Die letzten Meldungen

Wartung IdM-Portal: abgeschlossen

14. Februar 2017

Am heutigen Dienstag, den 14.2.2017 von 14 bis 17 Uhr, werden einige IdM-Datenbanken aktualisiert.

HPC cluster „LiMa“ unavailable (SOLVED)

11. Februar 2017

Due to a failure of the management/admin node, the LiMa cluster was unavailable from Saturday Feb. 11th approx. 11:10 until approx. 15:30.

Wartung des Servers „FAUPRINT2“

10. Februar 2017

Sehr geehrte Damen und Herren,

Meldungen nach Thema


TinyGPU Cluster

TinyGPU complements the GPU nodes in Emmy cluster and has nodes with two different types of consumer GPUs. The TinyGPU hardware consists of:

  • 7 compute nodes (tg00x), each with two Intel Xeon 5550 "Nehalem" chips (2x 4 physical cores) running at a base frequency 2.66 GHz with 8 MB Shared Cache per chip, 24 GB of RAM (DDR3-1333) and 200 GB of local scratch disk; two NVIDIA GeForce GTX980 GPUs. These GPUs were financed by one user group and replaced the original NVIDIA Tesla M1060 GPU boards which were no longer supported by recent CUDA releases.

  • 1 compute node (tg010) with two Intel Xeon 5650 "Westmere" chips (2x 6 physical cores + SMT) running at 2.66 GHz with 12 MB Shared Cache per chip, 48 GB of RAM (DDR3-1333) and 500 GB of local scratch disk. In total four NVIDIA Tesla C2050/C2070/C2075 "Fermi" GPU Boards (with varying ECC settings)

    This node has been removed from TinyGPU and is now part of RRZE's TestCluster

  • 1 compute node (tg020) with two AMD Opteron 6276 "Interlagos" chips (2x 16 cores) 64 GB of RAM (DDR3-1333), 500 GB of local scratch disk. One NVIDIA Tesla K20c "Keppler" GPU Board and one NVIDIA GeForce GTX 680 GPU Board

    This node has been removed from TinyGPU and is now part of RRZE's TestCluster

  • 5 compute nodes (tg03x), each with two Intel Xeon E5-2620v4 "Broadwell" chips (2x 8 physical cores) running at a non-avx base frequency of 2.1 GHz with 20 MB Shared L3 Cache per chip, 64 GB of RAM (DDR4) and a small local SSD scratch disk; four NVIDIA GeForce GTX1080 GPUs.

This website shows information regarding the following topics:

Access, User Environment, and File Systems

Access to the machine

Access to TinyGPU is through the Woody Frontends. So, connect to


and you will be randomly routed to one of the frontends for Woody, as there are no extra frontends for TinyGPU. See the documentation for the Woodcrest cluster for information about these frontends. Although the TinyGPU compute nodes actually run Ubuntu LTS, the environment is compatible. Programs compiled for Woody will just run on TinyGPU as well. In some cases, you even can compile CUDA programs on the Woody frontends (after loading the cuda module), although no GPU hardware is available there. In case of problems, try to compile your GPU programs on one of the TinyGPU compute nodes (e.g. within an interactive job).

For submitting Jobs, you will have to use the command qsub.tinygpu instead of the normal qsub.

In general, the documentation for Woody applies. This page will only list the differences to Woody.

File Systems

Node-local storage $TMPDIR

Each node has a small local hard drive or SSD for temporary files available under /tmp/ (also accessible via /scratch/).

Compiling CUDA/OpenCL codes

The Woody Frontends only have a limited software installation with regard to GPGPU computing. It is recommended to compile code on one of the TinyGPU nodes, e.g. by requesting an interactive job on TinyGPU.

The GTX1080 GPUs can only be used with CUDA 8.0 (or higher). Host software compiled on the new tg03x nodes might not run on the older tg00x nodes if AVX/AVX2 instructions are used.

Batch Processing

The batch system works just like on Woody, the few notable differences are:

  • The command for job submission is qsub.tinygpu instead of just qsub.
  • Jobs have to request multiples of 4 cores. One dedicated GPU will be assigned per four cores, i.e. ppn=4 will give one GPU, ppn=8 two GPUs, etc.
  • To request a specific type of GPU you can add :gtx980 or :gtx1080 to your PBS request. Otherwise the job will run on any available GPU.

Further Information

InfiniBand Interconnect Fabric

The InfiniBand network on TinyGPU has been retired. It originally was a double data rate (DDR) network, i.e. the links run at 20 GBit/s in each direction. All 8 original nodes were connected to a small DDR switch and can thus communicate with each other fully non blocking.

Letzte Änderung: 20. Oktober 2016, Historie

zum Seitenanfang

Startseite | Kontakt | Impressum

RRZE - Regionales RechenZentrum Erlangen, Martensstraße 1, D-91058 Erlangen | Tel.: +49 9131 8527031 | Fax: +49 9131 302941


  1. Studierende
  2. Beschäftigte
  3. Einrichtungen
  4. IT-Beauftragte
  5. Presse & Öffentlichkeit