HPC – High Performance Computing

Supercomputing capabilities for large-scale scientific applications

GRNET national HPC ARIS (Advanced Research Information System) Infrastructure provides supercomputing capabilities for large-scale scientific applications.

ARIS infrastructure consists of 533 computational nodes seperated in five compute islands, based on Intel x86 architecture, interconnected into a single Infiniband FDR14 network of low latency and high bandwidth, which offer multiple processing capabilities and architectures.

In particular, the ARIS system has:

  • Thin nodes island, based on the IBM NeXtScale platform and Intel Xeon E5-2680v2 processors. It has 426 computing nodes and offers a total of 8,520 cores (CPU cores).
  • Fat nodes island with large memory nodes, consisting of 44 Dell PowerEdge R820 servers. Each server offers 4 Intel Xeon E5-4650v2 processors and 512 GB of central memory.
  • GPU nodes island with GPU accelerated nodes, consisting of 44 Dell PowerEdge R730 servers. Each server contains 2 Intel Xeon E5-2660v3 processors, 64 GB of memory and 2 NVidia K40 GPU cards.
  • Phi nodes with Xeon Phi accelerated nodes consisting of 18 Dell PowerEdge R730 servers. Each server contains 2 Intel Xeon E5-2660v3 processors, 64 GB memory and 2 Intel Xeon Phi 7120P co-processors.
  • Machine learning node island, consisting of 1 server, which contains 2 Intel E5-2698v4 processors, 512 GB of central memory and 8 NVIDIA V100 GPU cards.

The theoretical computing power of the system reaches 535 teraFlops (trillions of floating point operations per second). The infrastructure includes a high-performance 2 Petabyte storage system based on the IBM Elastic Cloud, an evolution of the IBM General Parallel File System (GPFS) and an IBM TS3500 tape library with capacity of about 6 PB.

Regarding security, the system offers full redundancy with each server to have dual power supply, and there is an integrated UPS installation in a separate room which ensures the uninterrupted operation of the server for 15 minutes, which is enough time for closing the system safely in the event of power failure.

HPC - High Performance Computing