What's new?¶
Access¶
There is a new direct entry point to NT via nt.swin.edu.au. The existing login to the OzSTAR machine at ozstar.swin.edu.au remains accessible.
Hardware¶
Ngarrgu Tindebeek has 11,648 CPU cores and 88 GPUs in total.
160 standard compute nodes:
2x 32-core AMD EPYC 7543 CPUs
256 GB RAM
2 TB NVMe SSD
10 high-memory compute nodes:
1024 GB RAM
(other specs same as standard compute nodes)
22 GPU compute nodes:
512 or 1024 GB RAM
4x NVIDIA A100 80 GB GPUs
(other specs same as standard compute nodes)
Nodes are connected via NVIDIA Quantum-2 NDR InfiniBand switches with a data throughput of 200 Gbps.
Projects and filesystem¶
Ngarrgu Tindebeek and the existing OzSTAR cluster shares the same project management system and filesystem. The existing Lustre file system (/fred) will be upgraded to 19 PB, which is accessible on OzSTAR and Ngarrgu Tindebeek.
Limitations¶
Maximum wall time for a job is 7 days.
Maximum CPUs per project is 2500.
Maximum GPUs per project is 100.
Queues¶
As with with the OzSTAR cluster, you do not need to specify a Slurm queue. This is determined automatically by the node you are submitting from. Jobs submitted via tooarrana1/2 will be sent to the new nodes.
Modules¶
Ngarrgu Tindebeek continues to use software modules, but contains a new software hierarchy and a fresh installation of the latest applications. See Environment Modules for more information.
Environment¶
The default number of OpenMP threads is now set to 1 using the environment variable
OMP_NUM_THREADS=1(previously not defined by default). Users must explicitly change this value if they wish to use more than 1 thread.