Home About HPC History News Partners

Home — about HPC

About HPC

High performance computing

High performance computing (HPC) means using (super)computers and computing clusters to solve problems that are numerically or data intensive problems freom various fields of research and technology. Examples include medicine (genetics, „in silico“ drug design, …), physics (meteorological a climatological models, nuclear and particle physics, …), chemistry (properties of atoms and molecules, correlation between structure and reactivity, ….), but also economics (investment risk, stock growth, …) and many other fields.


Supercomputers are systems with a high level of performance compared to common, personal computers. Standardized testing is used to determine the performance level, such as the LINPACK (LINear equations software PACKage) benchmark. Computing performance is measured is Flop (floating point operations per second). Modern supercomputers performance in the range of terraFlops (10^12) and petaFlops (10^15). Agregate computing power is influenced by other parameters such as latency and throughput of the data networks used for communication between the processors and the nodes of the supercomputer, delay and throughput of the communication with data storage and others. All of these parameters determine the scalability of the so called parallel computations that use multiple nodes, processors or cores at the same time.


Parallel computing is the dominant phenomenon in modern HPC. Multi-node, multi-processor and multi-core architectures are used by the vast majority or computing resources, with individual computing units communicating together and offering aggregate performance as a single system. The connections are implemented by high speed network, such as Infiniband or 10Gb/s ethernet, or some other dedicated network. Architectures of this type can be scaled (at least theoretically) to huge sizes, the only limiting factors being electrical consumption, cooling and room space of the data-center.

Medicine

  • — Genetics
  • — “in silico“
  • — Drug Design
  • ...

Physics

  • — Meteorological and Climate Models
  • — Nuclear and particle Physics
  • ...

Chemistry

  • — Properties of Atoms and Molecules
  • — Corelation between Structure and Reaktivity
  • — ...

Ekonomika

  • — Investment Risk
  • — Stock Growth
  • — ...

Cluster Computing

A computing cluster consists of multiple computers, connected by a high speed local network, that cooperate to behave like a single homogeneous system from the outside. Due to their low cost, clusters are usually used to increase performance and availability. They use software, that allows high performance distributed computing. Cluster performance is oftentimes increased by using graphical accelerators. HA – high availability can be implemented by redundancy of the key components of the cluster and ensures its continuous production even when some of its components fail.

System: IBM dx360 M3

Number of compute nodes: 46

Processor: 2x 6 core Intel Xeon X5640 @2,27GHz

Memory: 96GB

Hard drive SubSystem : 2 x 500GB

Compute network: 40Gb/s Infiniband

Operating system: Scientific Linux 6.3

2x NVIDIA Tesla M2070 6GB RAM, 448 CUDA cores

System: IBM iDataPlex dx360

Number of compute nodes: 52

Processor: 2x 6 core Intel Xeon X5670 @2,93GHz

Memory: 48GB

Hard drive SubSystem : 1x 2TB

Compute network: 40Gb/s Infiniband

Operating system: Scientific Linux 6.4

2x NVIDIA Tesla M2050 6GB RAM, 448 CUDA cores

System: IBM Blade system x HS22

Number of compute nodes: 24

Processor: 2x 6 core Intel Xeon X5640 @2,27GHz

Memory: 48GB

Hard drive SubSystem : 500GB

Compute network: 40Gb/s Infiniband, 10Gb/s Ethernet

Operating system: Scientific Linux 6.3


System: IBM Blade system x HS23

Number of compute nodes: 19

Processor: 2x 8 core Intel Xeon E5-2650 @2,6GHz

Memory: 64GB

Hard drive SubSystem : 500GB

Compute network: 40Gb/s Infiniband, 10Gb/s Ethernet

Operating system: Scientific Linux 6.3


System: IBM iDataPlex dx360 M3

Number of compute nodes: 2

Processor: 2x 6 core Intel Xeon X5640 @2,27GHz

Memory: 24GB

Hard drive SubSystem : 2x 500GB

Compute network: 40Gb/s Infiniband

Operating system: Linux

2x NVIDIA Tesla M2070 6GB RAM, 448 CUDA cores


System: IBM iDataPlex dx360 M4

Number of compute nodes: 1

Processor: 2x 8 core Intel Xeon E5-2650 @2,6GHz

Memory: 64GB

Hard drive SubSystem : 2x 500GB

Compute network: 40Gb/s Infiniband

Operating system: Linux

1x NVIDIA Tesla K20 5GB RAM, 2496 CUDA cores

System: IBM iDataPlex dx360 M3

Number of compute nodes: 30

Processor: 2x 6 core Intel Xeon X5640 @2,27GHz

Memory: 4GB

Hard drive SubSystem : 2x 500GB

Compute network: 40Gb/s Infiniband

Operating system: Linux

16x NVIDIA Tesla M2070 6GB RAM, 448 CUDA cores

10x NVIDIA Tesla K20 5GB RAM, 2496 CUDA cores

System: IBM iDataPlex dx360 M3

Number of compute nodes: 54

Processor: 2x Intel E5645 @2,4GHz

Memory: 48GB

Hard drive SubSystem : 2 x 500GB

Compute network: 40Gb/s Infiniband, 10Gb/s Ethernet

2x NVIDIA Tesla M2070 6GB RAM, 448 CUDA cores


System: IBM iDataPlex dx360 M4

Number of compute nodes: 8

Processor: 2 x Intel E5-2670 @2,6GHz

Memory: 64GB

Hard drive SubSystem : 2 x 500GB

Compute network: 40Gb/s Infiniband, 10Gb/s Ethernet

2x NVIDIA Tesla K20 5GB RAM, 2496 CUDA cores


System: IBM iDataPlex x3550 M4

Number of compute nodes: 1

Processor: 1 x Intel E5-2640 @2,5GHz

Memory: 8GB

Hard drive SubSystem : 2 x 500GB

Compute network: 40Gb/s Infiniband, 10Gb/s Ethernet

System: IBM iDataPlex dx360 M3

Number of compute nodes: 24

Processor: 2x Intel Xeon X5670 @2,93GHz

Memory: 48GB

Hard drive SubSystem : 2 x 500GB

Compute network: 40Gb/s Infiniband

Operating system: Linux

2x NVIDIA Tesla M2070 6GB RAM, 448 CUDA cores


System: IBM iDataPlex dx360 M4

Number of compute nodes: 8

Processor: 2x Intel Xeon E5-2670

Memory: 64GB

Hard drive SubSystem : 2x 500GB + 2x 900GB

Compute network: 40Gb/s Infiniband

Operating system: Linux


System: IBM iDataPlex dx360 M4

Number of compute nodes: 6

Processor: 2x Intel Xeon E5-2670

Memory: 128GB

Hard drive SubSystem : 2x 900GB

Compute network: 40Gb/s Infiniband

Operating system: Linux


System: IBM iDataPlex dx360 M4

Number of compute nodes: 3

Processor: 2x Intel Xeon E5-2670

Memory: 64GB

Hard drive SubSystem : 2x 900GB

Compute network: 40Gb/s Infiniband

Operating system: Linux

2x NVIDIA Tesla K20 5GB RAM, 2496 CUDA cores

Number of Cores

4096

Memory Space

12144GB

Hard Drive Space

483.2TB

Massively parallel processing

Massively parallel processing (MPP) means running complex scientific calculations on multiple computing cores. Unlike in cluster computing, the nodes are more densely integrated and better optimized for distributed computations. Higher density and number of processors and faster connections between computing nodes ensure better and faster communication between the processes and higher aggregate performance. MPP is appropriate for the most complicated of scientific calculations with a large amount of communication between processes.

System: IBM Power 775

Number of computing cores: 4096

Memory: 32TB

Compute network: 5-48GB/s Internal Optical Links, 10GB/s Ethernet prepojenie s úložiskom dát

Capacity of external data storage: 600TB

Capacity of internal data storage: 300TB

Operating system: AIX

pcs

Aurel (IBM Power 775)

Shared memory processing

Shared memory processing (SMP) means using various implementations od shared memory. Shared memory allows multiple processes to access a common memory space, without the need of a network. SMP computing jobs are usually run on a single node, where they are split into threads and parallelized among the processors. SMP architectures are appropriate for jobs that require a large memory, frequent memory access and use parallelism on the thread level.

Typ 1

System: IBM Power 755/750

Number of compute nodes: 18

Processor: 4x 8-core Power7 3,3GHz

Memory per compute core: 128GB

Hard drive SubSystem : 6 x 600GB

Compute network: 40GB/s Infiniband

Operating system: SUSE 11.3 Linux

Typ 2

System: IBM Power 755/750

Number of compute nodes: 2

Processor: 4x 8-core Power7 3,3GHz

Memory per compute core: 256GB

Hard drive SubSystem : 6 x 600GB

Compute network: 40GB/s Infiniband

Operating system: SUSE 11.3 Linux

Typ 3

System: IBM Power 780

Number of compute nodes: 1

Processor: 8x 8-core Power7 3,3GHz

Memory per compute core: 1024GB

Hard drive SubSystem : 6 x 600GB

Compute network: 40GB/s Infiniband

Operating system: SUSE 11.3 Linux