Grid5000:Home

Grid'5000 is a precursor infrastructure of SLICES-RI, Scientific Large Scale Infrastructure for Computing/Communication Experimental Studies. Content on this website is partly outdated. Technical information remains relevant.

Grid'5000 Grid'5000 is a large-scale and flexible testbed for experiment-driven research in all areas of computer science, with a focus on parallel and distributed computing, including Cloud, HPC, Big Data and AI. Key features: provides access to a large amount of resources: 15000 cores, 800 compute-nodes grouped in homogeneous clusters, and featuring various technologies: PMEM, GPU, SSD, NVMe, 10G and 25G Ethernet, Infiniband, Omni-Path highly reconfigurable and controllable: researchers can experiment with a fully customized software stack thanks to bare-metal deployment features, and can isolate their experiment at the networking layer advanced monitoring and measurement features for traces collection of networking and power consumption, providing a deep understanding of experiments designed to support Open Science and reproducible research, with full traceability of infrastructure and software changes on the testbed a vibrant community of 500+ users supported by a solid technical team Read more about our teams, our publications, and the usage policy of the testbed. Then get an account, and learn how to use the testbed with our Getting Started tutorial and the rest of our Users portal. Published documents and presentations: Presentation of Grid'5000 (April 2019) Report from the Grid'5000 Science Advisory Board (2018) Older documents: Slides from Frederic Desprez's keynote at IEEE CLUSTER 2014 Report from the Grid'5000 Science Advisory Board (2014) Grid'5000 is supported by a scientific interest group (GIS) hosted by Inria and including CNRS, RENATER and several Universities as well as other organizations. Inria has been supporting Grid'5000 through ADT ALADDIN-G5K (2007-2013), ADT LAPLACE (2014-2016), and IPL HEMERA (2010-2014).

Current status (at 2025-06-24 18:12): 3 current events, 6 planned (details)

Random pick of publications

Five random publications that benefited from Grid'5000 (at least 2774 overall):

• Mathis Valli, Alexandru Costan, Cédric Tedeschi, Loïc Cudennec. Towards Efficient Learning on the Computing Continuum: Advancing Dynamic Adaptation of Federated Learning. FlexScience 2024 - 14th Workshop on AI and Scientific Computing at Scale using Flexible Computing Infrastructures, Jun 2024, Pisa, Italy. pp.42-49, 10.1145/3659995.3660042. hal-04698619v2 view on HAL pdf
• Khaled Arsalane, Guillaume Pierre, Shadi Ibrahim. Toward Stream Processing Elasticity in Realistic Geo-Distributed Environments. IC2E 2024 - 12th IEEE International Conference on Cloud Engineering, IEEE, Sep 2024, Paphos, Cyprus. pp.1-9. hal-04655408v2 view on HAL pdf
• Duy Van Ngo, Yannick Parmentier. Towards Sentence-level Text Readability Assessment for French. Second Workshop on Text Simplification, Accessibility and Readability (TSAR@RANLP2023), Sep 2023, Varna, Bulgaria. hal-04192063 view on HAL pdf
• Thomas Firmin, Pierre Boulet, El-Ghazali Talbi. Asynchronous Multi-fidelity Hyperparameter Optimization Of Spiking Neural Networks. International Conference on Neuromorphic Systems (ICONS 2024), Jul 2024, Washington, United States. hal-04781629 view on HAL pdf
• Reda Khoufache, Anisse Belhadj, Mustapha Lebbah, Hanene Azzag. Distributed MCMC Inference for Bayesian Non-parametric Latent Block Model. 28th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2024, May 2024, Taipei, Taiwan. pp.271-283, 10.1007/978-981-97-2242-6_22. hal-04623748 view on HAL pdf

Latest news

Cluster "hydra" is now in the default queue in Lyon

We are pleased to announce that the hydra[1] cluster of Lyon is now available in the default queue.

As a reminder, Hydra is a cluster composed of 4 NVIDIA Grace-Hopper servers[2].

Each node features:

1 Nvidia Grace ARM64 CPU with 72 cores (Neoverse-V2)

1 Nvidia Hopper GPU

512GB LPDDR5 memory

96GB HBM memory

1x1To SSD NVME + 1x1.92To SCSI disk

Due to its bleeding-edge hardware, the usual Grid'5000 environments are not supported by default for this cluster.

(Hydra requires system environments featuring a Linux kernel >= 6.6). The default system on the hydra nodes is based on Debian 11, but **does not provide functional GPUs**. However, users may deploy the ubuntugh2404-arm64-big environment, which is similar to the official Nvidia image provided for this machine and provides GPU support.

To submit a job on this cluster, the following command may be used:

oarsub -t exotic -p hydra

This cluster is funded by INRIA and by Laboratoire de l'Informatique du Parallélisme with ENS Lyon support.

[1] Hydra is the largest of the modern constellations according to Wikipedia: https://en.wikipedia.org/wiki/Hydra_(constellation)

[2] https://developer.nvidia.com/blog/nvidia-grace-hopper-superchip-architecture-in-depth/

-- Grid'5000 Team 16:42, 12 June 2025 (CEST)

Cluster "estats" (Jetson nodes in Toulouse) is now kavlan capable

The network topology of the estats Jetson nodes can now be configured, just like for other clusters.

More info in the Network reconfiguration tutorial.

-- Grid'5000 Team 18:25, 21 May 2025 (CEST)

Cluster "chirop" is now in the default queue of Lille with energy monitoring.

Dear users,

We are pleased to announce that the Chirop[1] cluster of Lille is now available in the default queue.

This cluster consists of 5 HPE DL360 Gen10+ nodes with:

2 CPU Intel Xeon Platinum 8358 (32 cores per CPU)

512 GiB memory

1*1.92TB SSD NVME + 2*3.84TB SSD

2*25 Gbps Ethernet interface

Energy monitoring[2] is also available for this cluster[3], provided by newly installed Wattmetres (similar to those already available at Lyon).

This cluster was funded by CPER CornelIA.

[1] https://www.grid5000.fr/w/Lille:Hardware#chirop

[2] https://www.grid5000.fr/w/Energy_consumption_monitoring_tutorial [3] https://www.grid5000.fr/w/Monitoring_Using_Kwollect#Metrics_available_in_Grid.275000

-- Grid'5000 Team 16:25, 05 May 2025 (CEST)

Change of default queue based on platform

Until now, Abaca (production) users had to specify `-q production` when reserving Abaca resources with OAR.

This is no longer necessary as your default queue is now automatically selected based on the platform your default group is associated to, as shown at https://api.grid5000.fr/explorer/selector/ and in the message displayed when connecting to a frontend.

For SLICES-FR users, there is no change since the correct queue was already selected by default.

Additionally, the "production" queue has been renamed to "abaca", although "production" will continue to work for the foreseeable future.

Please note one case where this change may affect your workflow:

When an Abaca user reserves a resource from SLICES-FR (a non-production resource), they must explicitly specify they want to use the SLICES-FR queue, which is called "default", by adding `-q default` the OAR command.

-- Abaca Grid'5000 Team 10:10, 31 March 2025 (CEST)

---

Read more news

Grid'5000 sites

Grenoble Lille Luxembourg Louvain

Lyon Nancy Nantes Rennes

Sophia-Antipolis Strasbourg Toulouse

Current funding

INRIA	CNRS
Universities IMT Atlantique Université Grenoble Alpes, Grenoble INP Université Rennes 1, Rennes Institut National Polytechnique de Toulouse / INSA / FERIA / Université Paul Sabatier, Toulouse Université Bordeaux 1, Bordeaux Université Lille 1, Lille École Normale Supérieure, Lyon	Regional councils Aquitaine Auvergne-Rhône-Alpes Bretagne Champagne-Ardenne Provence Alpes Côte d'Azur Hauts de France Lorraine