ADA is an HPC (High-Performance Computing) cluster at IIIT Hyderabad. More here.

The official guide for ADA is on hpc.iiit.ac.in (requires IIITH intranet access). This is maintained by Avneesh Mishra from RRC (unofficially, in spare time, for personal reference).

To get access, you have to mail [email protected] and CC your guide. Please do not mail him until absolutely necessary. Explicitly state that you need a 100 GB share space (if you’re a IIITH student working in a research center). CVIT users may have to do extra.

NOTE: Before proceeding, please read the following

Login

Login to ADA using any of the following. The following commands will log you into the master node (no compute allowed over it).
Login to the allotted node (already using or busy) using SSH (use -J for jump)

Remote Access

Compute

Request

Request a quick interactive session (time limit 6 hours)

It is ideal to run large scripts (that take more than 6 hours) through sbatch.

ML & AI Workloads

For workloads pertaining to deep learning. It is best to load the following modules (check versioning) beforehand (especially for TensorFlow workloads). Modules can be loaded through the main node only (not through allocated nodes), unless in an sbatch script.

# Load CUDA and cuDNN modules
module load u18/cuda/11.6 u18/cudnn/8.4.0-cuda-11.6

PyTorch workloads

TensorFlow workloads

You can inspect the usage by running nvidia-smi after accessing the allocated gnode through ssh, or use nvitop.

Storage

Always remember this table

Storage Directory	Access	Maximum Quota
`/home2/$USER` or `~`	Main node and allocated nodes (home)	`25 GB`
`/share1/$USER`	Main node only (not on allocated nodes)	`100 GB` [1]
`/scratch/$USER`	Allocated nodes only (not on main node)	`2 TB` (7 days) [2]

[1]: CVIT users might have more space options. This table (and this page) is mainly for non-CVIT people