Skip to main content Skip to search Skip to main navigation

AI/BigData platform for scale-out data center

Customer

Our customer is a provider of scalable storage solutions for large, data-intensive applications. The company's end customers benefit from the ability to expand hardware resources without interruption as soon as they are needed. End customers include financial service providers, government agencies, and research and healthcare companies.

Project Date: 2023
Project Volume: ca. 1.100.000€

Project Description

The goal of the project was to build an AI/BigData platform based on the NVIDIA Delta A100-80 HGX/SXM4.

Project Realization

Under the premise of "maximum GPU computing performance with simultaneous use of ultra-fast NVIDIA® NVLink® technology", a system configuration based on Dual 64Core AMD EPYC 7002, 2TB (16x 128GB) Samsung DDR4 RDIMM modules and the currently fastest NVIDIA Delta HGX platform incl. 8x A100-80/SXM4 was realized here. Thanks to ultra-performant NVIDIA® NVLink® technology with 600 GB/s NVLink bandwidth per GPU, the eight turnkey systems based on Supermciro A+ Server 4124GO-NART+ in inter-process with a multi-GPU system configuration have a bandwidth that is more than 7 times higher than that of fifth-generation PCIe processes, thus ensuring particularly high scalability and enabling, for example, extremely fast deep learning training.

In addition to high-performance 25GbE SFP+ network interfaces, all systems feature fast enterprise SATA RAID and AllFlash U.2 NVMe storage for caching.