FAQs

FAQs

Questions

1. How do I get a Spiedie account?
2. What are the available resources?
3. Why can’t I log in from my laptop?
4. Why can’t I use the sudo command?
5. How do I run a program?
6. How do I install new programs?
7. When will my program run?
8. Why did my job abruptly finish?

How do I get a Spiedie account?

• A faculty member can request access through the system admin. You can learn more about the subscription policies and request forms here

What are the available resources?

• Spiedie currently has 166 Compute Nodes with 512 Knights Landing Cores, 9 nVidia p100 GPUs, and 292 TB of Infiniband connected NFS Storage.

Learn more about the various partitions and features here

Why can’t I log in from my laptop?

• You must be assigned a Spiedie account in order to gain access to Spiedie. You cannot use your PODS account. You can only log in through the school internet connection or through a vpn if you are off-campus. Follow the instructions here to log in

Why can’t I use the sudo command?

• All users have super user privileges revoked by default and will not be given root access.

How do I run a program?

• Spiedie uses SLURM workload manager to run the job queue and schedule users jobs. You can learn about SLURM and how submitting jobs work here and here

How do I install new programs?

• Since you will not have root privileges, you must either build from source (custom code should be always compiled on Spiedie prior to running),or use Conda or Singularity.

When will my program run?

• The job queue is managed by SLURM and depends on the job size, current availability, job and user priority.

• If you wish to run a quick test file, it is recommended you use the quick partition which has a run time of 10 minutes per job.

Why did my job abruptly finish?

• Job failures are often associated with mis-configured run parameters. Exceeding memory allocation, or time allocation will result a failed job. Only certain partitions can handle high memory, or high CPU jobs and exceeding node allocations will result in a job failure.

• For GPU-accelerated programs, it is recommended to use double the video memory allocation for memory allocation, in order to make sure the program is not canceled.