The supercomputers operated by the Governmental Agency for IT Development (KIFÜ) were placed geographically in four major Hungarian cities in close cooperation with local higher education institutions.
Chronologically, the first HP CP4000SL type machine operated at the Victor Hugo street site of KIFÜ until it was decommissioned in 2018.
The cluster in Szeged, which is the same type as the Budapest machine, provides about four times more computing resources in terms of performance, and currently operates with the Pécs’ UV 1000 storage system of SGI. In January 2020, it received a new operating system as well as a task scheduler and resource manager.
When the machine (SGI Altix ICE8400EX) in Debrecen was handed over in 2011, the user community received a much faster resource compared to the computing capacity available at that time, thanks to the Intel Xeon X5680, which is considered to be a modern high clock signal processor. In terms of performance, with 18 teraflops, it is barely behind the Szeged machine (20 teraflops), but it provides this with a total of 1536 processor cores distributed over 128 nodes.
Changes in the development of accelerator processors had a significant impact on scientific software development. These acceleration technologies have multiplied the generally available computing capacity compared to previous computing operations performed solely on processors. In 2015, with the delivery of the Budapest2, Debrecen2 and Debrecen3 machines, we tried to meet the demand for both graphics accelerator cards (GPU) and Intel Xeon Phi accelerator coprocessors.
The NUMA system SGI UV 2000 (Miskolc) machine, which is considered to be unique in Hungary and intended as a modern successor to the SGI UV 1000 located in Pécs, was built to serve computational tasks with a large memory requirement, in which a total of 1.4 TB memory available.
Budapest2 | Debrecen | Debrecen2 (Leo, GPU claster) | Debrecen3 (Apollo, Xeon Phi claster) | Miskolc | Szeged | |
---|---|---|---|---|---|---|
Type | HP SL250s | SGI ICE8400EX | HP SL250s | HP Apollo 8000 | SGI UV 2000 | HP CP4000BL |
CPU / node | 2 | 2 | 2 | 2 | 44 | 4 |
Cores / CPU | 10 | 6 | 8 | 12 | 8 | 12 |
Memory / node | 63 GB | 47 GB | 125 GB | 125 GB | 1.4 TB | 132 GB |
Memory / core | 3 GB | 2.6 GB | 7 GB | 5 GB | 3.75 GB | 2.6 GB |
CPU | Intel Xeon E5-2680 v2 @ 2.80GHz | Intel Xeon X5680 @ 3.33 GHz | Intel Xeon E5-2650 v2 @ 2.60GHz | Intel Xeon E5-2670 v3 @ 2.30GHz | Intel Xeon E5-4627 v2 @ 3.33 GHz | AMD Opteron 6174 @ 2.2GHz |
GPU | – | – | 68 x 3 Nvidia K20x + 16 x 3 Nvidia K40x | – | – | 2 x 6 Nvidia M2070 |
Intel Xeon Phi (KNC) | 14×2 Intel (R) Xeon Phi MIC SE10/7120 | – | – | 45×2 Intel (R) Xeon Phi MIC SE10/7120 | – | – |
Linpack performance (Rmax) | 27 Tflop/s | 18 Tflop/s | 254 Tflop/s | 106 Tflop/s | 8 Tflop/s | 20 Tflop/s |
Number of nodes | 14 | 128 | 84 | 45 | 1 | 50 |
Dedicated storage | 500 TB | 500 TB | 585 TB (common with Phi) | 585 TB (common with GPU) | 240 TB | 200 TB |
Interconnect | IB NB FDR | IB QDR | IB NB FDR | IB NB FDR | Numalink 6 | IB QDR |