Get Services >  High Performance Computing >  Resource Update >  Release of Quad-Core and Large Memory Linux Cluster Atlas4

SVU Announcement

Expansion of High-Performance Computing Resources with
New Quad-Core Compute Servers

Expansion of HPC Resources

The Supercomputing & Visualisation Unit of Computer Centre is pleased to announce the expansion of its HPC resources with a new Linux-based cluster (Atlas4), consisting of a total of 88 servers. All the servers are installed with Intel quad-core processors. More details of the new cluster are given below:

No. of Nodes

Processors Type

No. of CPUs

Total cores per node

RAM per node

80

Quad-Core Intel Xeon E5430

2

8

16 GB

7

Quad-Core Intel Xeon E7330

4

16

64 GB

1

Quad-Core Intel Xeon E7330

4

16

128 GB

Accessing the New Cluster

The new cluster is accessible via the login nodes at atlas4-c01 and atlas4-c81. Users can use one of the following methods to access the cluster:

  • Via NUS HPC Portal at: http://nushpc.nus.edu.sg
  • Batch jobs can be submitted to batch LSF queues quad_parallel and quad_large, for parallel jobs and serial/multithread large memory jobs, respectively.
  • General serial jobs can be submitted to batch LSF queue linux64.

Application Software

Most of the applications that are running on Atlas2/3 cluster systems should be able to run on the new cluster. Please refer to the software web site for more details here.


Compilation and Running of User Programs

Check the Program Compilers site at the following link if you need to run C/C++ and Fortran programs: http://www.nus.edu.sg/comcen/hpc/software/compilers.html.

Check the Parallel Computing site at the following link if you need to run MPI and OpenMP programs: http://www.nus.edu.sg/comcen/hpc/parallel/index.html.


Scratch Space

At present, there is no scratch space and/or /svutmp space directly connected to all Atlas4 compute servers. Users are recommended to run their jobs at the home directory, where 4GB space is given to each user by default. You can also make use of /lsftmp space, where 1GB is provided by default.

For application that needs to use large scratch space during execution, please submit the jobs to queue mcore_parallel or mcore_large and make use the scratch space of /svutmp directory.

We are planning to expand and share the current /svutmp on Atlas2/3 to the new cluster Atlas4 in the near future. Once this is implemented, users can run their jobs at the high performance storage space on all the HPC clusters Atlas2-4 transparently.


SVU Support

Please do not hesitate to contact us at if you have any queries, or if you encounter any problems when accessing the new cluster Atlas4.

 

top