Ticker

Re-Structured Dokuwiki

Dear Users,

In attempt to improve on the structure of the wiki - such that more will more readily find relevant information - we re-structured the wiki. In particular:

The entry on ATLAS is not top-level any more, but rather sorted into the physics section.
Everything on development, from editors to debugging, has been removed from the top-level page, too. It can now be found in a comprehensive page in a dedicated section.
In particular, notes on "scripting languages" are removed from the top-level page, too. Leaving only a short-cut link to a dedicated section within the development section.

Life-science users, at least those reading along at the bioinf-news mailing list, already know about a dedicated section on life-science, too.

Your HPC-Team

we are happy to announce that we can offer the access to two NVIDIA DGX-1 systems. NVIDIA promotes those as “the most powerful AI system in the world for the most complex challenges”. NVIDIA DGX-1 offers an AI supercomputer (in a box) for superior performance in the fields of artificial intelligence, machine learning and deep learning. Read more on NVIDIA Website.

The machines are equipped with 8 NVIDIA Tesla v100 GPUs with 16GB/GPU or 32GB/GPU, two Intel Xeon E5-2698 v4 2.2 GHz CPUs and 512GB 2nd generation high bandwidth memory yielding a peak performance of 1PFLOPS. The GPUs are interconnected via the NVIDIA NVLink and the network connection consists of 4x InfiniBand 100 Gbps EDR and 2x 10GbE

If you’re intested in using those machines send us an email at hpc@uni-mainz.de

New Scheme for Python-Modulefiles

Dear Users,

Our build framework adopted a new policy to avoid having too many module files. Along goes a change for the Python Modules (Python2 >= 2.7.16 and Python3 >= 3.7.4). In essence, Python modules no longer load the SciPy-Stack, but this has to be loaded separately, paying attention to the matching version.

You will find detailed information in our wiki.

Your HPC-Team

Creating tarballs the parallel way

Dear Users,

Creating tarballs may sometimes have taken some of your patience, occasionally. Compressing directories is especially useful when prior to transfer or archiving. Well, this too can be done in parallel. We just created a wiki entry on the topic.

Your HPC-Team

MOGON II weiter an der Spitze deutscher Hochleistungsrechner

09.07.2018: Der Mainzer Supercomputer MOGON II weiterhin an der Spitze der schnellsten Hochleistungsrechner an deutschen Universitäten und unter den 100 schnellsten Supercomputern der Welt.

Die Johannes Gutenberg-Universität Mainz (JGU) hat 2016 und 2017 die zwei Phasen des Hochleistungsrechners MOGON II in Betrieb genommen, der mit einer Rechenleistung von zwei PetaFLOPS respektive 2.000.000.000.000.000 Gleitkommaoperationen pro Sekunde in der Liste der weltweit 500 schnellsten Rechner im November 2017 auf Platz 65 startete und auch im Juni 2018 mit Platz 87 noch unter den 100 schnellsten Rechnern der Welt rangiert. Damit ist der Mainzer Supercomputer zugleich weiterhin der schnellste Hochleistungsrechner an einer deutschen Universität. In der Liste der energieeffizientesten Supercomputer wird MOGON II weltweit nun auf Position 62 geführt (November 2017: Position 51). Der Name MOGON wurde in Anlehnung an das römische Mogontiacum, dem lateinischen Namen der Stadt Mainz gewählt.

Die Landesregierung, der Bund, die Johannes Gutenberg-Universität Mainz und das Helmholtz Institut Mainz (HIM) haben seit 2016 insgesamt 10,6 Millionen Euro in den neuen Hochleistungsrechner investiert, um die rheinland-pfälzischen Wissenschaftlerinnen und Wissenschaftler im Rahmen der Allianz für Hochleistungsrechnen Rheinland-Pfalz (AHRP) bis 2020 mit Rechenleistung der internationalen Spitzenklasse zu unterstützen.

Archiving from Mogon II

Dear Users,

we have set up our archiving scripts on Mogon II, such that the Mogon I and II setup is identical.

You will find further information in our wiki.

Your HPC Team

Pi is Pi is Pi is Pi — numerical computing with Python

Dear Users,

Script languages are great, but not necessarily for writing numerics code for HPC systems. Here is why, with a (partial) solution with regard to Python. However, compiled languages always have the potential to out-perform an interpreted language.

Your HPC team

Frühjahrsputz / spring-cleaning

Dear Users,

A penny saved is a penny got.

Therefore: Many homedirectories still contain an old directory named "lc2.home". This, lc2 users might recall, is the transferred content from the old lc2 cluster. We would like you to consider to either delete the old stuff or to archive it.

The same is true for a number of old "lc2.<projectname>" directories in the respective project directories.

Documentation concerning archiving from mogon can be found here: https://mogonwiki.zdv.uni-mainz.de/dokuwiki/archiving:tsm

Why should you care?

You save some bits of your file system quota.
You adhere to the DfG or BMBF requirements (or other parties requireing you to backup data).

Your HPC team

A case for slow data transfer

Dear Users,

You like transferring your big data using scp and or rsync. You really do, don't you?

Your HPC-Team

Archiving data from Mogon

Dear Users,

We are glad to announce a new wiki site on archiving data. Long overdue as some of you are already using archives, but hopefully with the tested wrapper scripts a little bit simpler to use.

Your HPC team