module load cuda/last will load ver. 9.0 of CUDA Toolkit paths. Nvidia drivers are also updated to CUDA-9 compatible (384.81) on all K80 (zeus[400-409) nodes)
Alex Pedcenko
CUDA updated to 9.0 on Zeus HPC
Posted by admin
on March 21, 2018
No comments
SSH Tunnel to the HPC’s compute node from Uni network
Posted by admin
on March 21, 2018
No comments
For more recent Recommended approach see: http://zeus.coventry.ac.uk/wordpress/?p=1119
If you need to launch GUI application on the actual compute node the job was assigned to by slurm, you can do VNC-session through ssh-tunnel (tunneling from outside to the node via login node).
You can connect “directly” to one of the reserved nodes (node assigned by slurm to your job) from university network (or from outside via VPN) using ssh tunnel. E.g. on your local machine issue the command (we presume SSH is installed on your PC, if not, you can use cygwin from w:\Students\Shared\EC\STUDENT\HPC\cygwin\)
ssh -L 5901:zeus15:5901 username@zeus4.coventry.ac.uk
This will connect the port 5901 (i.e. VNC-server session :1 running on zeus15) to the port 5901 (you can hook it to any other if 5901 is already taken on your PC) of your local machine via login node zeus4. Obviously replace “username” by your HPC user name. Then, to connect to VNC session running on zeus15:1, just connect your vncviewer to localhost:1. If you need to forward any other non-privileged port from compute node to your local PC, just change the port numbers in the command above.
vncviewer localhost:1
PGI Community Edition [free]
Posted by admin
on December 8, 2017
No comments
PGI Community Edition
PGI Community Edition includes a no-cost license to a recent release of the PGI Fortran, C and C++ compilers and tools for multicore CPUs and NVIDIA Tesla GPUs, including all OpenACC, OpenMP and CUDA Fortran features. The PGI Community Edition enables development of performance-portable HPC applications with uniform source code across the most widely used parallel processors and systems.
http://www.pgroup.com/products/community.htm
module load pgi/2017
Compilers:
pgc pgcc pgc++ pgf77 pgf90 pgf95 etc...
OpenMPI for PGI compilers
module load pgi/mpi/1.10.2/2017
OR
module load pgi/mpi/2.1.2/2017
Then mpi versions of PGI compilers will be available as
mpicc, mpic++, mpif90, mpif77...
Compiling with Intel MPI compiler
Posted by admin
on October 26, 2017
No comments
Let’s compile parallel code with intel compiler, say version 2013
module load intel/13 (you can also choose intel/11/icc or intel/11/ifort)
module load intel/impi
C-code
mpicc -cc=icc main.c -o main.exe
Fortran code main.f90
mpif90 -fc=ifort main.f90 -o main.exe
Run the code
module load intel/13
module load intel/impi
Launch on 8 CPUs:
mpirun -np 8 ./main.exe
New Parallel File System testing
Posted by admin
on October 24, 2017
No comments
Dear Zeus HPC cluster users,
this is to let you know that we have new parallel file system available for testing and use on Zeus HPC. The new storage utilizes both existing file servers in parallel (zeus5 & 6) over 40Gbps Infiniband fabric, which should bring faster read/write access for jobs. As you may know, users’ home folders are served by one file server of Zeus (zeus5) over NFS network share, and this have led in some cases to slowing down of the file system and causing overall slowdown and poor responsiveness of HPC.
The new file system does not replace users’ homes, but is intended as a runtime space for jobs. The location of this new storage is at /mnt/beegfs/scratch/yourusername, where each user has a working folder. You can also access this space by the link from your home folder, i.e. /home/yourusername/scratch
If you like to launch jobs from this file system, you need to copy your job/project folder to /home/yourusername/scratch and submit your slurm job form there or from /mnt/beegfs/scratch/yourusername if you prefer. This would, in theory, reduce the load on Zeuse’s NFS server, which serves user’s homes.
After your job completes, it is advisable (although not necessary) to copy your results from /home/yourusername/scratch/… to your home folder /home/yourusername for safekeeping. Scratch folder is not backed up and during the test period, there may be a chance of some failures.
Best Regards,
Alex Pedcenko
Power cut 15/06/17 21:30
Posted by admin
on June 16, 2017
No comments
It seems there was a power cut in ECB comms rooms on 15 th June approx. 21:45 — all HPCs went off (If you wonder why your jobs have died)
Alex
New GCC-7.1.0 compiler is installed on zeus
Posted by admin
on May 18, 2017
No comments
module load gcc/7.1.0
Alex
Google is giving a cluster of 1,000 Cloud TPUs to researchers for free
Posted by admin
on May 18, 2017
No comments
HPC room was overheating again on Sunday 7 May
Posted by admin
on May 7, 2017
No comments
Chillers in the HPC room EC3-21 failed once again this Sunday. Broadwell nodes and half of Nehalem nodes (zeus[20-91,15]) were switched off until the cause of the faults will be finally found by Estates.
Compute nodes which are available : zeus[100-171, 200-217] (queues: all, long, GPU)
Regards,
Alex
Recent Comments