Frequently Asked Questions

Start with Quick Start for setup, or contact support for help.

Accounts and Access

Why do I see “No association” after signing in?

This is normal for a new account and usually means you have not been added to a project. Ask your professor to add you to an approved project.

Registration alone does not grant compute access. You need membership in an approved project and a provisioned HPC account. Check your project and allocation.

Why is Remote accounts empty?

Your HPC account may not be provisioned yet. If you have been notified that access is ready, check Remote accounts and contact support if the account is missing or not in state OK.

What if my name, institutional email, or affiliation is incorrect?

Open a ticket with NOC NTUA to correct institutional attributes. See Review Your ICARUS Profile.

Who can add members or change their roles?

A Project manager or Project administrator can manage the approved project's team. Members must sign in once before being added. See Project Administration.

Which username and account do I use?

Use your HPC Username for SSH and the project's SLURM account in job scripts. With multiple projects, select the account you want to charge. Find both in View Your Projects.

Do I need an SSH key if I only manage projects?

No. Upload your own SSH public key if you need to connect to the HPC infrastructure and run jobs. Managing projects only in the dashboard does not require an SSH key.

Why does SSH report “Permission denied (publickey)”?

Check your HPC username, active access, and that your SSH client uses the private key matching your registered public key. See Login.

Proposals and Allocations

Who can submit a proposal, and when?

Faculty can apply through Calls for proposals → Production Call of the High-Performance Computing offering while the call is open. Students and researchers join an approved project. See Calls and Project Proposals.

How many projects can a PI have?

The current limit is 4 projects per PI; exceptions may be considered. Contact support with your justification.

What do I need to complete in a proposal?

Enter the project name and summary. Under Resource requests, click Add resource, select Standard, and click Save. Add team members and assign roles now or after approval. Check Progress and click Submit when the required fields are complete. See the proposal checklist.

Does “In review” mean I can access the HPC infrastructure?

No. Wait for approval and notification that access is ready. Track the decision in your proposal.

What are the compute limits?

All projects use Standard (Slurm QoS standard): up to 8 nodes and 72 hours walltime, with nodes × requested walltime ≤ 72 node-hours per job. Choose the combination your job needs; see the table and figure in Compute Tiers. The 72 node-hour limit applies to each job, not to the project’s total allocation. For example, 2 nodes for 36 hours or 8 nodes for 9 hours fit the limit. Scheduling still depends on resource availability and priority.

How long does a project last, and can it be extended?

The fixed duration is 6 months. Extension requests are open and subject to approval; the current end date applies until a new one is confirmed. For extension requests, open a support ticket.

Do I need to specify a QoS in my job script?

The allocation uses standard as its default QoS. You can omit --qos or specify --qos=standard. Set the project’s Slurm account with --account and use the gh200 partition. See Compute Tiers.

Infrastructure and Software

How many compute nodes are available?

ICARUS consists of 24 compute nodes with 3,456 CPU cores and 48 NVIDIA GH200 GPUs. Currently, 20 of those nodes are available to the NTUA community. Each node has 144 CPU cores and two GPUs. See Infrastructure.

Is CPU memory the same as GPU memory?

No. CPU LPDDR5X memory and GPU HBM3e memory are separate pools. Slurm is configured with 960,000 MiB (937.5 GiB) of host memory per compute node; each GPU reports approximately 143.4 GiB. --mem requests host memory. See NVIDIA Grace Hopper and Job Submission.

Running Jobs and Checking Usage

Can I run an executable built on my laptop?

ICARUS uses Arm64 CPUs. Rebuild x86-64 executables with the available compilers, and load the same software modules when compiling and running.

Where should I compile and run my application?

Compile on a login node and run computations through Slurm. Keep your executable and inputs in your home or project scratch directory, and set the working directory in your job script. /tmp is local to each node.

Can my job download datasets?

Compute nodes have no internet access. Transfer code and datasets through a login node before submitting jobs.

Will my job stop if I disconnect?

No. A job accepted by sbatch continues under Slurm. Reconnect to check its status.

Why is my job pending, rejected, or timing out?

For pending jobs, check the reason in squeue -u "$USER" and the pending-reason guide. For rejection or timeout, check your account and requests against the approved compute tier.

Where can I check usage and my remaining balance?

Use the dashboard usage report for your allocation balance. Use Job Accounting for sacct job records and sreport project totals. Reports may lag; CPU-hours alone do not represent GPU-equivalent charges.

Where are job logs and performance metrics?

See job output paths and Job Accounting. Missing GPU metrics do not mean zero GPU usage.

Storage and Data

What are my storage limits?

Home storage is 25 GB per user. The Scratch tier provides 2 TB per project, shared by its members. See Storage Quotas for usage commands.

When is my home directory created?

Your home directory is created on your first successful SSH login. Run echo "$HOME" to display its path.

Where is project scratch, and who can access it?

The shared directory is /nfs/scratch/<slurm_account>. Replace <slurm_account> with your project’s account. Authorized members share its 2 TB quota. Added members gain access after membership synchronization; removed members lose access after synchronization. See Project Administration.

How do I check storage usage?

For home storage, run lfs quota -h -p $(id -u) /nfs. For project scratch, use the directory’s Lustre project ID as shown in Check Project Usage. The Slurm account name and Lustre project ID are different identifiers.

Does joining more projects increase my home quota?

No. Each project has its own scratch allowance, and your home quota stays the same. Use the correct project directory; selecting a Slurm account does not change where files are stored.

Are files backed up, and when are they deleted?

Files are not backed up. Project scratch data is deleted 3 months after the project ends. Home data is deleted after 6 months without being on any project. Keep independent copies and transfer files before access expires. See Data Retention Policy.

Why do writes fail when the filesystem has free space?

You may have reached a storage or file-count quota. Check your quotas; df shows filesystem capacity, not your remaining allowance.

Support

How do I report an issue or request an extension?

Open a support ticket from the dashboard. For HPC issues, select High-Performance Computing as the Organization and choose the relevant project and resource. Include your HPC username, project or allocation, job ID where relevant, and the error or requested change. Use Request for an extension; approval must be confirmed.

Ask your supervisor or project manager to report the problem on your behalf, including your institutional username and the error message.