Navigation
GPU facilities available for HPC-Europa visitors
GPU facilities available for HPC-Europa visitors
BSC
We can offer HPC-Europa users the following machine:
Resources should be required by number of nodes and number of GPU hours per node.
- System Overview
NVIDIA GPU is a cluster with 128 Bull B505 blades each blade with the following configuration:
- 2 Intel E5649 (6-Core) processor at 2.53 GHz
- 2 M2090 NVIDIA GPU Cards
- 24 GB of Main memory
- Peak Performance: 185.78 TFlops
- 250 GB SSD (Solid State Disk) as local storage
- 2 Infiniband QDR (40 Gbit each) to a non-blocking network
- RedHat Linux
- 14 links of 10 GbitEth to connect to BSC GPFS Storage
- Software stack available
- Red Hat Enterprise Server
- BullX Cluster Suite
- Intel Cluster Studio
- C/C++/Fortran Compilers
- MKL
- Intel MPI
- Intel Trace Analyzer and Collector
- PGI Accelerator Fortran Server
- NVIDIA CUDA Toolkit
- OpenMPI
CINECA
in CINECA we have a Linux Infiniband Cluster IBM PLX (iDataPlex DX360M3)
equipped with 274 nodes each with 2 six-cores Intel Westmere at 2.40 GHz, for a total of 3288 cores
Each node is equipped with two GPUs: 2 NVIDIA Tesla M2070 GPU per node (for 264 nodes) + 2 NVIDIA Tesla M2070Q per node (for 10 nodes) for a total of 548 GPUs
The system has a RAM of 48 Gb/node and the internal network is infiniband with 4x QDR switches.
CSC
The only open GPGPU resource we have at the moment is in Vuori (also available for HPCE2 visitors), see http://www.csc.fi/english/pages/vuori_guide/using_vuori/running_gpu_programs/index_html - 8 nodes with 2 CPUs and 2 Nvidia Fermi GPUs each.
EPCC
Here at EPCC we have a small test system as part of the HECToR service which has four quad-cores each with 32 GB of main memory:
gpu1: Quad-core Intel Xeon 2.4GHz 4x NVidia Fermi C2050 (3GB Memory)
gpu2: Quad-core Intel Xeon 2.4GHz 4x NVidia Fermi C2050 (3GB Memory)
gpu3: Quad-core Intel Xeon 2.4GHz 2x NVidia Fermi C2050 (3GB Memory) and 2x NVidia Fermi C2070 (6GB Memory)
gpu4: Quad-core Intel Xeon 2.4GHz 2x AMD FireStream 9270
There is a link here which gives a bit more information:
http://www.hector.ac.uk/support/documentation/guides/gpgpu/
GENCI-CINES
From GENCI/CINES side, JADE is officially dedicated to HPC-Europa2. But upon specific agreement, depending on the project itself and the availability of the platform, one may have access to a 32 nodes Nehalem Cluster, with 12 node Tesla S1060 equipped. As access to the platform is not guaranteed, it is preferable to consider the GPU part as optional in the workplan.
HLRS
HLRS provides access to the NEC Nehalem Cluster "LAKI" accessible to HPC-E2 guests. Of the 700 nodes, 32 nodes contain NVIDIA Tesla S1070 GPGPUS, 6 nodes with NVIDIA Quadro for visualization and 2 nodes with dual-NVIDIA Fermi C2050. More info on http://www.hlrs.de/systems/platforms/nec-nehalem-cluster/ and the Wiki.
SARA
We at SARA have a small GPU cluster. Description and information is available on the SARA website: http://www.sara.nl/systems/gpu-cluster/description. If needed, the equipment can be used by HPC Europa visitors.