WEKA previewed the first high-performance storage solution built on NVIDIA Grace CPU Superchip and delivered in a Supermicro ARS-121L-NE316R.
AT SC24 in Atlanta, WEKA previewed the industry’s first high-performance storage solution designed specifically for the NVIDIA Grace CPU Superchip. Developed in collaboration with NVIDIA, Arm, and Supermicro, this solution leverages the WEKA Data Platform software to provide unmatched performance density and energy efficiency for enterprise AI and high-performance computing (HPC) workloads. It features Arm Neoverse V2 cores, NVIDIA ConnectX-7 NICs, and BlueField-3 SuperNICs, running on a cutting-edge Supermicro storage server to address the growing challenges of data access, space, and power constraints in modern data centers.
Driving the Next Generation of AI Innovation
The NVIDIA Grace CPU Superchip integrates the capabilities of a flagship x86-64 dual-socket server into a single module powered by 144 Arm Neoverse V2 cores. This design delivers 2x the energy efficiency of traditional x86 servers. Paired with NVIDIA ConnectX-7 NICs and BlueField-3 SuperNICs, offering up to 400Gb/s throughput, the solution provides high-speed, low-latency networking essential for AI and HPC workloads. These advancements are coupled with WEKA Data Platform’s zero-copy architecture, eliminating I/O bottlenecks and dramatically reducing AI pipeline latency.
Combined with the WEKA Data Platform’s AI-native architecture, this capability enables organizations to accelerate AI workflows by up to 10x, reducing the time to first token and improving overall efficiency in AI data pipelines.
The Grace CPU’s LPDDR5X memory architecture, with up to 1 TB/s of memory bandwidth, eliminates bottlenecks, ensuring a seamless data flow. The WEKA Data Platform, leveraging distributed architecture and kernel-bypass technology, supports faster model training, reduced epoch times, and higher inference speeds, making it an ideal solution for scaling enterprise AI workloads.
This solution offers unprecedented energy efficiency, with the WEKA platform increasing GPU stack efficiency by 10-50x. By reducing data copies and leveraging cloud elasticity, the platform shrinks data infrastructure footprints by 4-7x, cutting carbon output by up to 260 tons of CO2e per petabyte annually and lowering energy costs by up to 10x. With the Grace CPU Superchip’s energy efficiency, organizations can meet sustainability goals while boosting AI performance.
Transforming data centers for enterprise AI
The WEKA, NVIDIA, Arm, and Supermicro collaboration underscores a shared vision for creating high-performance, energy-efficient solutions tailored to next-generation data centers. Nilesh Patel, WEKA’s Chief Product Officer, highlighted the transformative potential of this partnership, emphasizing how the solution accelerates AI and HPC workloads while reducing the time to actionable insights and addressing rising data center energy consumption.
The Supermicro ARS-121L-NE316R storage server features:
- 16 Gen5 E3.S NVMe SSD bays for high-performance storage.
- Three PCIe Gen 5 networking slots, supporting up to two NVIDIA ConnectX-7 or BlueField-3 SuperNICs and one OCP 3.0 network adapter.
- A design optimized for AI, data analytics, and hyperscale cloud applications.
Ivan Goldwasser, Director of Data Center CPUs at NVIDIA, praised the solution for seamlessly integrating the NVIDIA Grace CPU Superchip with the WEKA platform to improve data-intensive AI workloads. Patrick Chiu, Senior Director of Storage Product Management at Supermicro, described the Supermicro ARS-121L-NE316R Petascale storage server as the first storage-optimized server using the NVIDIA Grace Superchip, tailored for high-performance storage workloads such as AI, data analytics, and hyperscale cloud applications.
Availability
The WEKA and Supermicro storage solution featuring NVIDIA Grace CPU Superchips will be commercially available in early 2025.
Engage with StorageReview
Newsletter | YouTube | Podcast iTunes/Spotify | Instagram | Twitter | TikTok | RSS Feed
