San Jose, October 8, 2024 – Supermicro announced a high-density infrastructure platform optimized for AI inferencing at the network edge. Supermicro’s system delivers up to 10 double-width GPUs in a single system capable of running in traditional air-cooled environments, according to the company.
The SYS-322GB-NR includes two powerful Intel Xeon 6900 processors with P-cores, 8800 MT/s MRDIMM and up to 20 PCIe 5.0 expansion slots. The system supports a variety of single or double-width GPUs, or to use some of the expansion slots for high-performance I/O or other add-on cards. Additionally, it features up to 6TB of RDIMM memory and up to 14 E1.S or 6 U.2 NVMe drives.
As companies seek to embrace complex large language models (LLM) in their daily operations, there is a need for new hardware capable of inferencing high volumes of data in edge locations with minimal latency, Supermicro said.
“Owing to the system’s optimized thermal design, Supermicro can deliver all this performance in a high-density 3U 20 PCIe system with 256 cores that can be deployed in edge data centers,” said Charles Liang, president and CEO of Supermicro. “As the AI market is growing exponentially, customers need a powerful, versatile solution to inference data to run LLM-based applications on-premises, close to where the data is generated. Our new 3U Edge AI system enables them to run innovative solutions with minimal latency.”
One example use case that this system delivers is in the manufacturing industry, where Supermicro’s new system can be deployed on-site at an automated production environment to process data feeds from cameras and sensors without having to transfer the data to a remote location. This capability reduces networking requirements and improves response times. Another environment where the SYS-322GB-NR will excel is large-scale control rooms, where the AI accelerator cards can be partially replaced by multi-display cards to support up to 64 independent displays.
