REPACSS Facility Overview

The REPACSS is a high-performance computing data center and AI infrastructure designed to support intensive computational and data-driven research, powered by variable energy sources. The system consists of a combination of compute, GPU, and storage nodes, interconnected with high-speed networking to ensure efficient data transfer and processing. The facility is structured to accommodate diverse workloads, including large-scale simulations, AI training, and data analytics.

System Components

• Compute Nodes: REPACSS features 110 AMD-based compute nodes, each equipped with dual AMD EPYC 9754 processors, totaling 256 cores and 1.5TB of DDR5 memory per node. These nodes are designed to provide robust concurrent processing capabilities for demanding scientific computations.
• GPU Nodes: The system incorporates 8 GPU nodes, each featuring dual Intel Xeon Gold 6448Y processors and 4 NVIDIA H100 NVL GPUs with NVLink interconnects. These nodes are optimized for machine learning and deep learning applications, offering a total of 32 GPUs with 94GB of HBM memory per unit.
• Storage Nodes: A multi-tiered storage system is implemented with a total of 9 storage nodes, delivering a cumulative capacity of approximately 2.94 petabytes (PB). These nodes include high-speed NVMe drives and large-capacity HDDs to balance performance and storage density.
• Networking & Infrastructure: The infrastructure is supported by Dell PowerSwitch S5248-ON and S5232-ON switches, ensuring high-speed interconnectivity. All nodes are housed in 2U rack-mounted chassis with redundant power supplies for reliability.

REPACSS System Specifications