NVIDIA GTC 2025: Supermicro Introduces 1U Petascale All-Flash Storage Server using NVIDIA Grace CPU Superchip

At NVIDIA GTC 2025, Supermicro, Inc. announced a new optimized storage server for high-performance software-defined storage (SDS) workloads.

This first-of-its-class storage server from a major Tier 1 provider utilizes the company‘s system design expertise to create a high-density storage server for software-defined workloads found in AI and ML training and inferencing, analytics and enterprise storage workloads. Supermicro worked closely with NVIDIA Corp. and WekaIO, Inc. to enable customers to create power-efficient high-performance storage systems for their AI factories.

“The Supermicro Petascale server with the NVIDIA Grace CPU Superchip highlights Supermicro’s focus on innovation and customer choice,” said Charles Liang, president and CEO, Supermicro. “We have developed the Petascale storage server using the NVIDIA Grace CPU with 144 Arm Neoverse V2 cores enabling high-performance I/O for software-defined storage workloads. We have demonstrated that the system can fully unleash the system’s PCIe Gen5 performance SSD bandwidth with linear scalability. Supermicro continues to bring to market the most advanced and optimized storage solutions available.“

Utilizing the company‘s Building Block Solutions system design strategy, the ARS-121L-NE316R 1U storage system incorporates a symmetrical architecture perfect for scaling AI workloads. The Supermicro Petascale storage server using the energy efficient NVIDIA Grace CPU Superchip, supports 16xhot-swap EDSFF PCIe Gen5 E3.S NVMe drives enabling 983TB raw capacity using 61.44TB SSDs and a rack containing 40 systems provides 39.3PB of raw storage capacity.

“We worked closely together on the development and testing of Supermicro’s storage server with NVIDIA Grace, giving storage software providers a range of hardware choices to meet different application requirements,” said Rob Davis, VP, storage networking technology, NVIDIA Corp. “This new storage server allows customers to benefit from NVIDIA’s innovations in both CPUs and DPUs to accelerate networking protocols, such as GPUDirect Storage, while using less power than comparable x86 servers.“

Supermicro is collaborating with WekaIO, Inc., a leader in high-performance storage software for AI, HPC and enterprise data management workloads. The combination of the 144 Arm cores, integrated 960GB of LPDDR5X on the CPU platform, and ability to support two NVIDIA BlueField-3 or ConnectX-8 SuperNICs will unleash the full performance of the WEKA Data Platform‘s zero-copy architecture, providing the highest performance with lower power utilization compared with x86-based systems.

“WEKA is proud to be collaborating with Supermicro and NVIDIA in creating a high-performance solution optimized for enterprise AI and ML training, analytics, and inferencing,” said Nilesh Patel, CPO, WEKA. “Initial testing of WEKA Data Platform software running on Supermicro’s petascale storage server with the NVIDIA Grace CPU-based pre-production cluster shows significant performance scaling and power efficiency.“

Supermicro and NVIDIA will present this Petascale storage server using the Grace CPU in more detail at NVIDIA GTC in San Jose, CA, starting on March 17 in an On-Demand session (S74296).

click to enlarge

Resources:
Supermicro WEKA Distributed Storage Solution     
Supermicro storage solutions for AI