As part of the National AI Research Programme (PNRIA)*, CNRS Computer Science has the mission to build a network of engineers around the Jean Zay ecosystem and the user support missions of IDRIS.
To this end, a team of 9 PNRIA engineers is based at IDRIS, within the AI User Support team to be close to the Jean Zay supercomputer and the teams responsible for its operation. In parallel, smaller teams of 3 people are also established at GRICAD (Grenoble), IRISA (Rennes), IRIT (Toulouse) and LORIA (Nancy).
The PNRIA Deep Learning expert engineers have the mission to support the French academic community in adopting AI technologies to advance their research projects.
This service is intended for researchers from all scientific fields (environment, astrophysics, materials physics, fluid mechanics, chemistry, biology, health, humanities and social sciences, computer science, etc.).
Whether you are discovering the possibilities of applying Deep Learning in your research field, or you are already experienced in training deep neural networks, whether you want to exploit large databases or optimise the performance of your training algorithm, you can benefit from enhanced support for your project by the AI engineers of the programme for a period of up to six months.
A first call for projects was launched on 9 March 2026!
The submission period is open until 30 April 2026 inclusive.
For more information, visit our page dedicated to the PNRIA engineers network.
* The PNRIA is a programme led by CNRS Computer Science on behalf of the CNRS. It is funded under the France 2030 plan and operated by the French National Research Agency (ANR).
]]>Following a long collaborative effort, IDRIS has redesigned its website. The address http://www.idris.fr will remain unchanged. Therefore, do not be surprised when this new site becomes accessible (normally by the end of the month).
The next meeting of the IDRIS User Committee (UC) will take place on Tuesday 31 March 2026 at the IDRIS premises. Do not hesitate to send your requests to the address cu-elus at idris.fr before Tuesday 24 March 2026. More information: http://www.idris.fr/cu.html
For those who use the conda environments provided by IDRIS, a new command "idr_module_search" has been developed to help you find the most suitable Jean Zay module for your needs. It allows you to identify which environments contain certain packages, with or without version specification. For more information: http://www.idris.fr/jean-zay/gpu/jean-zay-gpu-logiciels-ia.html#%F0%9F%94%8D_commande_idr_module_search
We remind you that it is recommended not to exceed a few tens of thousands of elements (files and subdirectories) per directory to ensure the proper functioning of Jean Zay's file systems. If you have a very large number of files to store, adjust the depth of your directory trees accordingly. Following this rule will also ensure the best possible performance, as metadata can be efficiently distributed among the different storage servers.
We remind you that misuse of the "squeue" command can contribute to overloading the Slurm job manager. We invite you to limit the number of calls to "squeue" to a maximum of one or two per minute if you use a script to monitor the status of your jobs. Since Slurm's internal refresh rate is of the same order of magnitude, there is no point in making more frequent requests.
Remember to register now for the IDRIS training sessions scheduled until next summer:
For more information on the IDRIS training catalogue and registration procedures: http://www.idris.fr/formations/catalogue.html.
]]>The Deep Learning for Science Day is back!
Registration and information at https://jdls2026.sciencesconf.org/
IDRIS and the CNRS Fidle National Training Action are organising the fourth edition of the Deep Learning for Science Day (JDLS).
This event will take place on Monday 8 June from 9 am to 5 pm at
CNRS Headquarters - Paris Michel-Ange
3 rue Michel-Ange - 75794 Paris CEDEX 16
This event is open to everyone, free of charge, by registration only.
You do not need to register if you plan to attend the event remotely.
The content of the day will also be broadcast live on the FIDLE YouTube channel for those who cannot travel: https://www.youtube.com/@CNRS-FIDLE.
What is the Deep Learning for Science Day (JDLS)?
The Deep Learning for Science Day (JDLS), organised by IDRIS and the FIDLE training programme, is a national event dedicated to the use of deep learning and artificial intelligence in scientific research.
Open to all (researchers, engineers, PhD students, post-docs), the JDLS offers a series of high-level presentations. Around ten experts present cutting-edge work in Deep Learning applied to various fields such as physics, astrophysics, biology, chemistry, law, humanities and social sciences, etc.
⚠️ Please note that the presentations will be in French. ⚠️
Designed as a place for interdisciplinary exchange and meetings, the JDLS aims to bring scientific communities together around concrete solutions to advance research through deep learning. A significant place is also given to PhD students, who can present their work during dedicated flash presentations and poster sessions.
]]>IDRIS is hiring! Do not hesitate to consult the page on the CNRS Employment Portal.
]]>For the sixth consecutive year, IDRIS is organising, in conjunction with NVIDIA, a GPU HPC and AI Hackathon.
It will take place on May 12th (remotely) and from May 19th to 21st 2026 (at the IDRIS premises). The registration deadline is set for March 10th.
GPU Hackathons allow teams of three or four developers or researchers to improve the performance of their HPC or AI codes under the guidance of one or more mentors, who are experts in GPU programming and come from universities, national laboratories, computing centres, governmental institutions, and manufacturers.
It is not necessary to be a user of IDRIS to participate in this event, so feel free to share this message with your colleagues. You will have the opportunity to access IDRIS's computing resources to test your code on recent GPUs, including NVIDIA H100s on Jean Zay and NVIDIA GB200 NVL72s on the DALIA prototype machine.
For HPC projects, it could be a matter of starting to port a code to GPU using OpenACC, OpenMP, Kokkos, GPU libraries, CUDA, etc., or optimising an existing GPU implementation. For large codebases, it is recommended to prepare a representative mini-application to be more efficient during the event.
For AI projects, this Hackathon offers a unique opportunity to explore fine-tuning, developing, or evaluating state-of-the-art models such as LLMs and GenAI applications. You can optimise your models using advanced techniques such as mixed precision, multi-GPU or multi-node parallelisation (DDP, PP, TP, CP), using basic frameworks such as PyTorch or TensorFlow, or specialised tools such as DeepSpeed, Megatron-LM, Nanotron, or NVIDIA NeMo. Although full pre-training is out of reach within the Hackathon's timeframe, preparing a pre-training code for large-scale models is still possible.
Registration is done via the event page (in English): https://www.openhackathons.org/s/siteevent/a0CUP00003Ahxc22AB/se000449.
Hello,
During yesterday's maintenance (February 17th, 2026), the operating system of the Jean Zay supercomputer has been updated to Red Hat 9.6 (from 9.4 previously). This update aims at keeping a recent software environment available on Jean Zay.
This operating system version change should have a limited impact on your use of the machine. The majority of your executables should continue to work without any action on your part, however in case of issues we recommend that you try to recompile your codes.
The GNU compilers bundled with the operating system have been updated to version 11.5.0. The "gcc/11.4.1" module is now an alias for a new "gcc/11.5.0" module in order to ensure an easy transition.
In case of problems, contact the IDRIS support.
Best regards, The IDRIS user support team
]]>12 March 2026 - 2pm - Online
]]>IDRIS is organizing an online HPC and AI GPU profiling Bootcamp on February 10th, together with NVIDIA and the OpenACC organization.
This bootcamp provides an overview of profiling and code optimization on GPU through a mix of lectures and hands-on sessions which will in particular allow you to familiarize yourself with the NVIDIA profiling tools Nsight Systems and Nsight Compute. There will be two parallel hands-on sessions, one focusing on optimizing high performance computing codes and the other on optimizing artificial intelligence codes.
The hands-on sessions will take place on computing resources provided by NVIDIA. You can attend the bootcamp even if you do not have an IDRIS account.
The bootcamp will also be the occasion to discover the new NVIDIA Grace Blackwell NVL72 architecture which powers the prototype machine DALIA recently installed at IDRIS.
Attendees will have the opportunity to discuss their real-world projects and codes with NVIDIA experts at the end of the event.
Attending this bootcamp is highly recommended for those interested in the GPU hackathon [http://www.idris.fr/eng/annonces/idris-gpu-hackathon-2026-eng.html] which will be held at IDRIS in May.
Please register on https://www.openhackathons.org/s/siteevent/a0CUP00003MjHIf2AN/se000457 (registration deadline: January 27th).
]]>The IDRIS helpdesk will be exceptionally closed on Wednesday 24 December, Friday 26 December and Monday 29 December 2025. Requests sent to assist at idris.fr will only be processed from Tuesday 30 December. The administrative management service for users remains open on these dates, so you can contact gestutil at idris.fr for any request related to the management of your computing account (password reset, addition of a connection IP address, etc.).
The next meeting of the IDRIS User Committee (UC) will take place on Thursday 18 December 2025 at the IDRIS premises. Do not hesitate to send your requests to the address cu-elus at idris.fr before Friday 12 December 2025. More information: http://www.idris.fr/cu.html
Remember to register now for the IDRIS training sessions scheduled until next summer:
For more information on the IDRIS training catalogue and registration procedures: http://www.idris.fr/formations/catalogue.html.
]]>CC-FR Webinar to Discover MINERVA: 4 December 2025 at 2 PM
IDRIS Training
On Thursday 4 December 2025 at 2 PM, the French Competence Centre (CC-FR) invites you to a webinar open to all to discover MINERVA, the European centre for services dedicated to large-scale AI on supercomputers. Mathieu Lobet (CEA / Maison de la Simulation) will present how MINERVA supports researchers, engineers and companies in bringing their AI workflows to HPC infrastructures, optimising their models at scale and leveraging EuroHPC resources for their AI projects. Come and discover how to benefit from MINERVA support and discuss your AI and HPC needs! All information and the registration procedure are available here: https://cc-fr.eu/%C3%A9v%C3%A8nement/accelerez-vos-projets-dia-a-grande-echelle-avec-minerva-le-centre-europeen-dedie-a-lia-sur-supercalculateurs/.
The following IDRIS training sessions are confirmed and will take place before the end of the year:
For more information on the IDRIS training catalogue and registration procedures: http://www.idris.fr/formations/catalogue.html.
]]>Hello,
A maintenance operation on the technical infrastructure of IDRIS will require the centre to be shut down completely on Tuesday 7 October. The Jean Zay machine will be unavailable all day. The return to production is planned for the evening.
User Support will be closed for this day. The website will remain accessible and you will be able to check the machine availability on the usual page: http://www.idris.fr/status.html.
We are sorry for any inconvenience this operation might cause.
Best regards, The IDRIS User Support Team
]]>The next User Committee (UC) meeting of IDRIS will take place on Wednesday 10 September 2025 at the IDRIS premises in Orsay (91). Please send your requests to: cu-elus at idris.fr. More information: http://www.idris.fr/cu.html
The annual maintenance of the IDRIS technical infrastructure requires the shutdown of the centre on Tuesday 7 October 2025. The Jean Zay machine will be unavailable all day and the User Support service will be closed. The website will remain accessible and you can follow the availability of the machines on the usual page: http://www.idris.fr/statut.html.
Since 1 January 2025, IDRIS has joined the European large-scale AI support centre MINERVA: http://www.idris.fr/minerva.html. As part of this, a survey is underway to better understand the expectations and challenges faced by the AI community using HPC resources such as Jean Zay. Your contribution is valuable to help us evolve services, technical support, training programmes, etc. Link to the survey: https://docs.google.com/forms/d/e/1FAIpQLSeWdowBcyc1d9bMsh6BGcolZQj_pECPpoCsF9Mdyf60Fcr68w/viewform?usp=header.
A new partition "compil_H100" is available to users: http://www.idris.fr/jean-zay/cpu/jean-zay-cpu-hw.html#compil_H100. It contains a node equipped with an Intel processor identical to those in the Eviden H100 partition (gpu_p6). It is dedicated to the compilation of codes and libraries that will run on this partition. It has the same access characteristics as the "compil" partition.
To standardise Git behaviour and for security reasons, access to GitHub via the SSH protocol will be completely blocked on all IDRIS machines from 15 September 2025. Until now, some GitHub SSH access points were whitelisted, which resulted in variable behaviour depending on the connections (some allowing access, others not). This situation could be confusing. From now on, only the use of the HTTPS protocol with personal token authentication will be allowed to interact with GitHub (cloning, push, pull, etc.). 👉 To generate a GitHub token and configure your repository in HTTPS: https://docs.github.com/fr/authentication/keeping-your-account-and-data-secure/creating-a-personal-access-token Please anticipate this change by adapting your Git configurations before 15 September 2025.
During the summer, we reached saturation of the shared DSDIR space, which contains the public datasets and models that we download on request. To free up space, we have moved the CommonCrawl and RedPajama-V2 text corpora to an annex storage space on rotating disks. From now on, any text corpus will also be stored on rotating disks. The access paths to these datasets remain unchanged: $DSDIR/<nom_du_dataset>. According to the tests we have carried out, this should not impact the performance of your training. Please contact support (assist at idris.fr) if this is the case.
Remember to register now for the IDRIS training sessions scheduled until the end of the year:
For more information on the IDRIS training catalogue and registration procedures: http://www.idris.fr/formations/catalogue.html.
]]>Dear Jean Zay users,
On January 1, 2025, IDRIS joined the European large-scale AI Support Centre MINERVA: http://www.idris.fr/minerva.html
In this context, we invite you to take part in a survey on the needs of AI users, in order to better understand the expectations and challenges faced by the AI community using HPC resources such as Jean Zay.
Your feedback will help improve services, technical support, training programmes, and more.
The survey takes about 10 minutes to complete and is aimed at individual users, teams, and organisations alike.
Survey link: https://docs.google.com/forms/d/e/1FAIpQLSeWdowBcyc1d9bMsh6BGcolZQj_pECPpoCsF9Mdyf60Fcr68w/viewform?usp=header
Your contribution is valuable to help us improve the accessibility and quality of AI services on Jean Zay and other HPC platforms.
Thank you in advance for your participation!
Best regards, IDRIS – MINERVA Project Team
]]>Hello everyone,
We are pleased to invite you to the
3rd Deep Learning for Science Day (JDLS 2025)
which will take place on Thursday, 5 June 2025 at the CNRS headquarters in Paris.
This event, free and open to all, is organised by IDRIS and the CNRS Fidle National Training Action, with the support of the CNRS, the Université Grenoble Alpes (UGA), the University of Montpellier, via the ISDM and the MIAI AI Cluster.
The objective of this day is to promote the sharing of experiences and exchanges around the applications of Deep Learning in science :-)
The entire day will be broadcast live on YouTube on the FIDLE channel: https://www.youtube.com/@CNRS-FIDLE.
Participation in this day is free and open to all, but registration is mandatory if you wish to attend in person.
It is also possible, especially if you are a PhD student, to contribute to the event by presenting a poster and/or a 3-minute flash presentation. Prizes will be awarded for the best posters/presentations!
Registration, poster submission and programme: https://jdls-2025.sciencesconf.org
Feel free to share/forward this announcement :-)
Looking forward to seeing you there!
The JDLS Organising Committee
Why participate?
Attend talks by recognised specialists on AI for science and science for AI,
Discover concrete cases of Deep Learning applications in many disciplines: chemistry, physics, language processing, numerical simulation...
Exchange with the scientific and AI community during dedicated discussion times and poster presentations,
Meet the actors and actresses involved in the appropriation of Artificial Intelligence tools in the various fields of research.
Confirmed speakers
We are pleased to welcome:
Prof. Jean-Philip Piquemal – Sorbonne University Director of the Theoretical Chemistry Laboratory (LCT), Honorary Member of the Institut Universitaire de France
Steeven Janny – LAGEPP/LIRIS, INSA de Lyon Former student of the École Normale Supérieure, his work focuses on the use of deep learning for the simulation of physical phenomena, particularly fluid mechanics.
Prof. Karën Fort – Université de Lorraine, LORIA / INRIA Talk on biases and ethics of LLM (Large Language Models) Member of the SEMAGRAMME project team (Semantic Analysis of Language) Specialist in automatic language processing and AI ethics.
The programme is being finalised. Other speakers will be announced soon on the conference website.
It is possible to contribute, especially if you are a PhD student, to the event in two formats:
Presentation of a scientific poster displayed throughout the day
"My project in 3 minutes" session: a flash presentation of your research project to the public.
Prizes will be awarded for the best posters/presentations :-) Free registration, poster submission, and programme: https://jdls-2025.sciencesconf.org/
You are receiving this email because you are a user of the IDRIS machines.
You can view this issue of the IDRIS Flash Info and previous issues on the IDRIS website: http://www.idris.fr/flash-info.html
]]>As previously announced, maintenance of the IDRIS technical infrastructure requires the centre to be shut down tomorrow, Tuesday 8 April. The Jean Zay machine will be unavailable all day.
The User Support service will be closed. The website will remain accessible and you can follow the availability of the machines on the usual page: http://www.idris.fr/statut.html.
IDRIS and the CNRS Fidle National Training Action are organising the third Deep Learning for Science Day (JDLS) on Thursday 5 June from 9 am to 5:30 pm at the CNRS headquarters, Paris Michel Ange. Speakers from various scientific fields will present their use of artificial intelligence methods throughout the day. It is also possible to participate in the day by taking part in the "My project in 3 minutes" session and/or the poster session that will follow. Information and registration at https://jdls-2025.sciencesconf.org.
Remember to register now for the IDRIS training sessions scheduled for the coming months:
For more information on the IDRIS training catalogue and registration procedures: http://www.idris.fr/formations/catalogue.html.
You are receiving this newsletter because you are a user of the IDRIS machines.
You can view this issue of the IDRIS Flash Info and previous issues on the IDRIS website: http://www.idris.fr/flash-info.html
]]>Hello,
During tomorrow's maintenance (April 1st, 2025), the operating system of the Jean Zay supercomputer will be updated to Red Hat 9.4 (from 9.2 currently). This update aims to provide a recent software environment on Jean Zay.
This operating system version change should have a limited impact on your use of the machine. The majority of your executables should continue to work without any action on your part; in case of issues, we recommend that you try to recompile your codes.
The GNU compilers bundled with the operating system will be updated to version 11.4.1. The "gcc/11.3.1" module will become an alias for a new "gcc/11.4.1" module to ensure an easy transition.
In case of problems, contact the IDRIS Support.
Best regards, The IDRIS user support team
]]>We remind you that the deadline for registration for our GPU HPC and AI hackathon is set for 4 March. As a reminder, this event, organised jointly with NVIDIA, will take place on 13 May (remotely) and on 20, 21 and 22 May (at the IDRIS premises).
Registration is via the event page (in English): https://www.openhackathons.org/s/siteevent/a0CUP00000sO9LV2A0/se000364.
IDRIS support invites you on Friday morning 28/02 at 10 am for a live broadcast of its "Panoram'IA" show. This show will feature our guest Jérôme Louradour, who will share his experience with the pretraining of Lucie on our Jean Zay supercomputer. The live broadcast and replays are available on our YouTube channel "Un oeil sur l'IDRIS": https://www.youtube.com/@idriscnrs.
As announced at the last User Committee meeting, the CEEMS energy measurement tool is now available on Jean Zay: see the page http://www.idris.fr/jean-zay/jean-zay-doc-energie.html for more information.
Remember to register now for the IDRIS training sessions scheduled for the coming months:
For more information on the IDRIS training catalogue and registration procedures: http://www.idris.fr/formations/catalogue.html.
]]>[Version française ci-dessous]
IDRIS is hosting its fifth IDRIS Open Hackathon (https://www.openhackathons.org/s/siteevent/a0CUP00000sO9LV2A0/se000364) from May 13 to May 22, 2025 and we are actively seeking applications (deadline: March 4th).
If you are not familiar, this hands-on coding event helps researchers and engineers to optimise, accelerate, and scale their HPC and AI applications on Jean Zay by pairing them with expert programming mentors whose guidance can take your HPC and AI projects to the next level.
This year NVIDIA invites you to an informational webinar “Getting Hackathon Ready” focused on AI applications, with the participation of IDRIS:
Join us on February 18th, 11AM CET to learn more about how to participate in this unique opportunity.
By attending the webinar, you will:
Space is limited, sign up now: https://bit.ly/IdrisGHR.
For a glimpse of the projects that benefited from attending last year’s event, read the 2024 IDRIS Open Hackathon recap: https://www.openhackathons.org/s/article/IDRIS-Contributing-to-the-Excellence-of-Scientific-Research.
We hope you will join us to learn more and receive personalised expert support.
Should you have any questions, please contact events@openhackathons.org.
IDRIS is organising its 5th Hackathon (https://www.openhackathons.org/s/siteevent/a0CUP00000sO9LV2A0/se000364) from 13 to 22 May 2025 and we are looking for interested teams to participate (registration deadline: 4 March).
What is a hackathon? It is an event dedicated to scientific code developers aimed at helping researchers and engineers to optimise, accelerate, and scale their HPC or AI applications on Jean Zay. To do this, you will be supported by programming expert mentors whose advice will boost your HPC or AI projects.
This year NVIDIA invites you to an informational webinar "Getting Hackathon Ready" focused on AI applications, with the participation of IDRIS:
Join us on 18 February at 11am to discover how to make the most of this opportunity.
By participating in the seminar, you will be able to:
The number of places is limited, register now: https://bit.ly/IdrisGHR.
For an overview of the projects that benefited from last year's hackathon, you can read the IDRIS Hackathon 2024 recap: https://www.openhackathons.org/s/article/IDRIS-Contributing-to-the-Excellence-of-Scientific-Research.
We hope you will participate to learn more and receive personalised support.
If you have any questions, please do not hesitate to contact events@openhackathons.org (in English).
]]>Migration of the DSDIR space to Lustre
[English version below]
Hello,
During the maintenance on January 21, 2025, the DSDIR environment variable was modified and now points to a new Lustre storage space: "/lustre/fsmisc/dataset".
All databases and models downloaded to DSDIR since the beginning of Jean Zay have been migrated to this new space. This new space is accessible from all Jean Zay nodes, including the nodes of the new H100 extension.
This new space will now be used for downloading public datasets or models. The old disk spaces "/gpfsdsdir/dataset" and "/gpfsdswork/dataset" are now deprecated and will no longer be accessible as of the end of January.
If you are already using the DSDIR environment variable in your scripts, the data migration will have no impact on your runs. However, if you are using hardcoded paths, you will need to update them. We invite you to take this opportunity to switch to systematic use of the DSDIR variable, as in the following script for example:
''' import os from datasets import load_dataset dataset = load_dataset(os.environ['DSDIR']+"/HuggingFace/wikimedia/structured-wikipedia") '''
As a reminder, DSDIR is a dedicated disk space for storing public databases or models needed for the use of Artificial Intelligence tools. It is visible to all Jean Zay users. For more information: http://www.idris.fr/eng/jean-zay/cpu/jean-zay-cpu-calculateurs-disques-eng.html#the_dsdir
Please do not hesitate to contact the support team at assist@idris.fr if you encounter any issue.
Best regards, The IDRIS support team
]]>For the fifth consecutive year, IDRIS is organising, together with NVIDIA, a GPU HPC and AI Hackathon.
It will take place on 13 May (remotely) and from 20 to 22 May 2025 (on-site at IDRIS). The registration deadline is 4 March.
GPU Hackathons allow teams of three or four developers or researchers to improve the performance of their HPC or AI codes under the guidance of one or more mentors, who are experts in GPU programming and come from universities, national laboratories, computing centres, government institutions and manufacturers.
You do not need to be an IDRIS user to participate in this event, so feel free to share this message with your colleagues. You will have access to IDRIS's computing resources, allowing you to test your code on recent GPUs, including NVIDIA H100s.
For HPC projects, it could be an opportunity to start porting a code to GPU using OpenACC, OpenMP, Kokkos, GPU libraries, CUDA, etc., or to optimise an existing GPU implementation. If your codebase is large, it is recommended to prepare a representative mini-application to be more efficient during the event.
For AI projects, this hackathon offers a unique opportunity to explore fine-tuning, developing or evaluating state-of-the-art models such as LLMs and GenAI applications. You will be able to optimise your models using advanced techniques such as mixed precision, multi-GPU or multi-node parallelisation (DDP, PP, TP, CP), using standard frameworks such as PyTorch or TensorFlow, or specialised tools such as DeepSpeed, Megatron-LM, Nanotron or NVIDIA NeMo. Although full pre-training is out of reach within the hackathon's time frame, preparing a pre-training code for large-scale models is still possible.
Registration is via the event page (in English): https://www.openhackathons.org/s/siteevent/a0CUP00000sO9LV2A0/se000364.
]]>