Description
Launch is a regional computational resource that supports researchers incorporating computational and data-enabled approaches in their scientific workflows at The Texas A&M University System Schools. A portion is offered to the national community. Researchers must be based in the US and associated with a US academic research institution.
Launch is a Dell Linux cluster with 45 compute nodes (8,640 cores) and 2 login nodes. There are 35 compute nodes with 384 GB memory and 10 GPU compute nodes with 768 GB memory and two NVIDIA A30s. The interconnecting fabric uses a single NVIDIA HDR100 InfiniBand switch.
RP Description
Launch is a compute cluster consisting of 45 Dell nodes, 35 CPU and 10 GPU, each with two AMD EPYC 9654 (Genoa) CPUs (192 cores); the GPU nodes add two NVIDIA A30 GPUs. It is particularly well suited to batch and interactive research computing across CPU and GPU workloads, and is often used for general-purpose simulation, data analysis, and portal-based interactive work.
Login Text
Log in to Launch through the [TAMU Launch Open OnDemand portal] in your browser; the [Launch access guide] covers signing in. Before your first sign-in:
- You need an ACCESS ID and an active ACCESS allocation that includes Launch.
- Multi-factor authentication is required through your ACCESS account.
- Launch has no direct SSH login. For a command line, open the cluster shell from the portal.
File Transfer Text
Globus is the recommended way to move data to and from Launch; it handles large or many-file transfers and works without an SSH login. For a few small files, the [HPRC Portal file browser] is simplest, though not for files over 2 GB. Launch has no direct SSH login, so scp, sftp, and rsync cannot reach it from your computer; see the [HPRC File Transfer guide] for details.
Storage Text
Launch provides global file space available from all compute and login nodes. Storage is organized into home, scratch, and project filesystems. Use showquota to view current scratch and project/group quotas. Scratch and Project storage are not backed up. For detailed information, view the Launch [File Systems] guide.
Jobs Information
You can run jobs at different sizes and durations on Launch. The following lists the different queues that you can submit to, describing how many nodes you get, how long you can run, the type of resources you get, and the average wait time.
Launch uses the Slurm scheduler. Submit resource-intensive work to the compute nodes; the two login nodes are for editing, compiling, file management, and other light work, where interactive processes are limited to one hour of CPU time and eight cores in total.
Submit a job script with sbatch, list your jobs with squeue -u $USER, cancel one with scancel, and check its efficiency after it finishes with seff. CPU jobs run on the default partition and need no partition flag; GPU jobs must request --partition=gpu with at least one GPU, for example --gres=gpu:a30:1 (up to two A30s per node). If you do not request memory, Slurm gives 1024 MB per CPU. Run sinfo to see the current partitions and their limits.
If you prefer not to write a Slurm script, the Drona Workflow Engine and the portal's Interactive Apps generate and submit jobs for you through a graphical form.
See the [Launch Batch System] guide, the [Drona Workflow Engine] guide, and the portal [Interactive Apps] guide for more information.