POWIEW Project focus is on building harmonized high performance computing resources infrastructure across three major Polish HPC centres. The systems will be available, along with scientific and technical support, for advanced computing projects with high identified scientific significance and documented computing needs.
The components are being installed in HPC centers of members of the POWIEW Consortium: Interdisciplinary Centre for Mathematical and Computational Modelling (ICM), Academic Computer Centre CYFRONET and Poznań Supercomputing and Networking Center (PSNC).
HPC systems will be available for scientific projects on the basis of open access.
The infrastructure consist of computer systems with high-end processing abilities:
- Massive parallel processing (MPP) optimized for highly scalable, parallel, fine-grained applications with dense algebraic structure.
- Systems optimized for coarsed-grained programs.
- Symmetric multiprocessing (SMP) systems containing processors with performance as high as several hundred GFLOPS adapted to parallel environment and with large amount of shared memory.
- Special modules accelerating the systems of multi-core and multi-threading structure.
- Professional systems:
- Real time simulations with possibility of hardware configuring (FPGA systems).
- Visual processing using multi-processor systems consisting of GPUs.
- Highly efficient disk arrays, centrally administrated (with 10 PB scale overall capacity).
- Power system with air conditioning and auxiliary power units.
- Pre- and postprocessing, system management, visualization and archiving workstations.
- Software (tools, applications).
The aim of POWIEW Project is installation, enabling and access to new generation of computing infrastructure not being substitute for campus HPC centers and grid environments.
Systems available with a support of the POWIEW Project
- Notos – massively parallel processing system (ICM, University of Warsaw):
- IBM Blue Gene/P architecture,
- 1024 quad-core compute nodes (4 GB RAM each),
- dedicated network realizing functionality of the MPI standard.
- Zeus GPU – accelerated system based on GPGPU architecture (Cyfronet, AGH Cracow)::
- Nvidia® TeslaTM M2050 and Intel Xeon X5670 heterogenous architecture,
- 4 twelve-core Intel Xeon X5670 compute nodes (with additional 896
Nvidia® TeslaTM M2050 cores),
- each node has 72 GB of RAM,
- network architecture based on 4xQDR Infiniband.
- Chimera – SGI Altix UV (PSNC, Polish Academy of Science, Poznan):
- Cane – GPU cluster (PSNC, Polish Academy of Science, Poznan):
- Boreasz – IBM Power 775 system (ICM, University of Warsaw):
- POWER7 architecture,
- 76 compute nodes, 128 GB RAM each,
- dedicated network (HFI).