Edit this page on GitHub


Working storage

Where to store your files

📖 Important: User homes are NFS mounted and shared by all nodes.


Storage space on head node


Storage space on computing nodes


Storage space on computing nodes for experiments

📖 Important: Fast SSD storage for IO-intensive workloads.


Dataset storage

📖 Important: These are NFS mounted and shared by all nodes.


PubMed Central Open-Access (Snapshot)



Why is Condor required to get GPUs/CUDA acceleration?

In essence, Condor allows sharing the GPU resources between all the users and, as configured, is the only sanctioned way to use GPU acceleration (CUDA) (i.e., processes launched directly via the command-line run on the CPU only).

We know it might take you a little bit of time to get acquainted with Condor, however after a bit of trial and error you will appreciate it. Here is a simple template for condor_submit to help you get started.

📖 Important: Note the use of request_GPUs = 1.

Universe  =  vanilla
GetEnv = True

Remote_Initialdir = /home/user/workspace
Executable = bin/myprogram
Arguments = -l 0.025 -dummy 0

RequestMemory = 2048
RequestCpus = 1
request_GPUs = 1

Output =  /home/user/workspace/myprogram.out
Error =  /home/user/workspace/myprogram.err


If the job is [H]eld or [I]dle when you check condor_q use condor_q -analyze JOB_ID to get debug information.


Check for the availability of software with module avail.