Skip to main content
⚠ INFORMATION
This page was translated by an AI (LLM) with a cursory human check and is awaiting full review.
IDRIS's new website: Your opinion matters!

You want to give your opinion, report an error, or suggest an improvement? We are listening! Share it via this short anonymous questionnaire which will only take you a few seconds to complete.
Note that every suggestion is considered but there is no guarantee it will be implemented.

Frequently Asked Questions (FAQ)

Remember to download the IDRIS Memento

IDRIS's new website​

IDRIS's new website: I want to give my opinion, report an error or suggest an improvement.

Your opinion matters! Feel free to share it via this short anonymous questionnaire which will only take you a few seconds to complete: Survey, we are listening to you. Note that every suggestion is considered but there is no guarantee it will be implemented.

Why this new website?

This new website allows you to access all our documentation, information, and news, while offering you a more modern, fluid, and intuitive experience, with a powerful search bar to find any information in an instant, and a display adapted to all your devices (computer, tablet, and smartphone).

The information I am looking for is no longer in the same place as before, what should I do?

Our new website features a highly efficient search tool: Type a title or keywords, and you will easily find what you are looking for.

Are all the documentation and information from the old site present?

Yes, the vast majority of the content from the old site is now available here and is regularly updated. Since this new site was launched recently, a few pages are still under construction; we are actively working on putting them online.

Is the old site still accessible?

Yes, a frozen version dating from 03/24/2026 of the old website is still accessible.

Does the change of website have an impact on resource access and account opening procedures?

No, the change of the IDRIS website has no impact on your account or on the procedures for accessing resources; it simply allows you to access all our documentation, information, and news, all on a more modern, fluid, and intuitive platform.

Connection problems​

How to access Jean Zay?

Each IDRIS user has a unique login for all the projects they participate in. This account is personal and must not be shared. This login is associated with a password that follows certain security rules. Before logging in, we advise you to consult the password management page. You can only log into Jean Zay from a machine whose IP address is registered in our filters. If this is not the case, consult the machine declaration procedure available on our website.

Interactive access to Jean Zay is possible via the SSH protocol on the machine's frontend nodes or via ourJupyterHub instance (note, a first connection via ssh is necessary to define your password).

For more information, you can consult the detailed hardware description of the cluster.

What are the available shells?

The Bourne Again shell (bash) is the only command interpreter supported as a login shell on IDRIS machines: IDRIS does not guarantee that the default user environment is correctly defined with other interpreters (ksh, tcsh, csh).

What is the initial password?

The initial password is the one you must use during your first connection or after a password reset.

It is the result of the concatenation of two parts:

  1. a first part randomly generated by IDRIS;
  2. a second part (8 characters) filled in by yourself in the account creation request form available on the DARI portal.

You will find more information regarding your password management in our documentation.

How to change the initial password upon the first connection?

You are required to change your password during your first connection: a password change procedure is automatically executed as soon as you log into the machine.

You will therefore be led to follow the procedure detailed on the web page accessible by clicking here.

I forgot my password. What should I do?

  • If you remember the user part (8 characters) of the initial password defined on the DARI portal, send an email togestutil@idris.fr to request the reset of the random part (IDRIS part) of the password.

  • If you have lost the user part (8 characters) of the initial password defined on the DARI portal, you must fill out the box entitled "Réinitialisation de la partie utilisateur du mot de passe initial" (Resetting the user part of the initial password) in the Account Management Form (FGC) and send it by email to gestutil@idris.fr.

New user: I received my Jean Zay login, why can I still not connect to the machine?

There are three likely reasons that may explain the situation.

  1. You received the account creation email very recently and you are trying to connect from a declared machine. But, as indicated in the email, there is one step left in the procedure before you can connect: we must attach the IP addresses you declared to your login.
    Usually, this step takes one working day. You should be able to connect the next day starting at 5:00 PM.

  2. You are not connecting from one of the machines whose IP address you have declared. You must then establish your connection via a machine with a declared IP address or request the addition of the IP address of the machine from which you are trying to connect using the FGC form.
    Warning: we only accept fixed institutional IP addresses.

  3. On average once a week, we perform maintenance operations announced on the IDRIS website: on the site's homepage, click on the drop-down menu entitled "Espace utilisateurs" (User space) then the "Disponibilité des machines" (Machine availability) section. During these maintenances, it is generally not possible to connect.

Apart from these reasons, do not hesitate to contact support (assist@idris.fr) by including the output of the ssh -vvv ... command.

Existing user: why can I not connect to the machine?

There are several likely reasons that may explain the situation.

  1. You are not connecting from a machine whose IP address you have declared. You must then establish your connection via a machine with a declared IP address or request the addition of the IP address of the machine from which you are trying to connect using the FGC form.
    Warning: we only accept fixed institutional IP addresses.

  2. On average once a week, we perform maintenance operations announced on the IDRIS website: on the site's homepage, click on the drop-down menu entitled "Espace utilisateurs" (User space) then the "Disponibilité des machines" (Machine availability) section. During these maintenances, it is generally not possible to connect.

Apart from these reasons, do not hesitate to contact support (assist@idris.fr) by including the output of the ssh -vvv ... command.

How do I add a machine from which I will connect to Jean Zay?

When requesting account creation, you must provide the list of machines with which you will connect to the IDRIS supercomputer (IP address and name). This declaration is made from the eDARI portal.

To add a new machine after account creation, you must fill out the box entitled "Ajout, suppression d’adresses IP de connexion" (Addition, deletion of connection IP addresses) of the account management form (FGC) and send it to gestutil@idris.fr. Note that this procedure requires the signatures of the user and the security manager.

Warning: we only accept fixed institutional IP addresses.

Can I connect from a machine with a dynamic/variable IP address?

This is the case, for example, of a personal PC connected via an ADSL link, whose IP number is not fixed because it is automatically chosen by the Internet Service Provider at the time of connection.

The IP number cannot then be integrated into the IDRIS filters for technical reasons (since the number changes with each connection) and especially for computer security reasons.

Therefore, you cannot connect directly to Jean Zay if your machine has a variable IP number.

To work around this problem, you have the solutions provided for teleworking and missions.

Can I use SSH keys for my connections with IDRIS?

SSH connections using an SSH key pair (private key / public key) are possible on Jean Zay.

WARNING: We are considering reinforcing our security policy regarding access to the Jean Zay machine. Therefore, we ask you to test, starting now, the use of certificates for your SSH connections instead of the usual SSH key pairs (private key / public key) by following the detailed procedures here.

How to use certificates for my SSH connections with IDRIS?

To reinforce our security policy regarding access to the Jean Zay machine, we ask you to test the use of certificates for your SSH connections instead of the usual SSH key pairs (private key / public key). The creation and use of certificates are done by following the detailed procedures here.

During the test phase and at least until the end of February 2023, connections via classic SSH keys will remain possible. Please report any problems you might encounter with the use of certificates.

Why can't I use Git on Jean Zay?

On Jean Zay, it is strongly recommended to access a Git repository via the HTTP(S) protocol. For example:

git clone https://github.com/myrepo/myrepo.git

Warning, this may require generating a specific authentication token, which is the case for example for GitHub.

If your Git repository is accessible only via the SSH protocol, you will have to declare the IP address of the Git server via the account management form. The part to complete is the second table (concerning outgoing flows) of the section "Ajout, suppression d'adresses IP de connexion" (Addition, deletion of connection IP addresses), at the bottom of page 2. The form must be returned signed to gestutil@idris.fr.

Warning: When access via the HTTP(S) protocol is possible (which is for example the case for Github.com and Gitlab.com), we will not accept the use of SSH.

Project management​

How to request computing hours at IDRIS?

GENCI offers several possibilities for making requests for computing hour allocations depending on the number of hours requested.

More information is available on our page detailing allocation requests.

How to add a user to a project?

If the user does not have an IDRIS account, they must follow the procedure documented on the Account management page to request the opening of an account on the compute machine for the project.

If the user already has an IDRIS account, they must follow the procedure for linking their account to the relevant project.

How to know the number of computing hours consumed per project?

You just need to use the IDRIS command idracct to know the hours consumed by each project collaborator as well as the percentage of the hours allocation and the total consumed.

Note that the information returned by this command is only updated once a day (as indicated on the first line of the command output).

If you have several projects at IDRIS, this command will give you the CPU and/or GPU consumptions of all the projects your login is linked to.

What to do when I am about to run out of computing hours?

It is possible to request additional hours:

  • either a so-called "au fil de l'eau" (continuous) request which can be made as long as your Dynamic Access (AD) or AI project has an ongoing hour allocation.
  • or a mid-term request for your Regular Access allocation in the case of an AR (Regular Access) to obtain additional hours for a period of six months.

These requests, which must be justified, are to be made via the eDARI portal as indicated on our web page detailing hour allocation requests.

Multi-project accounts​

How to know which project a Unix group corresponds to when attached to multiple projects?

To identify the Unix group corresponding to the project you are interested in (especially for counting computing hours), you must combine the information returned by the idracct and idrproj commands:

idracct command

The idracct command lists all the projects you are attached to. Each project has an IDRIS internal project number (different from the DARI number).

For example, if you are Jean Petit and you are attached to two projects, you will get an output like:

$ idracct

################################################################################

                           PROJET 123456 SUR JEAN-ZAY
################################################################################

================================================================================
 CPU (du 01-11-2019 au 31-10-2020)                   Allocation : 1000000 h.cpu
================================================================================

  Compte   Proprietaire                    Consommation Nb travaux Remboursement
  -------- ------------------------------ ------------- ---------- -------------
  login00  PETIT_Jean                          30000.00       1500             -
  login01  DURAND_Marie                        20000.00       1000             -
  -------- ------------------------------ ------------- ---------- -------------
                                   Totaux      50000.00       2500          0.00
                              Utilisation          5.00           


################################################################################

                           PROJET 654321 SUR JEAN-ZAY
################################################################################

================================================================================
 V100 (du 01-12-2020 au 31-11-2022)                    Allocation : 20000 h.gpu
================================================================================

  Compte   Proprietaire                    Consommation Nb travaux Remboursement
  -------- ------------------------------ ------------- ---------- -------------
  login00  PETIT_Jean                           3000.00        150             -
  login02  DUBOIS_Pascal                        2000.00        100             -
  -------- ------------------------------ ------------- ---------- -------------
                                   Totaux       5000.00        250          0.00
                              Utilisation         25.00

With this type of information, you should be able to identify the project that interests you based on the associated members and the corresponding allocation.

In the example above, the project for which you received 1,000,000 CPU hours in collaboration with Marie Durand is the one whose internal IDRIS number is 123456.

This information must then be combined with the output of the idrproj command, as explained below.

idrproj command

The Unix groups to which you belong are listed and associated with the IDRIS internal project numbers by the idrproj command. For example:

$ idrproj

Available projects:
-------------------
  abc (123456) [default][active]
  def (654321)

The Unix groups are indicated in the first column (here abc and def) and the corresponding IDRIS internal project numbers appear in parentheses.

In this example, project 123456 therefore corresponds to the Unix group abc.

How can I change the project of my current session?

To change the active project in your session you must execute the command:

eval $(idrenv -d grp1)

Where grp1 is the three-letter code (trigram) linked to your project.

More information on the multi-project user guide.

Account management​

How can I get a computing account?

There is no automatic or implicit account creation.
Each user must therefore explicitly request the creation of a user account to be able to work within a project. This request must be accompanied by a prior request to be linked to a project requiring the project manager's agreement.
The user account is then unique and allows collaborating on several projects by simply asking to be linked to the various projects.

For more information, please consult our documentation regarding account management at IDRIS.

You can also consult our documentation explaining the computing account declaration at IDRIS via the eDARI portal.

How to modify my personal data?

The modification of your personal data is done via the Extranet web interface.

  • For those who do not have an Extranet password or who have lost it, the access conditions are described on this page.
  • For those who have their password, click on Extranet, log in with your credentials then => Votre compte => Vos données => Coordonnées (Your account => Your data => Contact details).

The only data that can be modified online are:

  • email;
  • phone;
  • fax.

The modification of your postal address is done by sending us the account management form FGC via the section "Changement de structure de recherche ou de société sans changer de compte" (Change of research structure or company without changing account).

Data management​

What disk spaces are available on Jean Zay?

For each project, 5 distinct disk spaces are available on Jean Zay: HOME, WORK, SCRATCH/JOBSCRATCH, STORE and DSDIR.

You will find explanations concerning these spaces on our website, on the Disk spaces page.
Warning: HOME, WORK and STORE are subject to quotas!

If your login is linked to several projects, the IDRIS command idrenv will display all the environment variables referencing all the disk spaces of all your projects. These variables allow you to access the data of your various projects from any one of them.

Choose your storage space according to your needs (permanent, semi-temporary, temporary data, large or small files, etc.).

How to know the quotas and consumption of disk spaces?

You just need to use the IDRIS command idr_quota_user.

By default, it indicates the occupation of the disk spaces HOME, WORK or STORE of your active project (the one you are currently working on).

If your login is attached to multiple projects, you can use the -p ... option of the idr_quota_user command to select a specific project (see the output of the IDRIS idrproj command).

How to request a disk space or inode extension?

If your use of the disk space complies with its intended use and if you cannot delete or move data contained in this space, then your project manager can make a reasoned request for a quota increase (space and/or inodes) on the IDRIS extranet.

Is it possible to know the disk space consumption by each user of my project?

Yes, it is possible with the idr_quota_project command.

The information returned by the command is only updated once a day.

How to share data between members of a project?

To share data between members of the same project, you can use the common spaces $ALL_CCFRWORK, $ALL_CCFRSTORE and $ALL_CCFRSCRATCH which are accessible to all project members and attached to the various disk spaces.

You can manage access to your data simply by setting the Unix rights on subdirectories and files according to the desired access type (read, write or read/write) via the Unix chmod command.

Can I be notified when my files are going to be deleted from SCRATCH?

Unfortunately no. The amount of files stored in SCRATCH is such that we cannot scan the space in a reasonable time to set up this service.

Why are my files deleted from SCRATCH when they are less than 30 days old?

To detect files not accessed (read or modified) for 30 days, IDRIS takes into account the file timestamp. When downloading new data, the timestamp is not always updated and files may appear "older" than they are. This problem usually occurs when manipulating archives.

To avoid keeping an old timestamp when decompressing your data archives, you can use the -m or -DD options of the respective tar and unzip commands:

tar -m [...]

or

unzip -DD [...]
How to control access to my files with ACL (Access Control List) rights?

It is possible to use ACL (Access Control List) rights to share data between users more precisely than with classic Unix rights.

For more information, you can consult our documentation on ACL.

How to recover files that I have accidentally deleted?

Following the migration to the new Lustre storage spaces, the WORK disk space is no longer backed up. We recommend that you keep a copy of your important data as archives stored on your STORE.

For reasons of excessive size, neither the SCRATCH (semi-temporary space) nor the STORE (storage space) are backed up.

Can I ask IDRIS to transfer files from one account to another account?

IDRIS considers that data is linked to a project. Therefore, for the transfer to be possible, it is necessary that:

  1. both accounts (the owner and the recipient) are both attached to the same project;
  2. it is the project manager who makes the request, by signed fax or by email to support (assist@idris.fr) clearly specifying:
    • the machine concerned;
    • the source account and the destination account;
    • the list of files and/or directories to transfer.

Data transfers​

Which commands should I use to transfer data between IDRIS and my connection machine?

In interactive mode

For interactive transfers, you can use the commands related to ssh (sftp and scp), which must be available on your machine (they are on all IDRIS machines). But beware, these commands are very costly in CPU time and transfer time due to the encoding performed by ssh.

In batch mode

File transfers in a batch job can only be done if the transfer commands use an SSH authentication that does not ask for a password. You must therefore, beforehand, generate a public/private SSH key pair on the compute machine with an empty passphrase, then copy only the public key to your local machine. You can consult here the procedure to follow to set up this SSH authentication.

We recommend that you do an interactive test to see if the transfers work properly.

How to transfer data between the 3 national centers (CCFR network)?

Overview

The network of French Computing Centers (CCFR) is a dedicated very high-speed network that interconnects the three national computing centers CINES, IDRIS, and TGCC. This network is made available to users to facilitate data transfers between the national centers. The machines Joliot-Curie at TGCC, Jean Zay at IDRIS and Adastra at CINES are currently connected to this network.

Using this network requires that you have logins (different for each center) in at least two of the three centers and that they are authorized to access the CCFR network in the relevant centers.

Notes:

  • For your IDRIS login, the request for access to the CCFR network can be made:
    • when requesting your account creation from the eDARI portal,
    • or by sending an email togestutil@idris.fr from your address known to the center with the title "CCFR: your IDRIS login / your name" to request CCFR access and allow you to make your transfers to CINES and TGCC. The information will be transmitted to the other two centers so that they can also do what is necessary to make it operational.
  • Additionally, not all Jean Zay nodes are connected to this network. To use it from IDRIS, you can use the jean-zay.idris.fr and jean-zay-pp.idris.fr frontends.

For more information, do not hesitate to contact supportassist@idris.fr.

Data transfer via CCFR

Data transfer between the machines of the centers via the CCFR network is the main service for this network. A wrapper command ccfr_cp accessible via a module is provided to simplify usage:

$ module load ccfr

This ccfr_cp command automatically retrieves the connection information to the specified machine (domain name, specific port) and detects authentication possibilities. By default, the command will opt for basic authentication, using the traditional modalities in force on the targeted machine.
The ccfr_cp command is based on the rsync tool, configured to transit data with the SSH protocol. The copy performed is recursive and preserves symbolic links, access rights as well as the modification dates of the files.
The details of the command, as well as the list of accessible machines on the CCFR network are available by specifying the -h option to the ccfr_cp command.

For transfers from Jean Zay to the CINES and TGCC machines, you can use commands similar to these:

$ module load ccfr
$ ccfr_cp /path/to/datas/on/jean-zay login_cines@adastra:/path/to/directory/on/adastra:
$ ccfr_cp /path/to/datas/on/jean-zay login_tgcc@irene:/path/to/directory/on/irene:

To make transfers from Adastra, the procedure is similar except that you must use the machine adastra-ccfr.cines.fr (accessible from adastra.cines.fr) as indicated on the CINES documentation.
To make transfers from Joliot-Curie, the procedure is also similar and can be done directly from the irene-fr.ccc.cea.fr frontend. After connecting to the machine, the machine.info command will give you all the useful information.

A ccfr_sync command, a variant of ccfr_cp, allows a strong synchronization between the source and the destination by adding, compared to the ccfr_cp command, the deletion of files in the destination that are not present in the source.

Note: By default, these commands will use basic password authentication respecting the terms in force at the remote center (CINES or TGCC). You will therefore certainly be obliged to provide a password each time. To avoid this, you can use IDRIS certificates of the transfert-only type (7 days validity) whose instructions for use are defined on the IDRIS website. The use of such certificates will force you to initiate transfers from the remote machine adastra-ccfr.cines.fr (accessible from adastra.cines.fr) for CINES and irene-fr.ccc.cea.fr for TGCC after copying the transfert-only certificate on the remote machine and to build the rsync transfer commands yourself (therefore not using the wrappers ccfr_cp and ccfr_sync). You can then draw inspiration from the following examples to make your transfers:

# Simple copy from Jean Zay to remote machine (executed on the remote machine)
# using the transfert-only certificate contained in ~/.ssh/id_ecc_rsync on the remote machine
$ rsync --human-readable --recursive --links --perms --times --omit-dir-times -v \
  -e 'ssh -i ~/.ssh/id_ecc_rsync' \
  login_idris@jean-zay-ccfr.idris.fr:/path/on/jean-zay /path/on/adastra/or/irene

# Strong synchronization (--delete option) from Jean Zay to remote machine (executed on the remote machine)
# using the transfert-only certificate contained in ~/.ssh/id_ecc_rsync on the remote machine
$ rsync --human-readable --recursive --links --perms --times --omit-dir-times -v --delete \
  -e 'ssh -i ~/.ssh/id_ecc_rsync' \
   login_idris@jean-zay-ccfr.idris.fr:/path/on/jean-zay /path/on/adastra/or/irene

Warning: On the adastra-ccfr.cines.fr machine, the id_ecc_rsync certificate must be visible from your /home/login_cines/.ssh directory for the ssh command to use it (no environment variable defined for this disk space). You must therefore take care to unpack the certificate in this directory with a command such as:

login_cines@adastra-ccfr.cines.fr:~$ unzip ~/transfert_certif.zip -d /home/login_cines/.ssh
Archive: /lus/home/.../transfert_certif.zip
inflating: /home/login_cines/.ssh/id_ecc_rsync
inflating: /home/login_cines/.ssh/id_ecc_rsync.pub

Parallel file transfer

To speed up file transfers between three national centers, you can also use the parallel-sftp command. This is a tool developed by the CEA that allows using the standard sftp command in parallel to perform parallel file transfers. You will find more information on theCEA website.

parallel-sftp is installed on the login nodes of the Jean Zay IDRIS cluster. Its use is identical to sftp, except for the -n option which allows choosing the number of SSH connections used for parallel transfers.

For example, to perform a parallel transfer with 5 SSH connections:

$ parallel-sftp -n 5 <remote_login>@<remote_host>

Thus, if an sftp transfer is limited to 1 Gbps for example, this transfer will use a maximum of 5 Gbps.

Warning: for your transfers between national centers, you must use the nodes connected to the CCFR network. For IDRIS, this means that the <remote_host> parameter above must be adastra-ccfr.cines.fr or irene-fr-ccfr.ccc.cea.fr. But you can directly use jean-zay.idris.fr to execute the parallel-sftp command.

For use at IDRIS, you can follow the following example written for adastra (for irene simply use irene-fr-ccfr.ccc.cea.fr):

# 1/ Connect to Jean Zay:
$ ssh jean-zay_login@jean-zay.idris.fr
 
# 2/ To initiate a connection allowing transfers with the adastra machine
#    potentially using 5 threads in this case
$ parallel-sftp -n 5 adastra_login@adastra-ccfr.cines.fr
sftp>
 
# You are then in an sftp environment in which you can execute
# put and get commands to make transfers (see man sftp).
# WARNING: environment variables such as WORK, SCRATCH, STORE, ...
#          are not defined and therefore cannot be used to
#          specify the access paths to the data to be transferred.
#          You must indicate the full paths!
 
# 3/ Make the transfers
# To transfer a directory from Jean Zay to adastra, you can do:
sftp> put -r /path/to/jean-zay/src/directory /path/to/adastra/dest/directory
 
# To transfer a directory from adastra to Jean Zay, you can do:
sftp> get -r /path/to/jean-zay/dest/directory /path/to/adastra/src/directory
Can I recover files on external media?

It is no longer possible to request the transfer of your files to external media.

Job management​

When will my jobs start?

The Slurm job manager allows you to display the expected start date and time of your jobs using the --start option of thesqueue command (see the START_TIME column).

Notes:

  • It is possible that Slurm is not able to have a reliable enough estimate on the start time of certain jobs. In this case, the START_TIME column will display N/A as information.
  • As the list of pending jobs is constantly evolving, it is important to keep in mind that the information given by Slurm is an estimate that is likely to change depending on the machine load.
Why do my jobs stay in the queue for so long?

The machine is most likely busy and other jobs may have higher priority than yours.

But it is possible that you have misconfigured your job. For example, if the squeue command indicates (QOSMaxWallDurationPerJobLimit) in the column entitled NODELIST(REASON), this means that you are requesting too much execution time (#SBATCH --time=... directive in your job) for the selected QoS (either by default, or via the #SBATCH --qos=... directive in your job). In this case, you must delete your job and modify it before submitting it again.

Note: if your CPU jobs (respectively GPU jobs) meet the criteria of the CPU partitions and CPU QoS (respectively the GPU partitions and GPU QoS) defined on the machine, do not delete them. They will be executed as soon as possible.

The priority of a job depends on several criteria:

  • The chosen QoS (Quality of Service): "dev" QoS authorizing only short jobs (less than 2 hours) have higher priority than the default "t3" QoS (authorizing jobs up to 20 hours), and "t3" QoS have higher priority than "t4" QoS authorizing single-node jobs up to 100 hours.
  • Computing hours regulation modalities based on a fairshare criterion: a project that is ahead on the consumption of its allocated hours will see the priority of its jobs drop. You can check the status of your project with the idr_compuse command.
  • The waiting time in the queue: the longer a job has been submitted, the higher its priority becomes.

These different criteria ensure that all users can use the resources while maximizing the filling rate of the machine.

What is the maximum limit on the duration of jobs?

The maximum duration of jobs on Jean Zay is 100 hours except on the A100 partition which is limited to 20 hours.

To launch jobs exceeding 20 hours, you must use:

  • the qos_cpu-t4 QoS (Quality of Service) for CPU jobs by indicating #SBATCH --qos=qos_cpu-t4 in your job

  • the qos_gpu-t4 QoS for V100 GPU jobs by indicating #SBATCH --qos=qos_gpu-t4 in your job.

  • the qos_gpu_h100-t4 QoS for H100 GPU jobs by indicating #SBATCH --qos=qos_gpu_h100-t4 in your job.

You will find more details on using these two QoS in the following pages: CPU QoS and GPU QoS.

Note: with these QoS, your jobs may wait longer before starting because they have a lower priority and the number of compute nodes assigned to jobs of more than 20 hours is limited.

Why are my qos_gpu-t4 jobs in the PENDING state for the reason QOSGrpCpuLimit or QOSGrpGRES?

As indicated in our documentation concerning GPU QoS, all jobs in qos_gpu-t4 executed simultaneously by all users cannot occupy more than 32 converged nodes.

When this limit is reached, your jobs then remain PENDING with the reason QOSGrpCpuLimit or QOSGrpGRES.

Warning: do not delete them, they will be executed as soon as possible.

Can I use mpirun to start my parallel codes?

To avoid errors in the automatic distribution of MPI processes on compute nodes and to guarantee a distribution that complies with the resource specifications requested in your Slurm submission file, we recommend that you use the srun Slurm command to execute your code instead of the mpirun command.

A job interrupted due to a technical problem: is a refund possible?

Each project benefits from bonus hours via the possibility of consuming up to 125% of the allocated hours. This measure compensates for small losses of hours without having to make specific requests.

However, it is possible to make requests for additional hours via:

  • either a so-called "au fil de l'eau" (continuous) request which can be made as long as your HPC or AI project has an ongoing hour allocation.
  • or a mid-term request for your Regular Access (HPC) allocation to obtain additional hours for a period of six months.

Naturally, these requests must be justified. For example, by mentioning the tickets opened with IDRIS support and related to losses of hours due to technical problems.

These requests are to be made via the DARI portal as indicated on our web page detailing hour allocation requests.

If the loss of hours exceeds 5% of the allocation, a refund is possible. To do this, it is necessary to open a ticket with support.

Why do my jobs on the gpu_p5 partition fail with the "Illegal Instruction" error?

This error indicates that you have used modules and/or a code incompatible with the AMD CPUs equipping this gpu_p5 partition.

Unlike all other Jean Zay nodes which are equipped with Intel CPUs, the nodes of this partition are equipped with AMD CPUs which has two consequences:

  1. The modules accessible by default (module avail) are not compatible with this partition. It is necessary to load the cpuarch/amd module beforehand (module load cpuarch/amd then module avail) to be able to list and load the compatible modules.
  2. You must also recompile your codes for AMD CPUs using the compatible modules.

You can consult the section Modules compatible with the gpu_p5 partition for more information.

Machine maintenance and unavailability​

How can I know if the machine is unavailable?

The machine may be unavailable due to pre-programmed maintenance or a sudden technical problem. In both cases, the information is available on the IDRIS website's homepage via the drop-down menu entitled "Espace utilisateurs" (User space) then the "Disponibilité des machines" (Machine availability) section.

IDRIS users can also subscribe to the "info-machines" mailing list via the Extranet.

Atomistic Simulation​

Why do I get a "Permission denied" message when I use Gaussian?

By default, access rights to Gaussian are not open to all users. You must contact support to request access.

Why do I get a "Permission denied" message when I use VASP?

Access to VASP is controlled by a very strict authorization mechanism:

  1. To begin with, you must be personally registered on an official VASP software operating license issued by the developers. Registration page
  2. Then, you must make a request to use VASP on our machine via an email to support. Support must then contact VASP to ensure that you are registered before giving you access to the software.

Note that the procedure can take several days.

Why do I get a "Permission denied" message when I use CPMD?

Access to CPMD is controlled by a very strict authorization mechanism:

  1. To begin with, you must be personally registered on their site https://www.cpmd.org.
  2. Then, you must request access to CPMD on our machine by sending an email to support in which you will indicate the email address you used when you registered on the CPMD site. Support must then contact CPMD to ensure that you are registered before giving you access to the software.

Note that the procedure can take several days.

Scientific software and libraries​

Where can I find all the installed scientific libraries?

With each new installation, we update the list of scientific libraries available on the machine. They are classified into two categories: CPU libraries and GPU libraries.

What to do if the library or the version of a library I need is not installed?

You can ask IDRIS support:

  • to perform an installation of the product which will then be available to all users,
  • or to help you make your own installation.