direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Page Content

Scientific Computing

Lupe

The High Performance Computing cluster is available to scientists at the TU Berlin and offers about 3800 CPU cores and 46 GPU-Units. It can reach a peak performance of more than 348 trillion floating point operations per second, in total. With its architecture specifically tailored to data-intensive computing the cluster provides ideal conditions to achieve the ambitious research goals of the users during the scheduled operating time.

Access to cluster

Requirement for accessing is a TUB-Account, which is granted to every member of the TU Berlin by provisioning. For authorization a role administrator of a department must create a team in order to maintain permissions. Please ensure enabling group feature for the team. Every member will be inserted into the SLURM database after was notified about the team's name.

Login to the cluster is done via SSH connection to frontend servers (frontend-01 and frontend-02). In order to connect use the virtual address gateway.hpc.tu-berlin.de. The address is reachable within TUB subnetwork. For authentication (username/password) the TUB-Account is used.

Microsoft Windows requires additional software for accessing the cluster. Putty or MoabTerm are recommended.

Linux Systems can use built-in SSH to connect. Within the TUB subnetwork execute following command in a terminal shell:

ssh [TUB account name]@gateway.hpc.tu-berlin.de

Outside of the TUB subnetwork connecting requires either a VPN (Virtual Private Network) or the usage of the SSH jumphost. How to establish a VPN with TUB subnetwork is explained here. The jumphost is accessible via:

ssh [TUB account name]@sshgate.tu-berlin.de

From that point you can proceed to ssh gateway (see above).

Access to cluster data storage (BeeGFS) is available by scp or rsync (Linux), Windows systems will need additional software which complies with the scp/ssh standard. This storage also contains the user files. Use sshfs in order to access files directly (other methods will always produce copies on the remote system).

Help & documentation

Documentation for users is available at the HPC-Wiki. There you can find information about hardware, available software, to batch commands (how to start a programm) and notes regarding the use of gpu units. In order to access restricted areas within the wiki use your TUB account information to log in.

Calls can be answered within operation times of the ZE Campusmanagement (Mo-Do 10-16, Fr 10-15). You may also contact us via mail. For phone numbers or email contact see the service information at the right of the screen.

A forum for cluster users can be found at the special ISIS course (Key: rechenknecht).

Hardware

Computing Nodes
Count
Art
CPU
#
Memory
GPU
#
132
Standard
Xeon E5-2630v4
2
256 GiB
-
0
20
GPU
Xeon E5-2630v4
2
512 GiB
Tesla P100
2
1
Nachgerüstet
Xeon E5-2630v4
2
256 GiB
Tesla P100
2
26
Nachrüstbar
Xeon E5-2630v4
2
256 GiB
-
0
1
SMP
Xeon E7-4850v4
4
3 TiB
-
0
2
SMP-GPU
Xeon E7-4850v4
4
3 TiB
Tesla P100
2
Storage (BeeGFS)
Count Storage Server
Drives per Server
Drive Size
Storage Size (total)
4
60
6 TB
1440 TB
Prozessoren
CPU
Frequency
Cores
Threads/Core
FLOP/Circle
Intel Xeon E5-2630v4
2.2 GHz
10
2
DP/SP*: 16/32
Intel Xeon E7-4850v4
2.1 Ghz
16
2
DP/SP*: 16/32

* double precision / single precision

Zusatzinformationen / Extras

Quick Access:

Schnellnavigation zur Seite über Nummerneingabe

Auxiliary Functions

HPC Support
Room EN-K048
+49 (0)30 314 74591
+49 (0)30 314 74592
+49 (0)30 314 74593

Website