National University of Singapore

Release of Atlas7 Linux Cluster

Dear HPC Users,

We are pleased to announce the release of atlas7, a new Linux cluster which has joined the pool of HPC compute resources at Computer Centre. The cluster has 8 large memory compute nodes with each node having four Intel 10-core E7-4860 2.26GHz processors and 256GB memory, and 64 compute nodes with each node having dual hexa-core Intel X5650 2.66GHz processors and 48 GB of memory:

Total Number of Nodes	Server Model	Processors	RAM
8	HP Bl680c G7	Two Intel E7-4860 @ 2.26GHz	256GB
64	HP Bl460c G7	Two Intel X5650 @ 2.66GHz	48GB

Please note the following:

1. Logging in

To access atlas7, you can either ssh to atlas7-c01.nus.edu.sg (172.25.194.7), or login through HPC portal: Login to HPC portal, click atlas7-c01 cluster (xterm) to start a xterm. Since this is a large memory nodes, users who need more than 40GB memory to run their interactive pre- and post-processing jobs can make use of the node.

2. Batch jobs and batch queues

These queues are configured on the atlas7 cluster to run batch jobs:

Job description	Queue Name	Memory Limit	Runtime Limit
Serial large memory jobs	atlas7_large	200GB	Unlimited
Parallel jobs	atlas7_parallel (up to 32 threads)	40GB per-process	Unlimited
Parallel testing jobs	atlas7_parallel_test	40GB per-process	30minutes
OpenMP parallel jobs	openmp	16GB per-process	Unlimited
Short serial jobs	short	16GB	1440 Minutes
Short parallel jobs	short_parallel	40GB per-process	1440 Minutes

 

To run jobs in "openmp" on atlas7 cluster, you need to add an option "-m atlas7cluster" in your job submission.

"short" and "short_parallel" queues are created for short running jobs which can finish within one day. The jobs submitted to these two queues will be dispatched to the idle nodes in the HPC cloud host pools, and will NOT be counted in a user's total number of jobs when a total job number limit per users is applied.

3. Where to submit your jobs (work space)

The high performance parallel GPFS file system (/hpctmp) is accessible from Atlas7 clusters. Users are advised to submit and run their batch jobs from this file system. More details on the high performance work space may be found here. 

4. Compiling your code

Executables compiled on atlas6 could be used to on atlas7 cluster. Please test it first and if necessary, recompile your code on atlas7-c01 using one of the following compilers:
 

Compiler	Command
INTEL	ifort, icc
MPI	mpif90, mpif77, mpicxx, mpicc
GNU	gfortran, g77, gcc

More details on the compilers and code compilation may be found here.

Please check the HPC website for updated information and instructions. For problems or queries, please contact us at .