Skip to content

Latest commit

 

History

History
34 lines (21 loc) · 3.2 KB

systems.md

File metadata and controls

34 lines (21 loc) · 3.2 KB

Hardware Systems

System Name Details Hostname Hardware Specs Software Specs
NCI System Gadi Intel Xeon
Cascade Lake server
gadi-login 2 x Intel Platinum 8274
24-core, 2.6GHz; turboboost up to 3.2 GHz with HyperThreading
32 KB 8-way L1 Cache, 1MB 16-way L2 cache, 36 MB 11-way L3 Cache (shared)
196 GB DDR4 RAM
GCC 8.5.0
OpenMP
PAPI 5.7.0

Logging In

Briefly browse the NCI Gadi user guide to note the main topics.

Once you've logged in to your laptop, open a Terminal window and SSH into Gadi (gadi.nci.org.au) using your Gadi ID and password supplied during the workshop. The ID will be of the form j[k-l][a-z]777; so, for ID jka777, log in via ssh -X [email protected]. (The -X option requests X11 Forwarding, which is necessary to run graphical programs over SSH, e.g. emacs.)

You can use this link to change your NCI password.

Gadi uses environment modules to customize user environments. Run the command module avail to see what modules are available, and module list to see what modules you are using. What version of gcc are you using by default?

You will need gcc and papi for most exercises. If not already loaded, execute the command module load gcc papi to add them, and also add the above command into your ~/.bashrc file for later. Standard UNIX editors are installed including nano, vim and emacs (if you are not familiar with any of these, nano is recommended).

For the exercises, you may wish to open two sessions on remote system at once, one for editing files and one for commands. On Gadi, the command xterm & can be used to create a remote terminal window (saving you from typing your password in again).

Creating an Interactive Session

For exercises which require timing measurements, you should run them interactively on a single compute node of Gadi. To start an interactive job in your current working directory, enter the following command:

qsub -l wd -l ncpus=48 -l mem=192GB -l walltime=30:00 -I -X

This will allocate an entire node of Gadi (48 cores, 192GB of memory) for your exclusive interactive use for a period of 30 minutes, forwarding X windows through your login session. As soon as you've finished using the node, remember to exit to free up the resource for other people to use!