A computing cluster is a set of machines, several independent computers called nodes, which work together to allow global management and go beyond the limitations of a single computer.
ISIR has just acquired thirty new machines that increase its storage capacity as well as its computing resources. Formerly hosted at LIP6, this cluster was installed by our technical service in one of the data centres of the Information System Department of Sorbonne University, then attached to the ISIR network. A sensitive, delicate operation, carried out without a hitch and in record time!
The cluster responds to the research needs of the teams that use intensive computing with fine parallelism. This is the case in particular for the MLIA team recently integrated into ISIR since 1 January 2022. GPU computing is the basic architecture for work on learning and in particular deep learning, which is at the heart of MLIA’s experiments.
In addition to these local resources, there are infrastructures at the national level that allow scientists to carry out high-performance GPU or HPC calculations, such as the Jean Zay supercomputer, which our teams also use. Our internal cluster, with its rapid access and very flexible use, perfectly complements the national offer.
New computing cluster
Starting a calculation on a single computer is limited in memory and speed. Thanks to this new cluster, launching a calculation simultaneously on several computers allows to reduce considerably the calculation time: a real gain in productivity.
This new cluster is available to all ISIR researchers but also beyond the perimeter of the laboratory. For example, SCAI – Sorbonne Center for Artificial Intelligence – also uses this tool. As a partner, SCAI also invests in this cluster.
Following this redesign of the ISIR clusters, we can use a structure composed of two calculation clusters, GPU (Graphics Processing Unit) or CPU (Central Processing Unit), or a super cluster capable of performing all the calculations and managed locally.
The new ISIR cluster in figures
- 30 new machines that increase storage capacity and computing resources ;
- 73 GPU cards – soon 100 ;
- Current computing power of 1020 TFLOPS in FP32.
Referent contact : Loïc Charlery – IT manager ; loic.charlery(at)sorbonne-universite.fr