Solidigm announced today (October 7) the launch of its new AI Central Lab, creating the industry's highest-performance and highest-density storage testing cluster tailored for current AI workloads. This marks a new phase in the integration of AI and storage technologies.
This new lab is located in Rancho Cordova, California, within the FarmGPU facility adjacent to Solidigm's headquarters. Its primary purpose is to enable Solidigm to execute and analyze real-world AI workloads on NVIDIA's latest-generation B200 and H200 GPUs, delve into how cutting-edge storage technologies can enhance cluster computing performance, and open it to partners for validation and joint innovation.
Avi Shetty, Senior Director of AI Ecosystem and Strategic Partnerships at Solidigm, said: "The AI Central Lab combines the most powerful GPUs available today with Solidigm's leading storage architecture, allowing customers and developers to test performance, analyze bottlenecks, and continuously optimize in real-world AI workloads. These resources, previously limited to large enterprises, are now open to a wider range of partners, demonstrating the importance of bringing storage closer to GPU design."
Built using industry-standard reference architectures, the lab allows customers to test complete solutions powered by Solidigm SSDs in a standardized environment consistent with large global data centers. This not only helps simulate real-world AI application workloads but also enables comprehensive analysis at the server, rack, and even cluster levels.
Avi Shetty emphasized that simply testing storage devices in the past is no longer sufficient to meet the demands of the AI era. "We can now run real AI models directly in the lab and analyze system performance and energy efficiency through telemetry, while also understanding the actual storage requirements of different AI workloads."
Key highlights of the AI Central Lab include:
• The highest performance storage cluster:Using the ultra-high-speed Solidigm D7-PS1010 SSD, each node can achieve 116GB/s throughput in the MLPerf Storage test and supports multi-node expansion.
• Highest density storage architecture:配置192顆Solidigm D5-P5336 SSD，單顆容量達122TB，僅用16U機架即可達到23.6PB儲存容量。
• Full computing and network support:Equipped with NVIDIA B200/H200 GPUs, 800Gbps Ethernet technology, and integrated with high-efficiency server platforms from multiple partners.
The AI Central Lab supports a variety of AI workloads, including GPU-heavy model training and inference tasks, and tests emerging technologies such as KV cache offloading and VectorDB fine-tuning. The lab also measures the energy consumption of various system configurations to assist in the development of architectures optimized for energy efficiency. It also establishes industry-relevant AI performance evaluation standards using metrics such as tokens per dollar and tokens per watt.
Several organizations have already collaborated with Solidigm's AI Central Lab. Solidigm and the Metrum AI team have jointly developed a new technology that offloads AI data from memory to SSDs, reducing memory usage by up to 57% during RAG (Retrieval-Augmented Generation) inference.
In addition, Solidigm has also been working closely with FarmGPU to continuously optimize the performance and energy efficiency of AI infrastructure and promote the integrated development of storage and AI computing.
Through this new laboratory, Solidigm not only strengthens its leadership in AI storage, but also provides a practical verification platform for global partners to accelerate the innovation and implementation of AI applications.
