Video System Design
Systems like Leonardo are incredibly unique, there are no duplicates in the world. We can define them as huge prototypes intended to provide services in production 24/7. This is undoubtedly the next challenge to face.
In this video, however, we take a step back and talk about Leonardo’s system design.
Performance and efficiency are key elements that characterize the entire project. Since the initial design phase of the supercomputer, the goal has been to maximize performance but with special attention to sustainability, favoring highly efficient solutions from an energy point of view and thus reducing the environmental impact.
Let’s see how Leonardo’s architecture was designed and how the various components that compose it are integrated.
The architecture is modular and is divided into 4 main partitions with distinct functionalities.
- The front-end and service partition
- The storage partition
- The heterogeneous computing partition
- The conventional computing partition
In detail, the front-end and service partition provides access points to the supercomputer – both command line and graphical interface – and hosts the system management services.
The storage infrastructure must be able to support the incredible processing capabilities achieved by Leonardo. This partition is both capacitive and performance. In fact, the flow of data to and from the computing units must be highly efficient in order not to constitute a performance bottleneck. The storage organization is the result of a co-design between Cineca and the technology suppliers and is based on Cineca’s long experience in providing services to users.
The computing partition called Booster is the performance heart of the Leonardo. This partition is based on heterogeneous compute servers with two different computational units. Each calculation server uses a third generation Intel Xeon CPU and 4 NVIDIA GPU accelerators specifically designed for the Leonardo system.
The Data Centric Compute Partition in turn offers state-of-the-art technology solutions that are only now entering the market. They are general purpose and very high density solutions with as many as 3 servers in a single blade. Removing the direct liquid cooling part – through the so-called cold plates – we find fourth generation Intel Xeon CPUs and DDR5 RAM for very high data transfer speeds and high performance solid state disks.
This video, like the teaser, was also produced by Cineca and made with Blender software, which takes us inside Leonardo itself through the high-detail 3D models provided by Atos.
It was previewed during Leonardo’s inauguration ceremony, and is now available on the official youtube channel.
- Production: Cineca
- Direction: Giovanni Bellavia
- Scientific Manager: Mirko Cestari
- Animation e graphics: AirPixel, Paolo Zuzolo, Daniele De Luca
- 3D Modelling: Paolo Zuzolo, Daniele De Luca
- Editing: Airpixel
- 3D Models: Atos, Cineca
- Music: cleanmindsounds, “Future Background Lounge”