquid-cooled servers, customers can unleash the full potential of their AI and ML applications at rack scale. This solution boasts an impressive 270 GB/s of read throughput and 3.9 million IOPS per storage cluster in its minimum configuration, with the ability to seamlessly scale up to hundreds of petabytes. Leveraging the latest Supermicro systems, PCIe 5.0, E3.S storage devices, and the WEKA Data Platform software, users can expect significant performance enhancements for their AI applications. Our new storage solution for AI training empowers customers to harness the full potential of our advanced GPU server rack scale solutions, reducing their Total Cost of Ownership (TCO) while enhancing AI performance.”
In the realm of large-scale AI training, where petabytes of unstructured data are crucial, the key to success lies in providing GPU servers with low latencies and high bandwidth to maintain GPU productivity. Supermicro’s extensive portfolio of Intel and AMD-based storage servers plays a pivotal role in the AI pipeline. Notable components include the Supermicro Petascale All-Flash storage servers, boasting an impressive capacity of 983.04 TB per server of NVMe Gen 5 flash capacity, and delivering a remarkable 230 GB/s of read bandwidth along with 30 million IOPS. The solution is further complemented by the Supermicro SuperServer 90 drive bay storage servers designed for the capacity object tier. This comprehensive and rigorously tested solution is readily available worldwide to cater to the needs of customers engaged in ML, GenAI, and other computationally intensive workloads.
Key components of the new storage solution include:
- All-Flash tier – Supermicro Petascale Storage Servers
- Application tier – Supermicro 8U GPU Servers: AS -8125GS-TNHR and SYS-821GE-TNHR
- Object tier – Supermicro 90 drive bay 4U SuperStorage Server running Quantum ActiveScale object storage
- Software: WEKA Data Platform and Quantum ActiveScale object storage
- Switches: Supermicro InfiniBand and Ethernet Switches
Jonathan Martin, President at WEKA, emphasizes the symbiotic relationship between Supermicro’s All-Flash Petascale Storage Servers and WEKA’s AI-native data platform software, stating, “The high performance and large flash capacity of Supermicro’s All-Flash Petascale Storage Servers perfectly complement WEKA’s AI-native data platform software. Together, they provide the unparalleled speed, scale, and simplicity demanded by today’s enterprise AI customers.”
Conclusion:
Supermicro’s advanced AI storage solution presents a significant advancement in the market, providing organizations with a comprehensive and scalable infrastructure to accelerate AI and ML applications. With its impressive performance capabilities and reduction in TCO, it positions Supermicro as a key player in catering to the growing demand for efficient AI data pipelines and storage solutions, further driving innovation in the AI market.
