CISM: Lemaitre4 cluster extension, with computing capacity multiplied by five!

CISM has recently acquired five new compute servers that will be integrated into the existing Lemaitre4 cluster.

Each of these servers is equipped with four AMD MI300A APUs (Accelerated Processing Units), an architecture that combines CPU cores, graphics processing units (GPUs), and high-bandwidth memory (HBM) on a single chip.

Unlike traditional systems where CPUs and GPUs are separate and communicate through external buses, APUs feature unified memory shared between these two types of processors. This design reduces data transfers between components, lowers latency, and allows for more efficient use of computing resources, particularly for hybrid applications combining sequential and massively parallel computations.

The addition of these five MI300A-equipped nodes will significantly increase the capacity of the Lemaitre4 cluster.

Overall, the available computing power will be multiplied by approximately a factor of five compared to the previous configuration.

This extension aims to meet the growing demand for resources required by compute-intensive workloads, particularly in numerical simulation and machine learning, while also enabling researchers to become familiar with a technology that is expected to eventually replace the current CPU+GPU combination in large international computing clusters.

Lemaitre4 is accessible to all researchers belonging to a university that is a member of the Consortium of High-Performance Computing Equipment in the Wallonia-Brussels Federation (CÉCI).

This new hardware was funded by the FNRS-F.R.S. as part of a grant "Grand Équipement".