Skip to content

Hardware and storage

This page tells you what hardware you share on blazar and how much disk space you get. Use this page to plan your jobs and to keep your account inside its storage quota.

Hardware you share

All users share one set of hardware. SLURM divides it between jobs — see Submit and monitor SLURM jobs.

Component Specification What you can request
GPU 1 × NVIDIA RTX PRO 4500 (Blackwell), 32 GB VRAM 1 GPU per job with --gres=gpu:1
CPU Intel Core Ultra 7 265K Up to 8 cores per job with --cpus-per-task
Memory 64 GB DDR5-5600 Memory per job with --mem, for example --mem=32G
Storage 4 TB NVMe SSD, PCIe Gen4 ×4, up to 7,100 MB/s read and 6,800 MB/s write Shared disk, limited by your quota — see Your storage quota

Login session restrictions

Run heavy work inside a SLURM job, not in your SSH login session. The server reserves its fast hardware for jobs, so a login session cannot use it:

  • No GPU. A login session cannot reach the GPU. You get the GPU only inside a job that requests it — see Request the GPU.
  • Slow CPU cores only. The CPU has fast performance cores and slower efficiency cores. A login session runs on the efficiency cores. SLURM keeps the performance cores for jobs.
  • Limited, shared memory. All login sessions together can use up to 8 GB of RAM and 8 GB of swap — one shared budget for every logged-in user, not a separate budget per person.

    Warning

    If combined memory use goes over the limit, the system can stop any process in any login session, not only the one that used the most memory.

These limits apply to your SSH sessions and to anything else you start outside SLURM, such as a personal cron job.

Use a login session to edit files, manage data, submit jobs, and build container images. Run your training, analysis, and other heavy work inside a SLURM job.

Where your files go

Your files live in two places. Both are on the same disk.

Location Purpose
/home/<username> Your home directory. Your private files, container images, and job scripts.
/home/shared Shared data. Every user can read and write here.

/home is one storage pool of about 3.3 TiB. All users share this pool, so a quota keeps one account from filling the disk for everyone.

The server has no separate scratch or data disk. Keep large datasets in your home directory, inside your quota.

Your storage quota

You get a soft limit and a hard limit on /home.

Limit Default What happens
Soft limit 250 GB You can go above it for a short time. The server warns you at login.
Hard limit 500 GB You cannot write more data.
Grace period 7 days The time you can stay above the soft limit. After this, the soft limit becomes a hard block.

When you go above your soft limit, the server prints a warning the next time you log in over SSH. The warning shows your usage, your limits, and the days left in your grace period. Free up space before the grace period ends, or the server blocks new writes.

Note

An admin can set a different quota for your account. Contact the system administrator if you need more space for your work.

Check your disk usage

Run quota to see how much space you use and what your limits are:

quota -s

The output shows your current usage, your soft limit, and your hard limit.

To find which directories use the most space, run du:

du -h --max-depth=1 ~ | sort -h

To see how much space the shared pool has left, run df:

df -h /home

This shows the total size and free space of the pool that all users share. Your own quota still limits how much of it you can use.

Free up disk space

Container images and containers use disk space in your home directory. They are often the largest files in an account.