A New Scalable AI Infrastructure Based on NVIDIA DGX BasePOD for the University of Padua
The University of Padua is one of the oldest and most renowned universities in Europe, internationally recognised for the quality of its research, teaching and technological innovation.
Within the University, the IT and Telematic Services Area (ASIT) plays a strategic role in managing the university’s digital infrastructure and ICT services.
CUSTOMER NEEDS
The University of Padua has embarked on a programme to enhance its computing capabilities, with the aim of providing its research staff with an advanced infrastructure capable of supporting the high computational loads generated by AI algorithms and complex, simultaneous processing tasks from different research groups.
The University was looking for a high-performance solution designed to operate in multi-user environments and scale easily, ensuring maximum efficiency, flexibility and continuity of service over time.
A new AI infrastructure based on the NVIDIA DGX BasePOD reference architecture, a solution designed natively for large-scale artificial intelligence.
NVIDIA DGX BasePOD is designed to be an AI-first infrastructure, optimised in every component – from hardware to networking to software stack – to deliver maximum performance, simplicity of management and scalability over time.
Unlike alternative assembled solutions, NVIDIA DGX BasePOD is an integrated, validated platform designed to bring artificial intelligence into production quickly, with repeatable and predictable results.
A cluster consisting of the following has been set up at the University of Padua:
- Compute Nodes: 5 NVIDIA DGX H100 systems, the fourth generation of the world’s leading AI-engineered system, equipped with 8 NVIDIA H100 GPUs for a total of 640 GB of HBM3 memory with an aggregate memory bandwidth of 24 TB/s, and an architecture engineered to deliver 32 petaFLOPS with FP8 precision per system;
- Compute Fabric: NVIDIA Quantum QM9700 NDR 400 Gbps for the 400 Gb/s InfiniBand fabric, enabling ultra-fast communication between nodes without bottlenecks;
- In-band management network: NVIDIA SN4600 and SN2201 for Ethernet networking and out-of-band (OOB) management, essential for centralised and secure data access;
- 5 x86 management nodes, configured for high reliability, to orchestrate workloads via NVIDIA Base Command, Kubernetes and Slurm.
In addition to hardware, the solution also includes the entire NVIDIA software stack:
NVIDIA Base Command: for centralised monitoring, management, and orchestration of one to hundreds of nodes.
NVIDIA NGC™: for access to pre-trained models, optimised containers, and AI libraries.
NVIDIA AI Enterprise: to accelerate the deployment of AI solutions with certified tools and enterprise support.
A state-of-the-art AI platform
Discover all the advantages of the new NVIDIA DGX BasePOD infrastructure designed for the IT and Telematic Services Area (ASIT) of the University of Padua.
The NVIDIA DGX BasePOD infrastructure at our University is at the heart of UPSCALE (University of Padova Super Computing Architecture for Leading-Edge research), an important asset available to researchers in all disciplines where high-performance computing has become essential, from engineering to physics, life sciences to economics, astronomy to pharmaceuticals. […] The readiness of NVIDIA and beanTech to listen to the University’s needs and translate them into technical solutions has been fundamental in enabling us to achieve this result, as has the active cooperation with Infocamere, which hosts the computing infrastructure in its data centre.
This project represents a significant step towards the future of research and innovation, positioning the University of Padua at the forefront of high-performance computing at the national level and opening new frontiers for interdisciplinary research.